GEAIRA 0.1 - Generic Enterprise AI Reference Architecture
Status: Working Draft - M17 complete; human-readable GEAIRA 0.1 architecture baseline frozen
Programme: AIPS companion reference architecture
Current slice: GEAIRA-M17 - Consolidation & 0.1 Baseline Freeze
Semantic baseline: GEAIRM 0.1 M01-M17 frozen; A1-A15 executable formalization closed release candidate
GEAIRA answers: How should enterprise AI be architected?
1. GEAIRA-M01 - Architecture Charter, Jurisdiction & Modelling Principles
1.1 Purpose
GEAIRA-M01 establishes the architectural jurisdiction and modelling contract for GEAIRA before any detailed reference-architecture decomposition is frozen.
The slice answers five foundational questions:
- What architectural problem does GEAIRA solve?
- Which concerns belong to GEAIRA and which remain authoritative elsewhere?
- What architectural constructs may GEAIRA define?
- Which principles constrain later architecture decisions?
- How will the 0.1 human-readable architecture programme proceed?
M01 deliberately does not freeze a final set of Architecture Building Blocks (ABBs). The purpose of M01 is to create the rules by which later ABBs may be justified.
1.2 Source baseline
GEAIRA 0.1 begins from the completed GEAIRM 0.1 executable release candidate.
The architectural source baseline therefore includes:
- the 39 GEAIRM-owned canonical semantic elements;
- the 41 GEAIRM-owned canonical relationships;
- the 15 externally governed relationship/property bridges;
- the 36 frozen semantic invariants;
- the M16 canonical positive patterns and counterexamples;
- the M17 hard-distinction and cross-model authority boundaries;
- the A1-A15 machine-readable and conformance artifacts.
GEAIRA treats these as semantic input, not as a pre-existing component model.
GEAIRA-M01-R01 - Semantic baseline preservation: GEAIRA SHALL consume the frozen GEAIRM 0.1 semantic baseline without redefining its canonical concepts, relationships, distinctions or authority boundaries.
GEAIRA-M01-R02 - No one-to-one projection assumption: A GEAIRM concept or relationship SHALL NOT by itself require a one-to-one GEAIRA component, service, registry, database, API or runtime.
Examples:
GEAIRM usesTool
!= mandatory Tool Registry ABB
GEAIRM hasAIContext
!= mandatory Context Service ABB
GEAIRM usesEvidence
!= mandatory Evidence Store ABB
GEAIRM authorizedBy
!= mandatory Authorization Service ABB
GEAIRM retrievesResource
!= mandatory vector database
GEAIRM supersedes
!= mandatory source-control system
2. Architectural Purpose and Scope
2.1 What GEAIRA is
GEAIRA is a technology-neutral enterprise AI reference architecture.
It defines reusable architectural responsibilities and structural guidance for realizing enterprise AI semantics across different organizations, products, agents, runtimes and deployment environments.
GEAIRA may define:
- architecture planes;
- Architecture Building Blocks (ABBs);
- architectural responsibilities;
- interactions and interaction contracts;
- architectural control points;
- reference views;
- cross-cutting concerns;
- canonical architecture patterns;
- conditional/optional realization choices;
- architecture invariants and conformance expectations.
GEAIRA-M01-R03 - Responsibility-first architecture: GEAIRA SHALL identify architectural responsibilities before deciding whether those responsibilities require distinct ABBs.
GEAIRA-M01-R04 - Technology neutrality: GEAIRA SHALL describe architecture in technology-neutral terms and SHALL NOT require a specific cloud, model vendor, vector database, orchestration framework, agent framework, message broker, API gateway or implementation language unless a later profile explicitly specializes the generic architecture.
2.2 What GEAIRA is not
GEAIRA is not:
- the definition of Artificial Intelligence;
- an AI Product qualification specification;
- a Product ontology;
- an Agent ontology;
- a Data reference model;
- an Analytics/Information reference model;
- an enterprise-specific target platform architecture;
- a deployment implementation specification;
- a mandatory vendor stack;
- a replacement for security, privacy, regulatory or technology standards.
GEAIRA-M01-R05 - Architecture/semantics separation: GEAIRA SHALL preserve the distinction between semantic meaning and architectural realization.
GEAIRA-M01-R06 - Product qualification boundary: GEAIRA SHALL NOT decide whether an offering is a Product or whether a Product Version qualifies as an AI Product; those decisions remain under BPS and AIPS authority respectively.
3. Architecture Jurisdiction
3.1 Authority map
| Concern | Primary authority | GEAIRA treatment |
|---|---|---|
| Enterprise-AI semantics | GEAIRM | Consume and architect for; do not redefine. |
| Generic Product ontology / Product Intrinsic Nature | BPS / PIN | Reference where Product architectural responsibilities arise. |
| AI Product qualification, Product-version AI semantics, risk/governance | AIPS | Consume where Product-specific architecture responsibilities arise. |
| Agent identity and taxonomy | KivaNura Agent Ontology | Consume; define architecture for Agent participation without redefining Agent identity. |
| Data semantics | GEDRM | Reference; architecture alignment occurs with GEDRA. |
| Data reference architecture | GEDRA | Peer architecture; define explicit interoperability boundaries. |
| Analytics / Information semantics | GEARM | Reference; architecture alignment occurs with GEARA. |
| Analytics reference architecture | GEARA | Peer architecture; define explicit interoperability boundaries. |
| Generic provenance semantics | W3C PROV | Reuse; do not create a shadow provenance model. |
| AI Product deployment semantics | AIPDS / applicable AIPS deployment specifications | Align rather than duplicate Product deployment rules. |
| Enterprise-specific Data/Information/Intelligence platform realization | HDIP | May realize/specialize GEAIRA; HDIP is not intrinsic GEAIRA. |
| Generic enterprise-AI architectural responsibilities and patterns | GEAIRA | Canonical GEAIRA jurisdiction. |
GEAIRA-M01-R07 - Adjacent-authority preservation: GEAIRA SHALL preserve the authority of GEAIRM, BPS, AIPS, Agent Ontology, GEDRM/GEDRA, GEARM/GEARA and W3C PROV rather than flattening those concerns into a single architecture namespace.
GEAIRA-M01-R08 - Peer-architecture interoperability: Where GEAIRA intersects GEDRA or GEARA, the architecture SHALL define interoperability responsibilities rather than silently absorbing Data or Analytics architecture concerns.
GEAIRA-M01-R09 - Enterprise realization boundary: HDIP MAY realize or specialize GEAIRA responsibilities, but GEAIRA SHALL remain generic and SHALL NOT inherit enterprise-specific HDIP topology as intrinsic architecture.
4. GEAIRA Architecture Metamodel
M01 establishes the architectural constructs that later slices may use.
4.1 Architecture Plane
An Architecture Plane is a coherent grouping of related architectural responsibilities used to organize the reference architecture.
A plane is not automatically:
- a deployable unit;
- a network zone;
- a microservice boundary;
- a team boundary;
- a Product boundary;
- a data ownership domain.
4.2 Architecture Building Block (ABB)
An Architecture Building Block represents a technology-neutral architectural responsibility or cohesive set of responsibilities that is useful to name and reason about independently.
An ABB MAY be realized by:
- one component;
- multiple components;
- a shared platform capability;
- a Product capability;
- an Agent/runtime capability;
- a managed service;
- an embedded library;
- a human or governance process where the responsibility is not fully automatable.
GEAIRA-M01-R10 - ABB realization independence: A GEAIRA ABB SHALL specify responsibility and required architectural behavior without assuming a single physical deployment form.
4.3 Architectural Responsibility
An Architectural Responsibility states what the architecture must enable, constrain, preserve, observe or govern.
Responsibilities are more fundamental than ABB boundaries.
4.4 Architectural Interaction
An Architectural Interaction describes a required or permitted exchange, invocation, delegation, observation or dependency between architectural responsibilities/ABBs.
An interaction does not automatically imply a synchronous API, message, event, queue, database join or network call.
GEAIRA-M01-R11 - Interaction/technology separation: Architectural interactions SHALL remain protocol-neutral unless a later profile or realization explicitly specializes them.
4.5 Interaction Contract
An Interaction Contract defines the obligations at an architectural boundary, such as:
- semantic contract;
- identity/authority expectations;
- input/output expectations;
- policy obligations;
- evidence/telemetry obligations;
- lifecycle/currentness expectations;
- error/failure behavior.
4.6 Control Point
An Architectural Control Point is a location in the logical architecture where an obligation can be evaluated, enforced, recorded or escalated.
A control point is not necessarily a separate service.
4.7 Cross-Cutting Concern
A Cross-Cutting Concern applies across multiple planes or ABBs and MUST NOT be forced into a single plane merely for diagrammatic convenience.
Candidate cross-cutting concerns include:
- security and privacy;
- identity and access;
- policy and governance;
- provenance and traceability;
- observability and evidence;
- resilience and operational safety;
- interoperability and contract management;
- lifecycle, versioning and currentness.
4.8 Reference View
A Reference View is a purposeful projection of the architecture for a specific stakeholder question or end-to-end concern. Views do not create new ABBs merely by presenting them.
4.9 Architecture Pattern
An Architecture Pattern is a reusable arrangement of responsibilities, ABBs and interactions that satisfies a recurring architectural problem under stated conditions.
GEAIRA-M01-R12 - Pattern conditionality: GEAIRA patterns SHALL state their applicability conditions and SHALL NOT be interpreted as mandatory universal pipelines.
5. Foundational Architecture Principles
P01 — Semantic Fidelity
Architecture decisions must preserve GEAIRM meaning and its hard distinctions.
GEAIRA-M01-R13: GEAIRA SHALL NOT collapse semantically distinct GEAIRM concepts merely because an implementation could combine their realization.
P02 — Responsibility Before Component
The architecture begins with responsibility, not boxes.
GEAIRA-M01-R14: Distinct semantic concerns MAY be realized by the same physical component when their architectural responsibilities remain explicit and independently governable.
P03 — Separation of Product, Agent and Runtime
Product identity, Agent identity and AI Execution are different concerns.
GEAIRA-M01-R15: GEAIRA SHALL preserve architectural separation between Product responsibilities, Agent responsibilities and execution/runtime responsibilities even where one implementation hosts several of them together.
P04 — Authority Is Explicit
Capability does not imply permission.
GEAIRA-M01-R16: Decision, action and tool-use paths SHALL make authority, authorization and oversight responsibilities explicit wherever those concerns are applicable.
P05 — Context and Grounding Are Architectural Responsibilities, Not Vendor Features
GEAIRA-M01-R17: GEAIRA SHALL describe context, grounding, retrieval and memory responsibilities independently of specific storage/indexing/retrieval technologies.
P06 — Evidence and Observability Are First-Class
GEAIRA-M01-R18: Architectural paths that materially affect enterprise decisions, actions, assurance or governance SHALL expose sufficient evidence and observability responsibilities for their applicable governance context.
P07 — Lifecycle and Currentness Are Explicit
GEAIRA-M01-R19: Architecture SHALL support version, lifecycle, provenance and currentness responsibilities without requiring historical semantic mutation.
P08 — Composability Without Identity Inheritance
GEAIRA-M01-R20: Composition, orchestration, federation and dependency SHALL NOT automatically transfer Product identity, Agent identity, authority, assurance or governance status between participants.
P09 — Human Oversight Is a Designed Responsibility
GEAIRA-M01-R21: Where human review, approval, intervention or escalation is required, it SHALL be modelled as an explicit architectural responsibility rather than an undocumented operational assumption.
P10 — Fail-Safe Governance Boundaries
GEAIRA-M01-R22: Where policy, authorization, safety or governance information is unavailable or indeterminate, later architecture patterns SHALL define explicit failure/escalation behavior rather than assume permissive continuation.
P11 — No Mandatory Universal AI Pipeline
GEAIRM established that Recommendation, Plan, Decision, Authorization and Action do not form a mandatory universal sequence.
GEAIRA-M01-R23: GEAIRA SHALL model multiple valid decision/action patterns and SHALL NOT impose a universal end-to-end pipeline where the semantic model does not require one.
P12 — Architecture Is Profileable
GEAIRA-M01-R24: The generic reference architecture SHALL allow profiles to specialize responsibilities and patterns for regulatory, industry, risk, deployment or organizational contexts without weakening mandatory GEAIRA invariants.
6. Working Architecture-Plane Hypothesis
The following seven-plane structure was previously identified as a provisional hypothesis during GEAIRM closure. M01 now admits it into the GEAIRA programme as a working architecture decomposition to be tested and refined, not yet as a frozen final plane model.
| Plane | Working label | Architectural question |
|---|---|---|
| P1 | AI Intent, Purpose & Authority | Why is AI being used, under whose authority, and within which policy/decision boundaries? |
| P2 | Knowledge, Context & Grounding | What knowledge, context, grounding, retrieval and memory responsibilities support execution? |
| P3 | Intelligence Capability & Primitive | Which intelligence capabilities and qualifying mechanisms must be made available and governed? |
| P4 | AI Execution & Agentic | How are AI executions, Agents, tools, control loops and runtime participation coordinated? |
| P5 | AI Serving, Product & Integration | How are AI capabilities exposed, integrated, consumed and aligned with Product/service boundaries? |
| P6 | Decision, Action & Human Oversight | How are recommendations, decisions, authorizations, actions, approvals and interventions architected? |
| P7 | Evaluation, Assurance, Observability & Evolution | How are performance, evidence, assurance, observability, lifecycle, currentness and change governed? |
6.1 Plane guardrails
GEAIRA-M01-R25 - Plane non-identity: Architecture planes SHALL be organizational views of responsibilities; they SHALL NOT be treated as Product types, ontology classes, mandatory deployable tiers or network zones.
GEAIRA-M01-R26 - Plane refinement: The seven-plane hypothesis MAY be refined, renamed, split or combined by M03 if responsibility analysis demonstrates a better architecture decomposition, provided semantic coverage and authority boundaries are preserved.
GEAIRA-M01-R27 - Cross-cutting allowance: A responsibility MAY legitimately span multiple planes; GEAIRA SHALL not force cross-cutting responsibilities into artificial single-plane ownership.
7. Architecture Coverage Expectations
GEAIRA 0.1 is expected to cover at least the following responsibility areas:
intent and purpose
policy and delegated authority
knowledge/context/grounding/retrieval/memory
intelligence capabilities and primitives
AI execution and runtime invocation
Agent participation and tool use
physical/digital environment interaction
control-loop participation
serving and integration
Product / AI Product boundary alignment
decision formation
authorization and action
human review / approval / escalation
evaluation and evidence
assurance and governance integration
observability and traceability
version/lifecycle/currentness
semantic and architectural evolution
composition / federation / orchestration
peer interoperability with Data and Analytics architectures
GEAIRA-M01-R28 - Coverage without component inflation: Coverage of a responsibility area SHALL NOT be interpreted as requiring a dedicated ABB for every listed concern.
8. Initial Reference-Architecture Views
M01 identifies candidate views that later slices should materialize. These are not yet frozen.
- Architecture Authority & Responsibility View — semantic and architectural authority boundaries.
- Enterprise AI End-to-End View — intent through execution/effect with optional decision/action paths.
- Context & Grounding View — context resources, retrieval, grounding and memory responsibilities.
- Agentic & Tool-Use View — Agent participation, delegated authority, tool use and execution.
- Physical AI / Control View — environment, sensing, actuation and control-loop realization.
- Product & Serving View — AI Product/service exposure and integration boundary.
- Decision, Authorization & Oversight View — decision/action patterns and human intervention.
- Evaluation, Assurance & Observability View — evidence, evaluation, assurance and telemetry.
- Lifecycle, Provenance & Evolution View — version, currentness, supersession and change.
- Federation & Composition View — composition, orchestration, federation and dependency boundaries.
GEAIRA-M01-R29 - View non-proliferation: Reference views SHALL project the architecture and SHALL NOT create duplicate architectural responsibility merely for presentation purposes.
9. GEAIRA 0.1 Human-Readable Work Programme
GEAIRA will proceed architecture-first and responsibility-first.
| Slice | Title | Principal focus |
|---|---|---|
| M01 | Architecture Charter, Jurisdiction & Modelling Principles | Purpose, scope, architecture authority, metamodel and principles. |
| M02 | Architectural Responsibility Model & ABB Rules | Responsibility taxonomy, ABB admission criteria, interaction/control-point rules. |
| M03 | Plane Model & Cross-Cutting Architecture | Validate/freeze planes and cross-cutting concerns. |
| M04 | P1 — AI Intent, Purpose & Authority | Intent, purpose, policy, delegated authority, authorization boundaries. |
| M05 | P2 — Knowledge, Context & Grounding | Context, resources, retrieval, grounding and memory architecture. |
| M06 | P3 — Intelligence Capability & Primitive | Capability/primitive availability, selection, access and governance responsibilities. |
| M07 | P4 — AI Execution, Agentic & Environment | Execution, Agents, tools, environment and control-loop architecture. |
| M08 | P5 — AI Serving, Product & Integration | Serving, integration, Product boundary and consumption architecture. |
| M09 | P6 — Decision, Action & Human Oversight | Decision, authorization, action, approval, intervention and escalation patterns. |
| M10 | P7 — Evaluation, Assurance, Observability & Evolution | Evaluation, evidence, assurance, observability, lifecycle/currentness and evolution. |
| M11 | Inter-Plane Interaction & Contract Model | Canonical interactions, boundary contracts, control points and responsibility hand-offs. |
| M12 | Adjacent Architecture Interoperability | Alignment with GEDRA, GEARA, AIPS/AIPDS, BPS, Agent Ontology, PROV and HDIP realization. |
| M13 | Deployment-Neutral Realization Patterns | Centralized, distributed, federated, edge/physical and mixed realization choices without vendor lock-in. |
| M14 | Canonical Architecture Patterns & Anti-Patterns | Positive patterns, unsafe conflations and architecture boundary counterexamples. |
| M15 | Canonical Reference Views & End-to-End Flows | Consolidated stakeholder views and representative end-to-end architecture flows. |
| M16 | Architecture Invariants, Conformance & Profiling | Mandatory architecture invariants, profile rules and conformance boundaries. |
| M17 | Consolidation & 0.1 Baseline Freeze | Cross-slice reconciliation and human-readable architecture baseline freeze. |
Executable/machine-readable formalization of GEAIRA will be planned after M17 rather than assumed in M01.
10. M01 Decisions
M01 establishes the following programme decisions:
GEAIRA consumes GEAIRM; it does not redefine it.
architecture responsibility != semantic concept
architecture interaction != semantic relationship one-to-one
ABB != mandatory deployable component
plane != network tier
control point != mandatory service
reference view != new architecture
Product != Agent != AI Execution
capability != authority
context != environment
retrieval != grounding
recommendation != decision != authorization != action
evaluation != assurance
provenance != dependency
lifecycle != currentness
technology-neutral first
responsibility before component
explicit authority and oversight
observable and evidentiary architecture
profileable without semantic weakening
11. M01 Completion State
GEAIRA-M01 — COMPLETE.
M01 introduces:
- one architecture charter;
- a clear jurisdiction map;
- nine architecture metamodel constructs;
- twelve foundational principles;
- 29 normative M01 rules;
- a seven-plane working hypothesis;
- ten candidate reference views;
- a 17-slice human-readable GEAIRA 0.1 programme.
M01 does not freeze final ABBs or plane boundaries.
Next slice
GEAIRA-M02 — Architectural Responsibility Model & ABB Rules
M02 will define how responsibilities are discovered from GEAIRM semantics and enterprise-AI obligations, how ABBs earn admission into the canonical architecture, and how interactions/control points are represented without component inflation.
12. GEAIRA-M02 - Architectural Responsibility Model & ABB Rules
12.1 Purpose
GEAIRA-M02 establishes the discipline that sits between the frozen GEAIRM semantic model and the later GEAIRA architecture decomposition.
M02 answers five questions:
- How is an Architectural Responsibility discovered and stated?
- How strong is that responsibility and when does it apply?
- When should one or more responsibilities be promoted into a canonical Architecture Building Block (ABB)?
- When should GEAIRA deliberately not create an ABB?
- How are responsibilities, ABBs, interactions, contracts and control points traced back to their semantic or external source obligations?
M02 does not define the final GEAIRA ABB catalogue. Plane-specific ABB candidates will emerge in M04-M10 and be reconciled across planes in M11-M17.
GEAIRA-M02-R01 - Responsibility derivation before ABB admission: Every canonical ABB SHALL be justified by one or more explicitly stated Architectural Responsibilities; a semantic noun, relationship, vendor feature or diagram box alone is insufficient justification.
GEAIRA-M02-R02 - No semantic-to-component shortcut: A GEAIRM concept, role, state, classification or relationship SHALL NOT be promoted directly into an ABB without architectural-responsibility analysis.
12.2 Responsibility Derivation Model
Architectural Responsibilities may be derived from five legitimate source classes:
| Source class | Examples | GEAIRA treatment |
|---|---|---|
| Semantic obligation | GEAIRM concepts, relationships, invariants, hard distinctions | Translate meaning into required architectural outcomes without redefining the semantics. |
| External authority obligation | AIPS, BPS, Agent Ontology, GEDRA, GEARA, PROV, security/privacy/regulatory standards | Architect for the obligation while preserving external ownership. |
| Architecture quality obligation | availability, resilience, safety, interoperability, observability, recoverability | State the required quality outcome and scope without prematurely selecting technology. |
| Enterprise control obligation | authorization, segregation, review, evidence retention, escalation | Represent as architecture responsibility/control-point need where generally applicable. |
| Pattern obligation | requirements introduced only when a selected canonical pattern is used | Bind to the pattern applicability condition rather than universalize it. |
The preferred derivation sequence is:
semantic / external obligation
|
v
required architectural outcome
|
v
Architectural Responsibility
|
+--> interactions / contracts
+--> control points
+--> evidence / failure obligations
|
v
ABB admission analysis
|
+--> canonical ABB
+--> responsibility remains distributed/embedded
+--> profile/pattern-specific realization
GEAIRA-M02-R03 - Source traceability: Every Architectural Responsibility SHALL identify the semantic, external-authority, quality, enterprise-control or pattern source that justifies it.
GEAIRA-M02-R04 - No invented semantic authority: Architecture-derived responsibilities MAY add structural, control, operational or quality obligations, but SHALL NOT be presented as new GEAIRM semantic truth.
GEAIRA-M02-R05 - Outcome orientation: Responsibilities SHALL be stated as required architectural outcomes or obligations rather than as technology selections or organizational tasks.
GEAIRA-M02-R06 - Conditional derivation: When a responsibility arises only if a particular semantic situation, risk condition, pattern or capability is present, its applicability condition SHALL be explicit.
12.3 Canonical Architectural Responsibility Model
An Architectural Responsibility is the smallest architecture-significant statement that says what the architecture must enable, constrain, preserve, expose, evaluate, coordinate or govern.
A responsibility should be expressible using the following grammar:
When/where
<applicability condition>, the architecture SHALL<required architectural outcome>and SHALL<required evidence/failure/contract behavior where applicable>.
Example:
Where an AI Execution can cause an enterprise action, the architecture SHALL make the applicable authorization determinable before the action is committed and SHALL expose evidence of the authorization outcome appropriate to the governing context.
A fully stated responsibility SHOULD capture the following fields in human-readable form:
| Field | Purpose |
|---|---|
| Responsibility ID | Stable local identifier once the responsibility is frozen. |
| Name | Concise architecture-oriented label. |
| Statement | Normative outcome-oriented responsibility. |
| Source trace | GEAIRM/external authority/quality/control/pattern source. |
| Applicability | Conditions under which the responsibility applies. |
| Required outcome | What must be true architecturally. |
| Interaction implications | Boundaries that may need exchange, invocation, delegation or observation. |
| Control implications | Whether an evaluable/enforceable/escalatable control point is required. |
| Evidence/observability | What must be observable or evidential. |
| Failure/escalation behavior | Required behavior when the responsibility cannot be satisfied. |
| Lifecycle/currentness | Version/change/currentness expectations where relevant. |
| Cross-cutting scope | Whether the responsibility spans more than one plane. |
GEAIRA-M02-R07 - Responsibility atomicity: A responsibility SHOULD express one coherent architectural outcome; unrelated obligations SHALL NOT be bundled merely to reduce the number of responsibility statements.
GEAIRA-M02-R08 - Responsibility independence from organization: Responsibilities SHALL NOT be defined by current team, department, funding line or operating-model ownership.
GEAIRA-M02-R09 - Responsibility independence from deployment: Responsibilities SHALL remain valid whether realized centrally, distributed, embedded, federated, at the edge or through human/governance process.
GEAIRA-M02-R10 - Evidence explicitness: Where satisfaction of a responsibility materially affects assurance, governance, authorization, safety or enterprise action, the responsibility SHALL state the applicable evidence/observability obligation.
GEAIRA-M02-R11 - Failure explicitness: Where failure to satisfy a responsibility could create unsafe, unauthorized or materially misleading behavior, the responsibility SHALL define fail-safe, fail-closed, degraded-mode or escalation expectations at the appropriate architecture level.
12.4 Responsibility Taxonomy
GEAIRA uses a non-exclusive responsibility taxonomy. A responsibility has one primary type and MAY carry secondary tags where this improves architecture analysis.
| Type | Architectural intent |
|---|---|
| Enablement | Make a required capability/resource/context available for use. |
| Mediation | Govern access, invocation or use between participants. |
| Coordination | Arrange participation, ordering, composition, federation or orchestration. |
| Authority & Control | Establish, evaluate or enforce authority, permission, policy or safety boundaries. |
| Contract & Interoperability | Preserve obligations at architectural boundaries and peer-architecture interfaces. |
| Observation & Evidence | Make material events, decisions, effects and state observable and evidential. |
| Evaluation & Assurance | Evaluate targets and support assurance conclusions without collapsing evaluation into assurance. |
| Lifecycle & Evolution | Manage identity/version/lifecycle/currentness/change responsibilities. |
| Resilience & Safety | Preserve safe and predictable behavior under failure, uncertainty or degraded operation. |
| Human & Governance | Represent review, approval, intervention, escalation and governance participation explicitly. |
GEAIRA-M02-R12 - Taxonomy non-exclusivity: Responsibility types are architectural classifications and SHALL NOT be interpreted as mutually disjoint ontology classes or mandatory service families.
GEAIRA-M02-R13 - Cross-plane responsibility: A responsibility MAY span multiple architecture planes and SHALL retain one identity rather than be duplicated solely to give each plane a local copy.
12.5 Responsibility Applicability and Obligation Strength
GEAIRA distinguishes the strength of an architectural responsibility from the physical form used to realize it.
| Applicability class | Meaning |
|---|---|
| Baseline-mandatory | Required for every conformant GEAIRA realization. |
| Conditional-mandatory | Required whenever its stated trigger/condition is present. |
| Pattern-mandatory | Required when a canonical GEAIRA pattern that declares it is selected. |
| Profile-mandatory | Required by a valid specialization/profile in addition to the generic baseline. |
| Optional enhancement | Valuable but not required for conformance unless promoted by a profile/pattern. |
GEAIRA-M02-R14 - Obligation/implementation separation: A mandatory responsibility requires demonstrable satisfaction of the architectural outcome; it does not automatically require a dedicated standalone ABB or component.
GEAIRA-M02-R15 - Conditional precision: Conditional-mandatory responsibilities SHALL define a testable architectural trigger rather than rely on vague phrases such as “where appropriate”.
GEAIRA-M02-R16 - Profile strengthening only: A profile MAY make a generic conditional or optional responsibility mandatory in its scope and MAY tighten its contract, but SHALL NOT weaken a baseline-mandatory GEAIRA responsibility.
12.6 Responsibility Aggregation and Separation
Responsibilities may be grouped for architecture comprehension, but aggregation must preserve independently important obligations.
Two responsibilities SHOULD remain distinct when any of the following differ materially:
- semantic/external authority source;
- applicability condition;
- governing authority;
- evidence obligation;
- failure behavior;
- lifecycle/currentness behavior;
- interaction contract;
- control/enforcement point;
- assurance or regulatory treatment.
Responsibilities MAY be aggregated when they share a cohesive outcome, the same applicability and governance boundary, and gain no architecture value from independent treatment.
GEAIRA-M02-R17 - No semantic collapse through aggregation: Responsibility aggregation SHALL NOT erase a GEAIRM hard distinction such as capability vs authority, context vs environment, evaluation vs assurance, or decision vs authorization.
GEAIRA-M02-R18 - No duplicate responsibility by view: The same responsibility SHALL NOT be cloned under multiple views/planes when cross-reference is sufficient.
12.7 ABB Admission Model
An ABB is admitted only when naming a technology-neutral responsibility-bearing boundary materially improves architecture reasoning, interoperability, governance, reuse or conformance.
A candidate ABB MUST satisfy all of these foundational admission criteria:
- Responsibility basis — one or more explicit Architectural Responsibilities justify it.
- Cohesive purpose — its responsibilities form a coherent architectural purpose.
- Technology neutrality — its definition remains meaningful across different implementation technologies.
- Traceable boundary — its purpose and contracts can be traced to semantic/external obligations.
- Non-duplication — an existing GEAIRA/peer-architecture ABB does not already own the same generic architectural responsibility.
- Realization independence — its definition does not depend on being a separate process, service, store, registry, model server, gateway or database.
A candidate should additionally demonstrate at least one architecture-significance reason, such as:
- a meaningful interaction/contract boundary;
- an independently governable responsibility boundary;
- a reusable responsibility across multiple patterns/contexts;
- a distinct authority/control point;
- an independently relevant lifecycle/currentness boundary;
- a meaningful resilience/failure-isolation concern;
- a peer-architecture interoperability boundary;
- useful substitution/specialization value in profiles.
GEAIRA-M02-R19 - ABB admission gate: A canonical ABB SHALL satisfy all foundational admission criteria and demonstrate architecture significance beyond merely naming a semantic concept.
GEAIRA-M02-R20 - No registry reflex: The existence of governed metadata SHALL NOT by itself justify a Registry ABB; registry-style realization is an implementation choice unless independent architecture responsibilities justify a canonical registry boundary.
GEAIRA-M02-R21 - No store reflex: The need to retain context, evidence, memory, provenance or lifecycle information SHALL NOT by itself justify a dedicated Store ABB.
GEAIRA-M02-R22 - No service reflex: A relationship that requires interaction SHALL NOT by itself justify a dedicated Service ABB.
GEAIRA-M02-R23 - No productization reflex: A canonical ABB SHALL NOT imply that the ABB itself is a Product, AI Product or independently consumable service.
12.8 ABB Non-Admission and Anti-Inflation Rules
The following are normally insufficient, on their own, to justify a canonical ABB:
a GEAIRM concept name
an ontology relationship verb
a document or data structure
a database/table/index
an API endpoint
an event topic/queue
an SDK/library
an vendor product
an implementation adapter
an operational runbook
an organizational team
an architecture view
an implementation-specific registry/store/cache
A concern may remain a distributed/embedded responsibility when its architectural outcome is important but no stable independent ABB boundary is justified.
GEAIRA-M02-R24 - Embedded responsibility legitimacy: GEAIRA SHALL allow responsibilities to remain embedded/distributed rather than manufacture an ABB solely to make diagrams symmetrical.
GEAIRA-M02-R25 - Granularity balance: ABBs SHALL avoid both “mega-ABB” aggregation that hides independently governable responsibilities and “micro-ABB” fragmentation that merely mirrors individual semantic nouns or technical functions.
12.9 ABB Status and Specialization
GEAIRA distinguishes the following architecture statuses:
| Status | Meaning |
|---|---|
| Responsibility only | Architecture outcome is required/recognized but no canonical ABB is justified. |
| Candidate ABB | Proposed boundary under analysis; not yet canonical. |
| Canonical ABB | Admitted by GEAIRA after responsibility and admission analysis. |
| Profile specialization | A valid specialization of a canonical ABB/responsibility for a scoped context. |
| Realization component | Implementation-specific component/service/store/process that realizes one or more responsibilities/ABBs; not itself made canonical by implementation existence. |
GEAIRA-M02-R26 - Candidate non-normativity: A Candidate ABB SHALL NOT be presented as a mandatory GEAIRA building block until admitted by the canonical architecture process.
GEAIRA-M02-R27 - Profile identity preservation: A profile specialization SHALL preserve traceability to the generic responsibility/ABB it specializes and SHALL NOT silently redefine its purpose.
12.10 ABB Realization Cardinality
Logical architecture and physical realization are deliberately many-to-many.
Valid realization patterns include:
1 ABB -> 1 component
1 ABB -> many components
many ABBs -> 1 consolidated component
many ABBs -> many cooperating components
ABB responsibility -> embedded library/runtime capability
ABB responsibility -> managed platform capability
ABB responsibility -> human/governance process
GEAIRA-M02-R28 - Many-to-many realization: GEAIRA SHALL permit one-to-many and many-to-one mappings between canonical ABBs and realization components.
GEAIRA-M02-R29 - Consolidation traceability: When several ABBs are consolidated into one realization component, the implementation SHALL preserve traceability to each responsibility and its independent contracts/control obligations.
GEAIRA-M02-R30 - Split traceability: When one ABB is realized by multiple components, the implementation SHALL demonstrate how the distributed realization collectively satisfies the ABB responsibilities and failure behavior.
GEAIRA-M02-R31 - Co-location non-collapse: Co-location of responsibilities in one component SHALL NOT imply semantic or architectural equivalence of those responsibilities.
12.11 Architectural Interaction Model
M02 defines generic interaction kinds without freezing the M11 inter-plane interaction catalogue.
| Interaction kind | Architectural meaning |
|---|---|
| Invocation | One responsibility invokes or requests behavior from another. |
| Provision / Exchange | Information, context, evidence, capability or state is supplied across a boundary. |
| Delegation / Authority | Authority or decision rights are established, constrained or referenced across a boundary. |
| Control / Enforcement | A policy, safety or governance condition is evaluated/enforced/escalated. |
| Observation / Evidence | Events, decisions, effects or state are exposed for observability/evidence. |
| Lifecycle / Change | Version, currentness, supersession or change information crosses a boundary. |
| Coordination | Participants are composed, federated, orchestrated or otherwise coordinated. |
These kinds are conceptual architecture interactions, not protocols.
GEAIRA-M02-R32 - Interaction necessity: Canonical architectural interactions SHALL be introduced only where an architectural boundary obligation exists; GEAIRA SHALL NOT mechanically mirror all 41 GEAIRM semantic relationships as architecture interactions.
GEAIRA-M02-R33 - Protocol neutrality: An Architectural Interaction SHALL NOT prescribe REST, gRPC, messaging, shared memory, database access, event streaming or any other transport unless specialized by a later profile/realization.
12.12 Interaction Contract Model
Where an architectural interaction is material, its contract SHOULD state the applicable subset of:
- semantic meaning and identifiers;
- producer/provider responsibility;
- consumer/recipient responsibility;
- authority/authentication/authorization expectations;
- input/output or resource expectations;
- policy/governance constraints;
- lifecycle/version/currentness expectations;
- provenance/traceability expectations;
- evidence/telemetry obligations;
- failure, retry, rejection or escalation behavior;
- confidentiality/integrity/safety constraints;
- compatibility/evolution rules.
GEAIRA-M02-R34 - Contract completeness proportionality: Interaction contracts SHALL be as explicit as required by architectural risk/significance, but GEAIRA SHALL NOT require irrelevant contract fields merely for template completeness.
GEAIRA-M02-R35 - Contract evolution safety: Changes to an interaction contract SHALL preserve required semantic meaning and SHALL define compatibility/migration expectations where incompatible evolution could affect consumers or control obligations.
12.13 Architectural Control Point Model
An Architectural Control Point is justified when an obligation must be evaluated, enforced, recorded, approved or escalated at a logical architecture boundary.
A control point SHOULD identify:
trigger / scope
obligation being evaluated
required inputs / evidence
applicable authority or policy source
allow / deny / constrain / escalate outcome
failure/default behavior
evidence to record
human involvement where applicable
A control point MAY be:
- embedded in an existing ABB/component;
- shared across several ABBs;
- distributed across a workflow;
- implemented by a human/governance process;
- specialized into a dedicated service by a profile/realization.
GEAIRA-M02-R36 - Control-point non-service rule: A canonical Control Point SHALL NOT imply a standalone control service unless later ABB admission analysis independently justifies such a boundary.
GEAIRA-M02-R37 - Authority-source traceability: Every authority/policy-sensitive control point SHALL identify the governing authority/policy source whose obligation it evaluates or enforces.
GEAIRA-M02-R38 - Fail-safe control behavior: A control point that protects authorization, safety or governance boundaries SHALL define behavior for unavailable, indeterminate or stale control information rather than silently defaulting to permissive continuation.
12.14 Architecture Traceability Chain
GEAIRA requires a traceable chain from meaning to realization:
GEAIRM / external authority / architecture-quality obligation
|
v
Architectural Responsibility
|
+----------+----------+
| |
v v
ABB (if justified) distributed/embedded responsibility
|
+--> Architectural Interaction
+--> Interaction Contract
+--> Control Point
+--> Evidence / Failure obligations
|
v
Pattern / Profile / Realization
Traceability must also work in reverse: an implementation component claiming GEAIRA alignment should be able to identify which canonical responsibility/ABB it realizes.
GEAIRA-M02-R39 - Bidirectional traceability: Canonical GEAIRA ABBs and materially significant interactions/control points SHALL be traceable back to their source responsibilities, and architecture realizations claiming conformance SHALL be traceable forward from responsibilities to their realization.
GEAIRA-M02-R40 - No orphan canonical ABB: A canonical ABB with no current responsibility trace is invalid and SHALL be removed, reclassified as realization-only, or explicitly deprecated through later governance.
12.15 Worked Responsibility Derivations
The following examples illustrate M02 method. They are not a final ABB catalogue.
| Semantic input | Architectural responsibility | What is not automatically implied |
|---|---|---|
geairm:hasAIContext | An AI Execution must be able to obtain an identifiable, bounded and applicable execution context where context is required. | Context Service, context database, prompt server. |
geairm:retrievesResource | Retrieval used by AI execution must preserve resource identity/authority/currentness obligations applicable to the retrieval context. | Vector database, search engine, embedding store. |
geairm:usesTool | Tool invocation must be mediated according to applicable identity, authority, policy, input/output and evidence obligations. | Tool Registry, plugin gateway, agent framework. |
geairm:authorizedBy | Where authorization is semantically required, the architecture must make an applicable authorization determinable before the governed decision/action is committed. | Standalone Authorization Service. |
geairm:usesEvidence | Material evaluation/assurance paths must expose and preserve sufficient evidence references for their governing context. | Evidence Store or evidence database. |
geairm:participatesInControlLoop | Closed-loop/feedback realization must make sensing/action/control responsibilities and applicable safety/oversight boundaries explicit. | A single Control Loop component. |
geairm:supersedes | Version/evolution architecture must preserve supersession/currentness relationships without mutating historical truth. | Source-control system or mandatory registry. |
| GEAIRM Product/Agent/Execution distinctions | Architecture must keep Product responsibility, Agent identity and execution/runtime responsibility independently traceable even if co-hosted. | Three mandatory physical services. |
GEAIRA-M02-R41 - Worked-example non-canonicality: M02 worked examples demonstrate derivation method and SHALL NOT be interpreted as freezing named ABBs before the relevant plane slices perform admission analysis.
12.16 ABB Anti-Patterns
M02 establishes the following architecture anti-patterns for later slices to avoid:
- Ontology-as-services — one service per semantic concept.
- Relationship-as-API — one mandatory API for every semantic relationship.
- Registry-everything — every governed term/resource becomes a registry ABB.
- Store-everything — every persistent concern becomes a dedicated store ABB.
- Vendor abstraction theatre — a vendor product renamed as a supposedly generic ABB.
- Mega-AI Platform ABB — all enterprise-AI responsibilities hidden behind one undifferentiated box.
- Micro-ABB explosion — every small function represented as a canonical ABB.
- Control-point inflation — every policy check drawn as an independent service.
- Team-boundary architecture — ABBs aligned to current org charts rather than architectural responsibility.
- Implementation leakage — transport/storage/runtime choices encoded as generic reference-architecture requirements.
GEAIRA-M02-R42 - Anti-pattern rejection: Later GEAIRA slices SHALL reject candidate ABB decompositions that materially exhibit these anti-patterns unless a specific scoped profile provides independent architectural justification.
12.17 M02 Decisions
M02 freezes the following architectural modelling decisions for GEAIRA 0.1:
responsibility is the primary unit of architectural obligation
responsibility != component
responsibility != ABB automatically
semantic concept != ABB automatically
semantic relationship != interaction automatically
mandatory responsibility != mandatory standalone component
canonical ABB != Product
canonical ABB != deployment unit
control point != service automatically
logical ABB <-> physical realization is many-to-many
profiles may strengthen but not weaken mandatory baseline responsibilities
all canonical ABBs require responsibility traceability
M02 deliberately does not freeze:
- the final seven-plane structure;
- a canonical ABB inventory;
- inter-plane interaction catalogue;
- canonical architecture patterns;
- deployment topology;
- technology choices.
Those remain work for M03-M17.
13. M02 Completion State
GEAIRA-M02 — COMPLETE.
M02 establishes:
- a five-source responsibility-derivation model;
- a canonical Architectural Responsibility statement model;
- ten non-exclusive responsibility types;
- five applicability/obligation classes;
- ABB aggregation/separation rules;
- six foundational ABB-admission criteria plus architecture-significance tests;
- explicit ABB non-admission and anti-inflation rules;
- five ABB/status distinctions;
- many-to-many ABB realization rules;
- seven generic Architectural Interaction kinds;
- an Interaction Contract model;
- an Architectural Control Point model;
- a bidirectional architecture traceability chain;
- eight worked semantic-to-architecture derivations;
- ten ABB anti-patterns;
- 42 normative M02 rules.
M02 introduces no final canonical ABB catalogue and makes no change to GEAIRM semantics.
Next slice
GEAIRA-M03 — Plane Model & Cross-Cutting Architecture
M03 will test and refine the seven-plane hypothesis using the M02 responsibility model, identify cross-cutting responsibilities that must not be trapped inside one plane, and freeze the first canonical GEAIRA architecture decomposition for detailed plane work in M04-M10.
14. GEAIRA-M03 - Plane Model & Cross-Cutting Architecture
14.1 Purpose
GEAIRA-M03 tests the seven-plane hypothesis introduced by M01 using the responsibility and ABB discipline established by M02. The objective is not to create seven large components. The objective is to establish a stable architecture decomposition that helps organize responsibility, interactions, views and later ABB admission without confusing architecture planes with deployment tiers, ontology classes, Products, teams or platform services.
M03 answers six questions:
- Does each proposed plane represent a coherent and architecture-significant responsibility area?
- Are the plane boundaries distinguishable without breaking end-to-end enterprise-AI behavior?
- Can the decomposition survive centralized, distributed, federated, edge and mixed realizations?
- Which responsibilities are primarily anchored in a plane, and which legitimately span several planes?
- Which concerns must be represented as cross-cutting architectural overlays rather than forced into one plane?
- Does the resulting model cover the frozen GEAIRM semantic baseline without creating a one-to-one semantic-to-architecture projection?
GEAIRA-M03-R01 - Plane purpose: Architecture planes SHALL organize coherent architectural responsibility and SHALL NOT be introduced merely to make diagrams visually symmetrical.
GEAIRA-M03-R02 - Plane test before freeze: A plane SHALL be retained in the canonical GEAIRA decomposition only if it satisfies the M03 plane-evaluation criteria.
14.2 Plane-Evaluation Criteria
A candidate plane is acceptable only when it satisfies all of the following tests.
| Criterion | Test |
|---|---|
| Responsibility cohesion | The plane groups responsibilities that answer a coherent architectural question. |
| Architectural distinguishability | The plane can be distinguished from adjacent planes without relying on vendor/product boundaries. |
| Realization independence | The plane remains meaningful whether its responsibilities are centralized, distributed, embedded, federated, edge-hosted or human-operated. |
| Semantic coverage | The plane contributes to realization of the GEAIRM baseline while preserving semantic authority and hard distinctions. |
| Cross-cutting openness | The plane can accept cross-cutting obligations without claiming exclusive ownership of them. |
| Interaction clarity | Material hand-offs to other planes can be expressed as architectural interactions/contracts rather than hidden coupling. |
| Stability | The plane is likely to remain useful across technology generations and organization changes. |
GEAIRA-M03-R03 - All-criteria admission: A canonical plane SHALL satisfy all seven plane-evaluation criteria.
GEAIRA-M03-R04 - No technology-defined plane: A plane SHALL NOT be defined by a cloud product, model family, database category, orchestration framework, API style or deployment technology.
GEAIRA-M03-R05 - No organization-defined plane: A plane SHALL NOT be defined by a current team, department, committee, funding line or delivery structure.
GEAIRA-M03-R06 - No ontology-defined plane: GEAIRM semantic domains D1-D8 SHALL NOT be mechanically converted into GEAIRA planes.
14.3 Canonical GEAIRA 0.1 Plane Model
The M03 analysis retains seven planes. One working label is refined to make the physical/digital environment responsibility explicit: P4 becomes AI Execution, Agentic & Environment.
| Plane | Canonical label | Primary architectural question | Primary responsibility focus |
|---|---|---|---|
| P1 | AI Intent, Purpose & Authority | Why may AI be used, for what purpose, and within which authority/policy envelope? | Intent, purpose, permitted/prohibited use, policy source, delegated-authority envelope, governance/risk context and architectural constraints derived from them. |
| P2 | Knowledge, Context & Grounding | What information, knowledge and execution context must be available and trustworthy enough for the applicable use? | AI Context, resource provision, retrieval, grounding, memory, source identity, contextual applicability and context-boundary obligations. |
| P3 | Intelligence Capability & Primitive | Which intelligence capabilities and qualifying mechanisms must be available, selectable, governable and compatible? | Capability availability, primitive/model/mechanism access, selection, compatibility, version/currentness inputs and capability-to-execution readiness. |
| P4 | AI Execution, Agentic & Environment | How are AI executions and participating Agents/tools/environments coordinated at runtime? | AI Execution, Agent participation, tool use, runtime coordination, environment interaction, sensing/actuation responsibilities and control-loop participation. |
| P5 | AI Serving, Product & Integration | How are enterprise-AI capabilities exposed, integrated and consumed while preserving Product/service boundaries? | Serving/exposure, integration contracts, Product/AI Product architectural alignment, consumption boundaries, composition/federation/orchestration integration and peer-system connectivity. |
| P6 | Decision, Action & Human Oversight | How are decision formation, point-of-use authorization, action and human participation architected? | Recommendation/decision/action paths, authorization at governed commit points, review, approval, intervention, escalation, override and action safety boundaries. |
| P7 | Evaluation, Assurance, Observability & Evolution | How are AI behavior, evidence, assurance, operational state and change made assessable and governable over time? | Evaluation, evidence, assurance support, observability, traceability, lifecycle, version/currentness, supersession, change and continuous evolution responsibilities. |
GEAIRA-M03-R07 - Seven-plane freeze: GEAIRA 0.1 SHALL use P1-P7 above as its canonical human-readable architecture-plane decomposition for M04-M17.
GEAIRA-M03-R08 - Plane label stability: Later GEAIRA 0.1 slices MAY refine descriptive wording inside a plane but SHALL NOT create, remove, merge or renumber a canonical plane without explicit M17 reconciliation and justification.
GEAIRA-M03-R09 - P4 refinement: The canonical P4 label SHALL be AI Execution, Agentic & Environment; environment/control-loop responsibility is explicit in P4 scope rather than implied by the shorter M01 working label.
GEAIRA-M03-R10 - Plane non-tier: P1-P7 SHALL NOT be interpreted as seven mandatory runtime tiers, network zones, process stages or deployment layers.
GEAIRA-M03-R11 - Plane non-sequence: P1-P7 numbering is organizational and SHALL NOT imply a mandatory execution sequence from P1 through P7.
GEAIRA-M03-R12 - Plane non-ownership: Assignment of a responsibility to a plane SHALL NOT transfer semantic, Product, policy, data or organizational ownership to that plane.
GEAIRA-M03-R13 - Plane non-product: A plane SHALL NOT be treated as a Product, AI Product, Agent, runtime or deployable application merely because it is shown as a named architecture region.
14.4 Plane Assignment and Responsibility Identity
GEAIRA assigns responsibility to planes for architecture organization while preserving the M02 rule that responsibility identity is more fundamental than diagram placement.
The assignment model is:
Architectural Responsibility
|
+--> primaryPlane: zero or one P1-P7
|
+--> participatingPlanes: zero or more P1-P7
|
+--> crossCuttingOverlays: zero or more X1-X8
|
+--> ABB realization: zero or more, only if M02 admission is satisfied
A responsibility normally has one primary plane when a natural architectural anchor exists. A genuinely cross-plane responsibility may have no exclusive primary plane.
GEAIRA-M03-R14 - Single responsibility identity: A cross-plane responsibility SHALL retain one canonical identity and SHALL NOT be copied into separate pseudo-responsibilities merely to populate each plane.
GEAIRA-M03-R15 - Primary-plane anchor: Where a natural anchor exists, a responsibility SHOULD identify one primary plane and MAY identify other participating planes.
GEAIRA-M03-R16 - Cross-plane allowance: A responsibility MAY span any number of planes when its architectural outcome genuinely crosses their boundaries.
GEAIRA-M03-R17 - ABB multi-plane allowance: A canonical ABB MAY realize responsibilities associated with multiple planes; plane boundaries SHALL NOT force ABB duplication.
GEAIRA-M03-R18 - One-plane/many-ABB allowance: A plane MAY contain responsibilities realized by many ABBs, embedded capabilities, controls and human/governance mechanisms; a plane SHALL NOT imply one canonical component.
GEAIRA-M03-R19 - Boundary interaction explicitness: Where a responsibility crosses a material plane boundary, the relevant interaction/contract/control obligation SHALL be made explicit in M11 or the applicable plane slice.
GEAIRA-M03-R20 - Co-location neutrality: Responsibilities from different planes MAY be co-located in one realization component without collapsing their architectural distinction.
GEAIRA-M03-R21 - Distribution neutrality: Responsibilities from one plane MAY be distributed across several components or locations without fragmenting the canonical plane identity.
14.5 Canonical Cross-Cutting Architecture Overlays
Some architecture obligations recur across multiple planes and would be distorted if assigned exclusively to one plane. M03 therefore defines eight Cross-Cutting Architecture Overlays.
An overlay is not an eighth/eighth-plus architecture plane. It is an orthogonal concern that constrains or enriches responsibilities, ABBs, interactions, contracts, control points and patterns across applicable planes.
| ID | Cross-cutting overlay | Architectural concern |
|---|---|---|
| X1 | Security, Privacy & Trust Boundaries | Confidentiality, integrity, isolation, privacy, trust boundaries, protected processing and security-relevant failure behavior. |
| X2 | Identity, Authority & Policy Enforcement | Actor/resource identity, authority context, policy binding, authorization inputs, delegated authority and enforcement evidence. |
| X3 | Evidence, Provenance & Traceability | Traceable inputs, outputs, decisions, actions, dependencies, provenance references, evidence chains and historical truth. |
| X4 | Observability, Telemetry & Auditability | Material runtime/operational signals, auditability, monitoring, diagnostic context and evidence-producing telemetry. |
| X5 | Resilience, Safety & Failure Management | Availability, degradation, fail-safe/fail-closed behavior, containment, recovery, uncertainty handling and operational/physical safety. |
| X6 | Lifecycle, Version, Currentness & Change | Identity/version alignment, lifecycle state, currentness, compatibility, supersession, migration and change control. |
| X7 | Interoperability, Contracts & Semantic Compatibility | Boundary contracts, semantic compatibility, interface obligations, protocol-neutral exchange and peer-architecture alignment. |
| X8 | Governance, Human Accountability & Compliance | Human accountability, governance participation, review/escalation, policy/control evidence and regulatory/enterprise obligation alignment. |
GEAIRA-M03-R22 - Overlay orthogonality: X1-X8 SHALL be treated as cross-cutting overlays and SHALL NOT be promoted into additional numbered planes solely because they span the architecture.
GEAIRA-M03-R23 - Overlay non-service: A cross-cutting overlay SHALL NOT imply a mandatory shared service, platform, registry, gateway or control plane.
GEAIRA-M03-R24 - Overlay applicability: An overlay applies only where its concern is architecturally relevant; GEAIRA SHALL NOT require meaningless boilerplate controls merely to mark every overlay on every responsibility.
GEAIRA-M03-R25 - Overlay traceability: Material overlay obligations SHALL be traceable to the affected responsibilities, interactions, control points, ABBs or patterns rather than existing only as diagram decorations.
GEAIRA-M03-R26 - No duplicate overlay responsibility: An overlay SHALL strengthen or qualify an existing responsibility where appropriate rather than creating redundant copies of the same architectural obligation.
GEAIRA-M03-R27 - Profile strengthening: A valid GEAIRA profile MAY strengthen overlay obligations for its context but SHALL NOT weaken mandatory generic responsibilities or semantic hard distinctions.
14.6 Overlay Applicability Across Planes
The following matrix is a coverage guide, not a cardinality rule. "Primary pressure" means the concern is especially architecture-significant in that plane; "recurring" means it commonly affects responsibilities there. Omission does not prohibit applicability.
| Overlay | Primary pressure | Recurring applicability |
|---|---|---|
| X1 Security, Privacy & Trust | P2, P4, P5, P6 | P1, P3, P7 |
| X2 Identity, Authority & Policy Enforcement | P1, P4, P6 | P2, P3, P5, P7 |
| X3 Evidence, Provenance & Traceability | P4, P6, P7 | P1, P2, P3, P5 |
| X4 Observability, Telemetry & Auditability | P4, P5, P6, P7 | P2, P3 |
| X5 Resilience, Safety & Failure Management | P4, P5, P6 | P2, P3, P7 |
| X6 Lifecycle, Version, Currentness & Change | P3, P5, P7 | P1, P2, P4, P6 |
| X7 Interoperability, Contracts & Semantic Compatibility | P2, P3, P5 | P1, P4, P6, P7 |
| X8 Governance, Human Accountability & Compliance | P1, P6, P7 | P2, P3, P4, P5 |
GEAIRA-M03-R28 - Matrix non-exclusivity: The overlay/plane matrix is informative coverage guidance and SHALL NOT be interpreted as forbidding an overlay from applying to another plane when justified.
GEAIRA-M03-R29 - Overlay proportionality: Overlay obligations SHALL be proportionate to architecture significance, risk, authority, sensitivity and failure consequence.
14.7 Primary Plane vs Cross-Cutting Overlay
A concern can have a natural architectural anchor while also remaining cross-cutting.
Authorization example
P1
establishes intent, policy source and authority envelope
P6
applies point-of-use authorization before governed decision/action commitment
X2
carries identity/authority/policy obligations across relevant P1-P7 boundaries
This does not create two competing meanings of Authorization. It separates architecture responsibility by context.
Evidence example
P7
anchors evaluation/evidence/assurance architecture
X3
requires traceable evidence and provenance where material across other planes
Lifecycle/currentness example
P7
anchors lifecycle/currentness/evolution assessment and governance
X6
propagates version/currentness/change obligations into capability, context,
serving, interaction and action responsibilities where applicable
Policy/governance example
P1
anchors purpose, governing intent and authority/policy context
X2/X8
carry enforceable authority, accountability and governance obligations
through execution, serving, decision/action and assurance boundaries
GEAIRA-M03-R30 - Anchor/cross-cut distinction: A concern MAY have a primary plane anchor and simultaneously impose cross-cutting obligations; these are complementary architecture roles, not duplicated concepts.
GEAIRA-M03-R31 - No catch-all plane: Cross-cutting obligations SHALL NOT be dumped into P1 or P7 merely because those planes contain governance/assurance concerns.
14.8 GEAIRM-to-GEAIRA Coverage Mapping
GEAIRA is intentionally not a one-to-one architectural projection of the eight GEAIRM semantic domains. The following mapping demonstrates coverage while preserving the semantic/architecture boundary.
| GEAIRM domain | Principal GEAIRA plane coverage | Architectural interpretation |
|---|---|---|
| D1 Core AI Semantics | P1, P3, P4, P5 | Purpose/intent constrains use; capability/primitive is organized in P3; execution in P4; Product/serving boundary in P5. |
| D2 Mechanism, Execution & Effect | P3, P4, P6, P7 | Mechanism/capability readiness, runtime execution, effect-to-decision/action implications and evaluation/observation. |
| D3 Context, Knowledge, Grounding & Memory | P2, P4, P7 | Context construction/provision in P2; consumption during execution in P4; evidence/currentness/evaluation concerns in P7. |
| D4 Agency, Authority, Decision & Action | P1, P4, P6 | Authority envelope in P1; Agent/runtime participation in P4; decision/authorization/action/oversight in P6. |
| D5 Environment, Embodiment & Control | P4, P6, P7 | Environment/control-loop participation in P4; action/oversight safety in P6; observability/evaluation/evolution in P7. |
| D6 Dependency, Composition, Product & Topology | P4, P5, M11 cross-plane model | Runtime composition/coordination, Product/integration boundaries and explicit inter-plane contracts. |
| D7 Evaluation, Assurance, Risk & Governance | P1, P6, P7 | Governing/risk context, action/oversight controls and evaluation/assurance/evidence architecture. |
| D8 Identity, Lifecycle, Currentness, Provenance & Change | P5, P7 with X3/X6 | Serving/version compatibility plus lifecycle/currentness/evolution, with provenance/change obligations cross-cutting. |
GEAIRA-M03-R32 - Coverage without ownership transfer: Mapping a GEAIRM domain into one or more planes SHALL NOT transfer semantic authority to GEAIRA or imply that the plane owns all concepts in that domain.
GEAIRA-M03-R33 - Many-to-many mapping: GEAIRM domains and GEAIRA planes SHALL be treated as many-to-many organizational views rather than corresponding hierarchies.
GEAIRA-M03-R34 - Hard-distinction preservation: Plane decomposition SHALL preserve GEAIRM hard distinctions including Product/Agent/Execution, capability/authority, context/environment, evaluation/assurance, provenance/dependency and lifecycle/currentness distinctions.
14.9 Canonical Inter-Plane Responsibility Hand-Offs
M03 does not yet freeze the M11 interaction catalogue, but it establishes the principal hand-off questions that M04-M10 and M11 must resolve.
| Boundary | Architectural hand-off question |
|---|---|
| P1 -> P2/P3/P4/P5/P6/P7 | How are applicable purpose, policy, authority and governance constraints made available and enforceable where needed? |
| P2 -> P4 | How is bounded, applicable context supplied to execution while preserving source identity, currentness, access and grounding obligations? |
| P3 -> P4 | How are required intelligence capabilities/primitives made invocable and compatible with the execution context? |
| P4 < - > P5 | How do serving/integration boundaries invoke or expose execution while preserving Product/service identity and contracts? |
| P4 < - > P6 | How do execution effects, recommendations, plans or runtime state participate in decision/action paths without creating a mandatory universal pipeline? |
| P6 -> P4/P5/external environment | How are authorized actions committed, mediated or rejected and how are outcomes returned/observed? |
| P2/P3/P4/P5/P6 -> P7 | What evaluation, evidence, observability, lifecycle/currentness and assurance-relevant information must be exposed? |
| P7 -> P1/P2/P3/P4/P5/P6 | How do findings, currentness/change and assurance outcomes constrain or inform continued operation without rewriting historical semantic truth? |
GEAIRA-M03-R35 - Hand-off contract requirement: Material inter-plane hand-offs SHALL eventually be represented by Architectural Interactions and Interaction Contracts consistent with M02 rather than by implicit shared-state assumptions.
GEAIRA-M03-R36 - No universal plane pipeline: GEAIRA SHALL NOT require all enterprise-AI scenarios to traverse P1-P7 in numeric order or to touch every plane at runtime.
GEAIRA-M03-R37 - Asynchronous/distributed allowance: Plane interactions MAY be synchronous, asynchronous, event-driven, embedded, federated, human-mediated or otherwise specialized by realization; M03 defines responsibility boundaries, not protocols.
GEAIRA-M03-R38 - Feedback allowance: Inter-plane interactions MAY form feedback/control loops; the plane model SHALL NOT be interpreted as a one-way layered stack.
14.10 Plane and Cross-Cutting Anti-Patterns
M03 adds the following decomposition anti-patterns to the M02 ABB anti-pattern catalogue:
- Seven-tier deployment stack — treating P1-P7 as mandatory physical tiers.
- Plane-per-team — assigning each plane to a department and redefining architecture around the org chart.
- Cross-cutting-as-platform — assuming every overlay requires one centralized shared platform/service.
- Duplicate ABB per plane — cloning a responsibility/ABB solely so each plane appears self-contained.
- Plane owns semantics — treating a plane as semantic authority for GEAIRM/AIPS/BPS/Agent/Data/Analytics concepts.
- P7 dumping ground — moving every operational, governance or lifecycle concern into P7 instead of preserving primary responsibility elsewhere.
- P1 policy-engine assumption — treating P1 as a mandatory central policy engine rather than an architecture plane for intent/authority responsibility.
- P5 Product redefinition — using serving/integration architecture to redefine BPS Productness or AIPS AI Product qualification.
- P4 agent-framework assumption — treating P4 as a mandate for a specific Agent framework/orchestrator/runtime.
- Overlay checkbox architecture — attaching X1-X8 labels without traceable obligations, control points or contracts.
GEAIRA-M03-R39 - Plane anti-pattern rejection: Later GEAIRA slices SHALL reject decompositions that materially exhibit the M03 plane/cross-cutting anti-patterns unless a scoped profile provides independent architecture justification without weakening generic rules.
14.11 M03 Frozen Decisions
M03 freezes the following architecture-organization decisions for GEAIRA 0.1:
P1 AI Intent, Purpose & Authority
P2 Knowledge, Context & Grounding
P3 Intelligence Capability & Primitive
P4 AI Execution, Agentic & Environment
P5 AI Serving, Product & Integration
P6 Decision, Action & Human Oversight
P7 Evaluation, Assurance, Observability & Evolution
planes organize responsibility; planes are not deployment tiers
plane number does not imply runtime sequence
responsibility identity survives cross-plane participation
ABB boundaries may span planes
cross-cutting overlays are orthogonal concerns, not extra planes
cross-cutting overlay != mandatory shared service
GEAIRM domain -> GEAIRA plane mapping is many-to-many
inter-plane boundaries require explicit architectural contracts when material
The following cross-cutting overlays are canonical for GEAIRA 0.1:
X1 Security, Privacy & Trust Boundaries
X2 Identity, Authority & Policy Enforcement
X3 Evidence, Provenance & Traceability
X4 Observability, Telemetry & Auditability
X5 Resilience, Safety & Failure Management
X6 Lifecycle, Version, Currentness & Change
X7 Interoperability, Contracts & Semantic Compatibility
X8 Governance, Human Accountability & Compliance
M03 does not freeze:
- final plane-specific ABB catalogues;
- the canonical M11 inter-plane interaction catalogue;
- deployment topology or runtime sequencing;
- centralized versus distributed realization choices;
- protocol/technology selections;
- canonical patterns, which remain M13-M15 work.
GEAIRA-M03-R40 - Detailed-plane discipline: M04-M10 SHALL use the canonical P1-P7 boundaries and X1-X8 overlays while applying the M02 responsibility/ABB admission rules independently within each plane.
GEAIRA-M03-R41 - No premature ABB freeze: M03 plane freeze SHALL NOT be interpreted as freezing any named ABB before the applicable detailed plane slice performs responsibility and admission analysis.
GEAIRA-M03-R42 - Cross-cutting preservation in later slices: M04-M10 SHALL identify material X1-X8 implications for their responsibilities rather than treating cross-cutting architecture as work deferred entirely to P7.
15. M03 Completion State
GEAIRA-M03 — COMPLETE.
M03 establishes:
- seven canonical architecture planes;
- one refinement of the M01 working labels: P4 is AI Execution, Agentic & Environment;
- seven plane-evaluation criteria;
- a primary-plane/participating-plane assignment model;
- eight canonical cross-cutting architecture overlays;
- a cross-cutting applicability guide;
- a GEAIRM-domain-to-GEAIRA-plane many-to-many coverage map;
- eight principal inter-plane hand-off questions for later contract formalization;
- ten plane/cross-cutting anti-patterns;
- 42 normative M03 rules.
M03 introduces no canonical ABB catalogue, prescribes no deployment topology, and makes no change to GEAIRM semantics.
Next slice
GEAIRA-M04 — P1: AI Intent, Purpose & Authority
M04 will apply the M02 responsibility model and M03 plane/cross-cutting framework to P1, derive the canonical P1 responsibilities, test candidate ABBs, define authority/policy interaction obligations, and distinguish governing intent from point-of-use authorization performed in P6.
16. GEAIRA-M04 - P1: AI Intent, Purpose & Authority
16.1 Purpose of P1
P1 answers the architecture question:
Why is enterprise AI being used, for what intended purpose, under whose authority and accountability, and within which governing boundaries?
P1 establishes the governing envelope within which later GEAIRA planes may supply context, capabilities, execution, serving, decision/action behavior and assurance. It does not itself require a central governance platform and it does not turn every policy, purpose statement, authority grant or risk record into a service.
P1 is principally concerned with:
- intended purpose and permitted-use boundaries;
- authoritative business/governance sponsorship for the AI use;
- accountable human/organizational authority;
- Delegated Authority where Agents operate under bounded mandate;
- applicable policy, governance-obligation and risk context without redefining those external semantics;
- distribution of governing constraints to the planes that must honor them;
- authority currentness, revocation, supersession and escalation;
- evidence that the governing envelope existed and was applicable when material activity occurred.
P1 does not own runtime AI Execution, AI Context construction, capability/primitive selection, Product qualification, point-of-use Authorization, Action, Evaluation or Assurance. Those responsibilities belong principally to other planes and authorities.
GEAIRA-M04-R01 - Governing-envelope responsibility: Every governed enterprise-AI realization SHALL establish sufficient P1 responsibility to determine its intended purpose, governing authority/accountability and applicable operating constraints before the relevant governed use is accepted for operation.
GEAIRA-M04-R02 - No central-governance assumption: P1 responsibilities SHALL remain valid whether realized centrally, federated, embedded in Product/deployment governance, distributed across enterprise control systems or partly through human governance process.
GEAIRA-M04-R03 - No semantic redefinition: P1 SHALL consume GEAIRM, AIPS, BPS, Agent Ontology and applicable policy/governance semantics without redefining Productness, AI Product qualification, Delegated Authority, Authorization, Risk Tier, Governance Obligation, Agent autonomy or organizational/legal authority.
GEAIRA-M04-R04 - Governing intent not runtime pipeline: P1 SHALL establish constraints that later planes can honor; it SHALL NOT imply that every runtime invocation must synchronously call a P1 component.
16.2 Semantic and External Authority Inputs
P1 relies on several externally or semantically authoritative inputs. Architectural use of these inputs does not transfer semantic ownership to GEAIRA.
| Concern | Authority | P1 architectural treatment |
|---|---|---|
| Artificial Intelligence / AI Execution semantics | GEAIRM | Referenced to identify what is being governed; not redefined. |
| Delegated Authority | GEAIRM | Consumed to bound Agent mandate; P1 does not equate it with Authorization. |
| Authorization | GEAIRM | Referenced as a P6 point-of-use permission concept; P1 provides governing basis/constraints, not the occurrence itself. |
| Agent identity/autonomy/control mode | KivaNura Agent Ontology | Referenced where Agent participation is governed; autonomy does not create authority. |
| Product / Product Intrinsic Nature | BPS / PIN | Referenced when an enterprise AI realization is also a Product; P1 does not decide Productness. |
| AI Product qualification / Product Version | AIPS | Referenced where applicable; P1 does not qualify the Product. |
| Agency / Autonomy classification | AIPS | Consumed as Product-Version governance/classification context where applicable. |
| Effective Risk Tier / Risk Driver Assessment | AIPS | Consumed only at valid AIPS Product-Version scope. |
| Governance Obligation / Governance Obligation Profile | AIPS | Consumed as governing obligations; P1 does not replace them with a GEAIRA policy ontology. |
| Policy / legal / regulatory mandate | Applicable external authority | Referenced through authoritative identifiers, rules or obligations appropriate to the enterprise context. |
| Governance Control / control binding | Applicable governance / AIPDS | Referenced where realization or deployment controls are governed externally. |
| Organizational accountability / decision authority | Applicable enterprise/legal authority | Referenced; GEAIRA does not invent a universal organizational authority ontology. |
GEAIRA-M04-R05 - External-authority preservation: P1 SHALL preserve the authority and scope of externally governed Product, risk, policy, Agent and organizational-authority semantics.
GEAIRA-M04-R06 - Product-scope restraint: AIPS Effective Risk Tier, Risk Driver Assessment and Governance Obligation Profile SHALL NOT be generalized through P1 to arbitrary non-Product AI participants.
GEAIRA-M04-R07 - Policy-reference restraint: P1 MAY require authoritative policy/obligation references and architecture constraints, but SHALL NOT define a universal enterprise policy language or control catalogue in GEAIRA 0.1.
GEAIRA-M04-R08 - Autonomy/authority separation: Agent autonomy, Agent Control Mode and AIPS Agency/Autonomy classifications SHALL NOT by themselves establish P1 delegated authority or permission to act.
GEAIRA-M04-R09 - Capability/permission separation: Availability of an AI capability, primitive, Tool or execution path SHALL NOT by itself establish governing authority, permitted use or Authorization.
16.3 P1 Architectural Responsibility Catalogue
M04 freezes twelve canonical P1 Architectural Responsibilities. These responsibilities are technology-neutral outcomes; they do not each imply a separate ABB or component.
| ID | Responsibility | Primary type | Applicability | Required architectural outcome |
|---|---|---|---|---|
| P1-RESP-01 | Intended Purpose Declaration | Human & Governance | Baseline-mandatory | The governed AI use has an explicit intended purpose/scope traceable to an authoritative owner or mandate. |
| P1-RESP-02 | Permitted / Prohibited Use Boundary | Authority & Control | Baseline-mandatory | Material allowed, disallowed and constrained uses can be determined for the governed scope. |
| P1-RESP-03 | Authority Source & Accountability Binding | Human & Governance | Baseline-mandatory | The authoritative source and accountable human/organizational role for the governed AI use are identifiable. |
| P1-RESP-04 | Delegated Authority Envelope | Authority & Control | Conditional-mandatory | Where an Agent operates under delegated mandate, the bounded authority scope, constraints and oversight are explicit. |
| P1-RESP-05 | Applicable Policy & Governance Obligation Projection | Contract & Interoperability | Conditional-mandatory | Applicable external policy/governance obligations can be projected to the architecture without semantic redefinition. |
| P1-RESP-06 | Product / Risk Governance Context Association | Contract & Interoperability | Conditional-mandatory | Where BPS/AIPS Product boundaries apply, the correct Product Version, risk and governance context can be associated with the governed AI use. |
| P1-RESP-07 | Governing Constraint Distribution | Mediation | Baseline-mandatory | Relevant P1 constraints can be made available to P2-P7 responsibility holders without requiring implicit shared state. |
| P1-RESP-08 | Authority Currentness, Revocation & Supersession | Lifecycle & Evolution | Conditional-mandatory | Time-bounded, revocable or superseded authority/constraint state can be detected and propagated where material. |
| P1-RESP-09 | Governing Conflict & Escalation | Human & Governance | Conditional-mandatory | Conflicting, incomplete or indeterminate governing inputs have an explicit precedence/escalation path rather than silent permissive resolution. |
| P1-RESP-10 | Human Accountability & Escalation Envelope | Human & Governance | Baseline-mandatory | Accountable participation, review/escalation responsibilities and material exception ownership are explicit where applicable. |
| P1-RESP-11 | Governing Evidence & Traceability | Observation & Evidence | Baseline-mandatory | Material purpose/authority/constraint decisions are traceable enough for applicable assurance, governance and audit needs. |
| P1-RESP-12 | Assurance / Governance Feedback Intake | Evaluation & Assurance | Conditional-mandatory | Findings, currentness concerns or assurance outcomes that constrain continued use can feed back into governing intent/authority decisions. |
GEAIRA-M04-R10 - Purpose explicitness: P1-RESP-01 SHALL identify intended purpose with enough specificity to distinguish permitted enterprise use from materially different use that would require reassessment.
GEAIRA-M04-R11 - Use-boundary explicitness: P1-RESP-02 SHALL allow material prohibited or constrained uses to be represented without assuming that all constraints are executable machine policy.
GEAIRA-M04-R12 - Accountability explicitness: P1-RESP-03 SHALL identify an authoritative accountability source or role; technical ownership alone SHALL NOT be presumed to establish organizational authority.
GEAIRA-M04-R13 - Delegation conditionality: P1-RESP-04 SHALL apply when an Agent is permitted to operate under a bounded delegated mandate and SHALL preserve the GEAIRM distinction between Delegated Authority and specific Authorization.
GEAIRA-M04-R14 - Governance applicability precision: P1-RESP-05 and P1-RESP-06 SHALL carry explicit applicability conditions so that Product-specific or regulatory obligations are not silently generalized.
GEAIRA-M04-R15 - Downstream constraint availability: P1-RESP-07 SHALL make governing constraints available at architectural boundaries where they are needed, but SHALL NOT mandate a centralized runtime lookup for every invocation.
GEAIRA-M04-R16 - Currentness discipline: When purpose, authority, policy or governance applicability can expire, be revoked or be superseded, P1-RESP-08 SHALL make that state architecture-visible to affected responsibilities.
GEAIRA-M04-R17 - Conflict non-silence: P1-RESP-09 SHALL NOT resolve conflicting authoritative constraints by selecting the least restrictive interpretation merely to preserve availability.
GEAIRA-M04-R18 - Evidence proportionality: P1-RESP-11 evidence obligations SHALL be proportionate to authority, risk, consequence and governance scope; evidence presence SHALL NOT by itself prove governance compliance.
16.4 P1 Hard Distinctions
M04 freezes the following architecture distinctions:
intended purpose != Prompt
intended purpose != Instruction
intended purpose != AI Context
business intent != runtime invocation
permitted use != technical capability
policy obligation != runtime Authorization
Governance Obligation != concrete control occurrence
Delegated Authority != Authorization
Delegated Authority != Agent autonomy
Authorization != AIPS Agency / Autonomy classification
accountability != technical ownership
assurance finding != approval
risk tier != permission
Product qualification != permission to execute every use
P1 governing envelope != P6 point-of-use authorization
P1 constraint projection != mandatory policy-engine call
GEAIRA-M04-R19 - Purpose/context distinction: Intended purpose and permitted-use scope SHALL NOT be modelled as synonyms for GEAIRM AI Context, Prompt or Instruction.
GEAIRA-M04-R20 - Policy/authorization distinction: Policy or Governance Obligation applicability SHALL NOT itself be represented as a concrete GEAIRM Authorization occurrence.
GEAIRA-M04-R21 - Delegation/authorization distinction: P1 Delegated Authority responsibility SHALL remain distinct from P6 point-of-use Authorization responsibility even when both are realized by one component.
GEAIRA-M04-R22 - Risk/permission distinction: Effective Risk Tier or other risk classification SHALL NOT by itself grant or deny a concrete execution/action; it informs applicable governance and control obligations.
GEAIRA-M04-R23 - Accountability/ownership distinction: Product owner, model owner, engineering owner or platform owner SHALL NOT automatically be treated as the accountable organizational authority for every governed AI use.
GEAIRA-M04-R24 - Assurance/approval distinction: Evaluation or Assurance findings SHALL NOT be treated as purpose approval or authority grants unless an applicable governance process explicitly establishes that consequence.
16.5 P1 ABB Admission Analysis
M04 applies the M02 admission criteria to the P1 responsibility catalogue.
16.5.1 Canonical ABB: Governing Intent Boundary
ID: P1-ABB-01
Label: Governing Intent Boundary
Purpose: Provide a technology-neutral architectural boundary through which intended purpose, permitted/prohibited use scope, authoritative sponsorship and lifecycle identity for the governed enterprise-AI use can be established and exposed.
Primary responsibilities:
- P1-RESP-01 Intended Purpose Declaration;
- P1-RESP-02 Permitted / Prohibited Use Boundary;
- the intent-oriented portion of P1-RESP-03;
- P1-RESP-07 Governing Constraint Distribution;
- P1-RESP-11 Governing Evidence & Traceability.
Why it passes the M02 admission gate:
- it has a cohesive governing-intent purpose;
- its contract boundary is meaningful across P2-P7 and adjacent governance architecture;
- intended-purpose/use-boundary identity has independent lifecycle/change significance;
- it is implementation-neutral and reusable across Product, non-Product, Agentic and physical-AI patterns;
- it is independently governable even when co-located with other governance functions.
It does not imply a standalone Intent Service, registry, database or workflow.
16.5.2 Canonical ABB: Authority & Accountability Boundary
ID: P1-ABB-02
Label: Authority & Accountability Boundary
Purpose: Provide a technology-neutral architectural boundary through which the applicable authority source, accountable party/role, delegated mandate, policy/governance references, currentness/revocation state and escalation basis can be established and exposed to dependent responsibilities.
Primary responsibilities:
- P1-RESP-03 Authority Source & Accountability Binding;
- P1-RESP-04 Delegated Authority Envelope;
- P1-RESP-05 Applicable Policy & Governance Obligation Projection;
- P1-RESP-06 Product / Risk Governance Context Association;
- P1-RESP-08 Authority Currentness, Revocation & Supersession;
- P1-RESP-09 Governing Conflict & Escalation;
- P1-RESP-10 Human Accountability & Escalation Envelope;
- P1-RESP-12 Assurance / Governance Feedback Intake.
Why it passes the M02 admission gate:
- authority/accountability is an independently governable control boundary;
- it has distinct currentness, revocation, escalation and evidence obligations;
- it forms a reusable hand-off boundary to P4/P6 and enterprise governance systems;
- it remains meaningful across centralized, federated, embedded and human-mediated realizations;
- failure of this boundary can require materially different fail-safe behavior than failure of ordinary business metadata.
It does not imply a standalone Authorization Service, IAM service, policy engine or entitlement repository.
16.5.3 Responsibilities deliberately not promoted to separate ABBs
The following are retained as responsibilities/interactions rather than separate canonical ABBs in M04:
| Candidate label | M04 disposition | Reason |
|---|---|---|
| Policy Engine | NOT ADMITTED | An implementation pattern; policy semantics and enforcement may be distributed or externally authoritative. |
| AI Governance Registry | NOT ADMITTED | Registry realization is not required by the architecture outcome. |
| Delegated Authority Registry | NOT ADMITTED | Delegated Authority is a semantic concept; persistence/discovery does not independently justify a registry ABB. |
| Risk Tier Service | NOT ADMITTED | AIPS risk semantics remain externally authoritative; service form is implementation-specific. |
| Purpose Store | NOT ADMITTED | Persistence need does not justify a dedicated store ABB. |
| Governance Obligation Service | NOT ADMITTED | Governance obligations may be projected from AIPS/external governance without one generic service boundary. |
| Approval Service | NOT ADMITTED in P1 | Concrete approval/Authorization behavior belongs principally to P6 and may be human/system mediated. |
GEAIRA-M04-R25 - P1 ABB freeze: P1-ABB-01 Governing Intent Boundary and P1-ABB-02 Authority & Accountability Boundary are canonical logical ABBs for GEAIRA 0.1 P1.
GEAIRA-M04-R26 - ABB non-deployment rule: Neither canonical P1 ABB SHALL imply a separately deployable service, process, store, registry or product.
GEAIRA-M04-R27 - Intent/authority separability: P1-ABB-01 and P1-ABB-02 SHALL remain independently traceable even when realized by the same governance platform, Product metadata mechanism or human process.
GEAIRA-M04-R28 - No policy-engine mandate: P1 SHALL NOT require a universal policy engine; executable policy enforcement MAY be selected by profiles/patterns where justified.
GEAIRA-M04-R29 - No registry mandate: P1 SHALL NOT require a universal governance, intent, authority or delegation registry.
GEAIRA-M04-R30 - P6 approval boundary: A generic approval/Authorization mechanism SHALL NOT be promoted as a P1 ABB merely because P1 provides governing authority context; point-of-use authorization is principally a P6 responsibility.
16.6 P1 Architectural Interactions and Contract Obligations
M04 freezes eight P1 interaction responsibilities. M11 will later reconcile these into the canonical cross-plane interaction catalogue.
| ID | Interaction | Direction | Architectural purpose |
|---|---|---|---|
| P1-INT-01 | Intent / Purpose Establishment | authoritative business/governance source -> P1 | Establish intended purpose, scope and permitted/prohibited use. |
| P1-INT-02 | Authority / Accountability Binding | authoritative organizational/governance source -> P1 | Bind authority source, accountable role and delegated mandate where applicable. |
| P1-INT-03 | Policy / Obligation Projection | policy/AIPS/governance authority -> P1 | Supply applicable policy, obligation and control-context references. |
| P1-INT-04 | Product / Risk Governance Association | P5/AIPS governance context -> P1 | Associate qualified Product-Version governance context where valid. |
| P1-INT-05 | Governing Envelope Provision | P1 -> P2/P3/P4/P5/P6/P7 | Make relevant intent/authority/constraint information available to dependent responsibilities. |
| P1-INT-06 | Authority Basis Provision to P6 | P1 -> P6 | Provide authority/delegation/policy basis needed for point-of-use Authorization and oversight decisions. |
| P1-INT-07 | Assurance / Currentness Feedback | P7 -> P1 | Feed findings, currentness concerns, safety/assurance outcomes or semantic change into continued-use governance. |
| P1-INT-08 | Revocation / Supersession Notification | P1 -> affected planes | Propagate material change in purpose, authority or governing constraints to dependent responsibilities. |
A P1 interaction contract MAY include, where applicable:
governed-use identity
intended purpose / scope identifier
permitted / prohibited / constrained use declarations
authority source / accountable role reference
delegated-authority reference and bounded scope
applicable policy / obligation / profile references
Product / AI Product Version reference when valid
risk/governance context reference when valid
effective-from / expiry / revocation / supersession state
currentness / version information
precedence / conflict / escalation metadata
evidence / provenance references
fail-safe or degraded-mode requirement
This is a contract-content model, not a mandatory payload schema.
GEAIRA-M04-R31 - Explicit P1 hand-offs: Material transfer of P1 governing constraints across plane boundaries SHALL use explicit Architectural Interactions/Contracts rather than rely solely on undocumented shared configuration.
GEAIRA-M04-R32 - Contract minimality: P1 interaction contracts SHALL carry only governing information needed by the receiving responsibility and SHALL respect security/privacy/minimization obligations under X1.
GEAIRA-M04-R33 - Currentness carriage: Where P1 information is versioned, revocable or time-bounded, the interaction contract SHALL expose enough identity/currentness information for the receiver to detect materially stale governing state.
GEAIRA-M04-R34 - No synchronous lookup mandate: P1 contract requirements MAY be satisfied by embedded signed policy/context, cached governed state, event propagation, federation, human process or synchronous lookup; GEAIRA does not mandate one runtime interaction style.
GEAIRA-M04-R35 - Downstream non-expansion: A receiving plane SHALL NOT broaden the permitted-use or authority scope supplied by P1 unless an authoritative P1 change establishes the broader scope.
GEAIRA-M04-R36 - Feedback non-mutation: P7 feedback MAY trigger reassessment, suspension, narrowing or supersession of P1 state, but SHALL NOT silently mutate historical governing records that were applicable to prior occurrences.
16.7 P1 Control Points, Failure and Escalation
P1 introduces logical control-point requirements without mandating central services.
16.7.1 Governing Admission Control Point
Before a governed enterprise-AI use is accepted into its applicable operational scope, the architecture must be able to determine that:
- intended purpose is established;
- accountable authority is identified;
- applicable constraints/obligations are known to the required degree;
- required Product/AI Product governance context is associated where applicable;
- unresolved governance conflicts have an escalation path.
This control point may occur through design governance, deployment admission, Product onboarding, release promotion, runtime activation or another enterprise-specialized process.
16.7.2 Authority Change / Revocation Control Point
Where authority or permitted use is revocable, time-bounded or supersedable, the architecture must provide a control point that can prevent continued reliance on materially invalid authority state and can trigger reassessment/escalation.
16.7.3 P1 failure classes
| Failure condition | Generic P1 expectation |
|---|---|
| Intended purpose missing or materially ambiguous | Do not silently treat the system as unrestricted; require clarification or governed exception. |
| Authority source unavailable/indeterminate | Fail closed for authority-dependent activation/action or escalate according to profile; informational/degraded behavior may be separately permitted. |
| Delegated Authority expired/revoked | Prevent further reliance on that authority and propagate invalidation to affected responsibilities. |
| Conflicting authoritative constraints | Escalate/resolve by defined precedence; do not select least restrictive constraint by default. |
| Product/risk context missing where required | Do not fabricate or generalize AIPS governance classification; block or escalate the Product-scoped use as defined by profile. |
| P7 assurance/currentness finding invalidates continued use | Trigger P1 reassessment/suspension/narrowing as appropriate; preserve historical records. |
GEAIRA-M04-R37 - Governing admission control: A conformant architecture SHALL provide a Governing Admission Control Point, though its realization MAY be procedural, embedded, distributed or automated.
GEAIRA-M04-R38 - Authority-dependent fail-safe: Where an operation materially depends on valid authority and that authority is unavailable, expired, revoked or indeterminate, the architecture SHALL define fail-closed or governed-escalation behavior rather than permissive continuation by default.
GEAIRA-M04-R39 - Conflict escalation: Material conflict between authoritative governing inputs SHALL have an explicit resolution/escalation responsibility and SHALL be evidentially traceable when it affects continued operation.
GEAIRA-M04-R40 - Revocation propagation: Material revocation or supersession of governing authority SHALL be propagated to affected dependent responsibilities with bounded delay appropriate to the consequence/risk profile.
GEAIRA-M04-R41 - Historical truth preservation: P1 changes SHALL preserve the governing state that applied historically to prior executions/decisions/actions; currentness change SHALL NOT rewrite historical applicability.
16.8 P1 Cross-Cutting Overlay Application
All X1-X8 overlays can apply to P1, but the strongest recurring P1 pressures are X1, X2, X3, X6, X7 and X8.
| Overlay | P1 application |
|---|---|
| X1 Security, Privacy & Trust Boundaries | Protect sensitive purpose, authority, policy and accountability information; minimize downstream disclosure. |
| X2 Identity, Authority & Policy Enforcement | Establish authoritative identity/authority sources, bounded delegation and policy applicability. |
| X3 Evidence, Provenance & Traceability | Preserve who/what established purpose/authority, under which version/mandate, and what applied when. |
| X4 Observability, Telemetry & Auditability | Observe material activation, authority changes, exceptions and governing constraint propagation where needed. |
| X5 Resilience, Safety & Failure Management | Define safe behavior when governing state is unavailable, contradictory, stale or revoked. |
| X6 Lifecycle, Version, Currentness & Change | Version and reassess purpose, authority and governance applicability without historical mutation. |
| X7 Interoperability, Contracts & Semantic Compatibility | Preserve governing meaning across P1-to-plane and peer-governance boundaries. |
| X8 Governance, Human Accountability & Compliance | Keep accountable participation, exception ownership, escalation and external governance alignment explicit. |
GEAIRA-M04-R42 - Cross-cutting non-locality: P1 SHALL identify its X1-X8 obligations but SHALL NOT absorb the corresponding responsibilities from other planes into one centralized governance boundary.
GEAIRA-M04-R43 - Least-disclosure governing contracts: P1 SHALL expose governing information on a need-to-know basis consistent with X1 while still providing enough information for downstream conformance and safe behavior.
GEAIRA-M04-R44 - Evidence/accountability integrity: P1 evidence/provenance SHALL preserve the distinction between evidence that an authority/purpose record existed and proof that every downstream action complied with it.
16.9 P1 vs P6: Governing Authority and Point-of-Use Authorization
This boundary is canonical for GEAIRA 0.1.
P1
establishes intended purpose
establishes permitted/prohibited use boundary
establishes authority/accountability basis
projects applicable policy/governance constraints
establishes Delegated Authority envelope where applicable
maintains authority currentness/revocation/escalation context
P6
participates in Decision Formation / action paths
determines or consumes point-of-use Authorization where required
applies human approval / intervention / escalation patterns
commits, mediates or rejects governed Actions
The relationship is therefore:
P1 governing envelope
|
v
P6 authorization / oversight control
|
v
specific Decision / Execution / invocation / Action occurrence
but this is not a mandatory universal runtime pipeline. A low-consequence informational execution may require no P6 Authorization occurrence; a physical/high-impact action may require strong point-of-use authorization and human/safety controls.
GEAIRA-M04-R45 - P1/P6 separation: P1 SHALL establish the governing authority basis while P6 SHALL own the principal architecture responsibility for context-specific Authorization, approval, action and human-oversight control.
GEAIRA-M04-R46 - No authorization pre-grant by intent: A P1 permitted-use declaration or Delegated Authority envelope SHALL NOT automatically authorize every downstream Decision, invocation, Execution or Action occurrence.
GEAIRA-M04-R47 - P6 non-expansion of P1: P6 Authorization/control SHALL operate within applicable P1 governing bounds and SHALL NOT silently expand them.
GEAIRA-M04-R48 - Runtime-dependency restraint: P6 MAY use locally available, cached, federated or embedded governing state when valid/current; GEAIRA SHALL NOT require synchronous P1 availability for every authorization event.
16.10 P1 Product / AI Product Governance Boundary
P1 can govern enterprise AI whether or not the governed participant is a Product.
Product / AI Product case
When a valid BPS Product and AIPS AI Product Version exists, P1 may consume:
- Product/Version identity;
- AI Product qualification state;
- AIPS Agency/Autonomy classification;
- Effective Risk Tier;
- Risk Driver Assessment;
- Governance Obligation Profile;
- applicable assurance/currentness information.
P1 uses these as governing inputs. It does not compute Productness or change the AIPS semantic scope.
Non-Product AI case
For an experimental primitive, internal Agent, composition or execution that is not a Product, P1 may still establish:
- intended purpose;
- authority/accountability;
- permitted-use boundaries;
- enterprise policy obligations;
- evidence/escalation expectations.
But P1 SHALL NOT invent an AIPS Product Version, R0-R4 tier or Governance Obligation Profile merely to make the architecture look uniform.
This distinction permits generic enterprise AI governance while preserving Product ontology and AIPS scope.
16.11 Canonical P1 Reference View
The following is the canonical logical P1 view. It shows responsibility flow, not deployment topology.
BUSINESS / GOVERNANCE / LEGAL / PRODUCT AUTHORITIES
|
+----------------+----------------+
| |
v v
+---------------------------+ +------------------------------+
| P1-ABB-01 | | P1-ABB-02 |
| Governing Intent Boundary | | Authority & Accountability |
| | | Boundary |
| intended purpose | | authority source |
| permitted/prohibited use | | accountable role |
| scope / lifecycle identity| | delegated authority |
+-------------+-------------+ | policy/obligation references |
| | currentness/revocation |
| | escalation/feedback |
| +---------------+--------------+
| |
+------------------+----------------+
|
Governing Envelope Provision
|
+----------+------------+------------+----------+
| | | | |
v v v v v
P2 P3 P4 P5 P6 ----> P7
authority |
basis |
|
Assurance / currentness feedback <---+
The two P1 ABBs MAY be co-located, distributed, federated, embedded in Product/governance platforms or partly realized by human process.
16.12 Representative P1 Scenarios
Scenario A — Advisory AI without action authority
An AI capability produces recommendations for a human analyst.
P1 establishes:
- intended purpose: decision support;
- prohibited use: no autonomous external action;
- accountable business owner;
- applicable policy/evidence requirements.
The architecture does not infer Delegated Authority merely because the AI is technically capable of calling a Tool. P6 may require no concrete Authorization for producing the recommendation itself, while any downstream action remains separately governed.
Scenario B — Agent operating under bounded delegated authority
A Software Agent may perform specified low-value operational tasks.
P1 establishes a Delegated Authority envelope bounded by:
- permitted task types;
- resources/systems;
- transaction/action thresholds;
- time/effective period;
- escalation conditions;
- human accountability.
P4 coordinates Agent execution/tool use. P6 applies point-of-use Authorization/approval where the selected action requires it. Agent autonomy does not expand the delegated scope.
Scenario C — Qualified AI Product with AIPS governance context
A BPS Product Version qualifies as an AIPS AI Product Version.
P5/AIPS provides the valid Product-Version identity and governance classification. P1 associates applicable Effective Risk Tier / Governance Obligation Profile with the governed use and projects relevant constraints. P1 does not re-run AI Product qualification or treat the risk tier as an Authorization.
Scenario D — Non-Product experimental AI mechanism
An internal research AI Primitive is evaluated before Productization.
P1 can establish research purpose, permitted datasets/environments, accountable sponsor and enterprise-policy constraints. The architecture does not assign AIPS R0-R4 or a Product Governance Obligation Profile merely because the mechanism uses AI.
Scenario E — Physical / high-consequence AI
A physical-AI pattern has constrained operating conditions and safety obligations.
P1 establishes permitted operating purpose, accountable authority and governing restrictions. P4/P6 implement environment/action/control-loop responsibilities. X5 safety obligations and P7 assurance feedback may cause P1 to suspend or narrow permitted use without rewriting historical executions.
16.13 P1 Anti-Patterns
M04 defines the following P1-specific anti-patterns:
- Policy-engine architecture by default — assuming P1 requires one central policy-decision service.
- Purpose-as-prompt — using runtime prompts/instructions as the authoritative intended-purpose model.
- Risk-tier-as-permission — treating AIPS R0-R4 as direct Authorization.
- Autonomy-as-authority — treating Agent autonomy or Product AR as unrestricted delegated mandate.
- Governance-registry reflex — creating a canonical registry solely because purpose/authority records must be discoverable.
- Owner-equals-authority — assuming technical/Product ownership automatically establishes organizational/legal decision authority.
- P1/P6 collapse — treating standing governing intent and point-of-use authorization as one undifferentiated control.
- Governance inheritance by dependency — copying risk/obligations/authority across dependency or composition without authoritative applicability.
- Silent stale authority — continuing operation under expired/revoked/superseded authority without bounded currentness behavior.
- Assurance-is-approval — converting evaluation/assurance PASS into purpose approval or action permission without explicit governance authority.
- Centralization by diagram — assuming all P1 responsibilities must be realized by one enterprise governance platform.
- Uniform Product governance — forcing non-Product AI into AIPS Product-Version risk/governance semantics for diagram consistency.
16.14 M04 Frozen Decisions
M04 freezes the following P1 architecture decisions for GEAIRA 0.1:
P1 establishes the governing envelope for enterprise AI.
intended purpose is architecture-governed but is not a GEAIRM AI Context, Prompt or Instruction
permitted use != technical capability
authority/accountability must be explicit at the applicable governance scope
Delegated Authority != Authorization
autonomy != authority
risk tier != permission
Governance Obligation != concrete Authorization/control occurrence
P1 governing envelope != P6 point-of-use Authorization
P1 does not require synchronous runtime consultation
P1 can govern Product and non-Product AI without flattening their semantics
P1 currentness/revocation/change must preserve historical truth
The following P1 ABBs are canonical:
P1-ABB-01 Governing Intent Boundary
P1-ABB-02 Authority & Accountability Boundary
The following remain responsibilities/realization choices rather than canonical generic ABBs:
Policy Engine
AI Governance Registry
Delegated Authority Registry
Risk Tier Service
Purpose Store
Governance Obligation Service
Approval Service (P1)
M04 does not freeze:
- an enterprise policy language;
- a universal authority or entitlement model;
- a runtime authorization protocol;
- a centralized governance platform;
- organization-specific accountability hierarchies;
- regulatory control libraries;
- implementation-specific registries/stores/workflows;
- P6 authorization/action patterns, which remain M09 work;
- the final M11 inter-plane contract catalogue.
17. M04 Completion State
GEAIRA-M04 — COMPLETE.
M04 establishes:
- one canonical P1 purpose and scope;
- twelve canonical P1 Architectural Responsibilities;
- two canonical logical P1 ABBs;
- seven deliberately non-admitted implementation-style ABB candidates;
- eight P1 interaction responsibilities;
- two principal P1 logical control-point classes;
- six generic P1 failure/exception classes;
- explicit P1 application of X1-X8 cross-cutting overlays;
- a canonical P1/P6 governing-authority vs point-of-use Authorization boundary;
- five representative P1 scenarios;
- twelve P1 anti-patterns;
- 48 normative M04 rules.
M04 introduces no change to GEAIRM semantics, does not prescribe a deployment topology, and does not require a central policy/governance runtime.
Next slice
GEAIRA-M05 — P2: Knowledge, Context & Grounding
M05 will apply the M02 responsibility/ABB discipline and M03 overlay model to context construction, knowledge resources, retrieval, grounding, memory, source identity/currentness and the P2-to-P4 context-provision boundary.
18. GEAIRA-M05 - P2: Knowledge, Context & Grounding
18.1 Purpose and semantic baseline
GEAIRA-M05 establishes the canonical architecture baseline for P2 — Knowledge, Context & Grounding.
P2 answers the architectural question:
How should enterprise AI identify, retrieve, qualify, assemble, ground, maintain and provide the context and knowledge resources required by an AI Execution without confusing context with environment, retrieval with grounding, grounding with correctness, or memory with learning?
M05 consumes the frozen GEAIRM D3 semantic baseline, especially:
Knowledge Resource;Rule;AI Context;Instruction;Prompt;Retrieval;Memory;AI Context ResourceRole;Grounding SourceRole;hasAIContext;includesContextResource;containsInstruction;groundsAIContext;retrievesResource;suppliesToAIContext;memoryContributesToAIContext.
M05 does not redefine those semantics. It derives technology-neutral architecture responsibilities from them.
GEAIRA-M05-R01 - Semantic preservation: P2 SHALL consume GEAIRM context, knowledge, grounding, retrieval and memory semantics without redefining their intrinsic meaning.
GEAIRA-M05-R02 - Responsibility-before-component: P2 SHALL derive architectural responsibilities before admitting any canonical P2 ABB.
GEAIRA-M05-R03 - Technology neutrality: P2 SHALL NOT require a vector database, knowledge graph, search engine, document repository, prompt store, memory database, embedding model, retrieval framework or other implementation technology merely because the corresponding semantic responsibility exists.
18.2 P2 architecture boundary
P2 is the principal GEAIRA plane for the architecture responsibilities concerned with:
- determining what context is required for a governed execution/use;
- locating and obtaining eligible knowledge/context resources;
- qualifying source identity, applicability, provenance and currentness;
- constructing a bounded AI Context;
- representing and carrying Prompt / Instruction material where applicable;
- associating Grounding Source roles with applicable resources;
- incorporating Memory contributions without treating Memory as learning;
- providing context to P4 execution;
- preserving traceability of what was made available;
- invalidating or refreshing context when governing source/currentness conditions materially change.
P2 does not own:
- intended purpose or standing governing authority — P1;
- AI Primitive / capability realization — P3;
- execution semantics or Agent runtime coordination — P4;
- Product serving/integration concerns — P5;
- point-of-use Authorization / action control — P6;
- independent evaluation, assurance or architecture-wide observability — P7;
- Data / Information intrinsic semantics — GEDRM / GEARM and their architecture counterparts;
- generic provenance semantics — W3C PROV.
GEAIRA-M05-R04 - Context/environment separation: P2 AI Context SHALL remain distinct from P4 Environment; environmental state MAY become context material, but the architecture SHALL NOT collapse the two concepts.
GEAIRA-M05-R05 - Retrieval/grounding separation: P2 SHALL treat Retrieval and Grounding as distinct responsibilities; successful retrieval SHALL NOT automatically establish Grounding Source status.
GEAIRA-M05-R06 - Grounding/correctness separation: P2 SHALL NOT interpret grounding as proof that a source or resulting output is correct, complete, authoritative or fit for purpose.
GEAIRA-M05-R07 - Memory/learning separation: P2 Memory responsibilities SHALL NOT imply model training, retraining, parameter update or learned adaptation.
GEAIRA-M05-R08 - Prompt/authority separation: Prompt and Instruction handling within P2 SHALL NOT create Authorization, Delegated Authority or governing purpose.
18.3 Canonical P2 Architectural Responsibilities
M05 establishes the following canonical P2 responsibilities.
| ID | Architectural Responsibility | Principal obligation class | Architectural intent |
|---|---|---|---|
P2-RESP-01 | Context Requirement Determination | Baseline-mandatory | Determine the categories, constraints and scope of context required for the governed execution/use. |
P2-RESP-02 | Context Resource Eligibility | Baseline-mandatory | Determine whether a candidate resource is eligible to participate in a context under applicable policy, identity, scope and quality/currentness rules. |
P2-RESP-03 | Knowledge Resource Access & Mediation | Conditional-mandatory | Obtain governed access to knowledge/resources without assuming a particular repository or access protocol. |
P2-RESP-04 | Retrieval Coordination | Conditional-mandatory | Locate/select/obtain resources where retrieval is part of context construction. |
P2-RESP-05 | Grounding Source Qualification | Conditional-mandatory | Establish when an eligible resource is intentionally used in the Grounding Source Role. |
P2-RESP-06 | Context Assembly & Bounding | Baseline-mandatory | Construct the bounded set of resources, state, conditions, prompts/instructions and other information to be made available to an execution. |
P2-RESP-07 | Prompt & Instruction Contextualization | Conditional-mandatory | Preserve applicable prompt/instruction identity, ordering, scope and context contribution without treating them as authority. |
P2-RESP-08 | Memory Contribution Governance | Conditional-mandatory | Govern retained/retrievable state contributed to context, including scope, isolation, retention and currentness. |
P2-RESP-09 | Source Identity, Version & Currentness | Baseline-mandatory where source identity matters | Preserve sufficient identity/version/currentness information for materially relevant resources. |
P2-RESP-10 | Context Provenance & Traceability | Baseline-mandatory | Record or expose enough lineage/evidence to reconstruct what resources and transformations contributed to the provided context. |
P2-RESP-11 | Context Provision to Execution | Baseline-mandatory | Provide the assembled context to P4 through a governed architecture interaction. |
P2-RESP-12 | Context Change, Invalidation & Refresh | Conditional-mandatory | React to source revocation, supersession, staleness, policy change or changed context requirements. |
P2-RESP-13 | Context Conflict & Degradation Handling | Conditional-mandatory | Handle contradictory, unavailable, insufficient, over-large or otherwise degraded context without silently converting uncertainty into certainty. |
GEAIRA-M05-R09 - Explicit context requirement: A governed P2 realization SHALL make context requirements explicit enough to distinguish required, optional and prohibited context categories for the applicable use.
GEAIRA-M05-R10 - Resource eligibility before inclusion: A resource SHALL NOT be treated as an eligible context contribution solely because it is technically retrievable.
GEAIRA-M05-R11 - Bounded context: P2 SHALL provide a bounded context appropriate to the applicable execution/use rather than treating all accessible enterprise information as implicitly available AI Context.
GEAIRA-M05-R12 - Context-role preservation: A resource included in an AI Context SHALL retain its externally governed intrinsic identity while playing the contextual AI Context Resource and/or Grounding Source Role where applicable.
GEAIRA-M05-R13 - Traceable construction: P2 SHALL preserve traceability from the provided context to materially relevant source resources, retrieval/selection activity and context transformation/assembly steps where required by risk, governance or assurance needs.
GEAIRA-M05-R14 - Currentness-aware contribution: P2 SHALL consider applicable identity/version/currentness constraints before presenting a resource as current or eligible context.
GEAIRA-M05-R15 - Context provision independence: P2 context provision SHALL be defined as an architectural responsibility independent of whether context is pushed, pulled, embedded, streamed, cached, federated or assembled in-process.
18.4 Semantic-to-architecture derivation
M05 applies the M02 derivation discipline as follows.
| GEAIRM semantic fact | P2 architectural consequence | Not automatically implied |
|---|---|---|
AIExecution hasAIContext AIContext | Context assembly/provision responsibility | Context Service |
AIContext includesContextResource Resource | Resource eligibility/inclusion responsibility | Central context repository |
Prompt containsInstruction Instruction | Prompt/instruction integrity and contextualization | Prompt-management platform; Authorization |
GroundingSource groundsAIContext AIContext | Grounding qualification and source traceability | Vector database; correctness |
Retrieval retrievesResource Resource | Retrieval coordination/access responsibility | Intelligent Search capability; search engine ABB |
Retrieval suppliesToAIContext AIContext | Governed hand-off from retrieval to context assembly | Grounding status |
Memory contributesToAIContext AIContext | Governed retained-state contribution | Learning, retraining or Agent identity |
KnowledgeResource exists | Knowledge/resource mediation may be needed | Knowledge graph, knowledge base or Product status |
GEAIRA-M05-R16 - No ontology-as-service projection: P2 SHALL NOT create a canonical ABB merely by suffixing a GEAIRM concept or relationship with Service, Store, Registry, Manager or equivalent implementation terminology.
GEAIRA-M05-R17 - Retrieval non-intelligence: Retrieval architecture SHALL NOT be classified as an intelligence capability merely because it selects resources for AI use.
GEAIRA-M05-R18 - Source-role contextuality: Grounding Source and AI Context Resource SHALL remain contextual roles; P2 SHALL NOT mint intrinsic architecture types that erase the underlying resource authority.
18.5 Canonical P2 ABBs
After applying the M02 ABB-admission criteria, M05 admits three canonical logical P2 ABBs.
P2-ABB-01 — Context Assembly & Provision Boundary
Purpose: Establish the architecture boundary responsible for constructing a bounded AI Context and making it available to P4 execution.
Principal responsibility coverage:
P2-RESP-01Context Requirement Determination;P2-RESP-06Context Assembly & Bounding;P2-RESP-07Prompt & Instruction Contextualization;P2-RESP-08Memory Contribution Governance;P2-RESP-11Context Provision to Execution;- applicable portions of
P2-RESP-13.
This ABB defines a logical responsibility boundary. It may be embedded in an application/runtime, distributed across multiple components, implemented through managed capabilities, or partly realized through deterministic configuration.
It does not imply a standalone Context Service.
P2-ABB-02 — Knowledge & Retrieval Mediation Boundary
Purpose: Establish the architecture boundary responsible for governed discovery/access/retrieval of candidate knowledge and context resources.
Principal responsibility coverage:
P2-RESP-02Context Resource Eligibility;P2-RESP-03Knowledge Resource Access & Mediation;P2-RESP-04Retrieval Coordination;- relevant access/currentness aspects of
P2-RESP-09.
This ABB may federate across Data Products, Information Products, document stores, APIs, event streams, search/index capabilities, knowledge graphs or other resource authorities.
It does not imply one central enterprise knowledge repository or a vector database.
P2-ABB-03 — Grounding & Source Qualification Boundary
Purpose: Establish the architecture boundary responsible for qualifying source use as grounding, preserving source identity/currentness and exposing traceability of the grounding basis.
Principal responsibility coverage:
P2-RESP-05Grounding Source Qualification;P2-RESP-09Source Identity, Version & Currentness;P2-RESP-10Context Provenance & Traceability;P2-RESP-12Context Change, Invalidation & Refresh;- applicable portions of
P2-RESP-13.
This ABB does not certify source truth or output correctness. P7 may evaluate quality, evidence sufficiency, fitness or assurance independently.
GEAIRA-M05-R19 - Context ABB admission: P2-ABB-01 is canonical because bounded context construction/provision forms a cohesive, reusable, independently traceable architecture responsibility boundary.
GEAIRA-M05-R20 - Retrieval ABB admission: P2-ABB-02 is canonical because governed access/retrieval across heterogeneous authorities forms a reusable interoperability and mediation boundary independent of repository technology.
GEAIRA-M05-R21 - Grounding ABB admission: P2-ABB-03 is canonical because Grounding Source qualification, source identity/currentness and grounding traceability form an independently governable semantic/control boundary distinct from raw retrieval.
GEAIRA-M05-R22 - ABB deployment neutrality: None of the three P2 ABBs SHALL imply a separately deployed service, process, database or platform component.
GEAIRA-M05-R23 - ABB co-location: Two or more P2 ABBs MAY be co-located in one realization without losing their independent architectural traceability.
18.6 Deliberately non-admitted P2 ABB candidates
The following implementation-style candidates are not admitted as generic canonical GEAIRA ABBs in M05:
Vector Database
Knowledge Graph
Knowledge Base
Enterprise Search Service
Document Store
Embedding Service
Prompt Store / Prompt Registry
Memory Store / Memory Service
Context Cache
Context Registry
RAG Service
Grounding Service
Any of these MAY be valid realization components or profile-level specializations.
GEAIRA-M05-R24 - No vector-database mandate: Retrieval/grounding responsibility SHALL NOT imply vector similarity search, embeddings or a vector database.
GEAIRA-M05-R25 - No knowledge-store mandate: Knowledge Resource architecture SHALL NOT imply one canonical enterprise knowledge store or graph.
GEAIRA-M05-R26 - No prompt-store mandate: Prompt/Instruction governance SHALL NOT require a standalone prompt registry or store at generic reference-architecture level.
GEAIRA-M05-R27 - No memory-service mandate: Memory contribution SHALL NOT require a standalone memory service or persistent memory store.
GEAIRA-M05-R28 - RAG pattern restraint: Retrieval-augmented generation is a valid realization pattern but SHALL NOT define the generic scope of P2.
18.7 Canonical P2 architectural interactions
M05 establishes the following technology-neutral interaction responsibilities.
| ID | Interaction | From | To | Architectural meaning |
|---|---|---|---|---|
P2-INT-01 | Governing Context Constraints | P1 | P2 | Supplies purpose/policy/authority constraints that affect eligible context and sources. |
P2-INT-02 | Context Requirement / Request | P4 or governed caller | P2 | Expresses what context is required for an execution/use. |
P2-INT-03 | Knowledge / Resource Access | P2 | external resource authorities / GEDRA / GEARA / P5 integrations | Obtains candidate resources under applicable contracts and controls. |
P2-INT-04 | Retrieval Result Contribution | P2-ABB-02 | P2-ABB-01 | Makes retrieved resources available for context assembly without asserting grounding. |
P2-INT-05 | Grounding Qualification | P2-ABB-03 | P2-ABB-01 | Supplies grounding-role/source-identity/currentness metadata for qualifying resources. |
P2-INT-06 | Context Provision | P2 | P4 | Provides the bounded AI Context for an execution/interaction. |
P2-INT-07 | Context Trace / Evidence | P2 | P7 | Exposes source/context-construction evidence required for evaluation, assurance or observability. |
P2-INT-08 | Source / Context Currentness Feedback | P7 / source authority | P2 | Signals findings, supersession, invalidation or currentness concerns affecting future context use. |
P2-INT-09 | Context Invalidation / Refresh Notification | P2 | P4/P5/affected consumers | Signals that previously prepared/cached context is no longer valid under declared conditions. |
GEAIRA-M05-R29 - Contract neutrality: P2 interactions SHALL be defined independently of REST, event, query, RPC, file, database or model-provider protocol choices.
GEAIRA-M05-R30 - P1 constraint preservation: P2 SHALL preserve applicable P1 purpose/policy/authority constraints when determining context eligibility and shall not broaden governed use through context construction.
GEAIRA-M05-R31 - P4 request non-authority: A P4 context request SHALL NOT itself authorize access to or inclusion of resources prohibited by P1, external authority or cross-cutting controls.
GEAIRA-M05-R32 - Retrieved/result separation: P2-INT-04 SHALL NOT infer that every retrieved result is included in the final context or used as grounding.
GEAIRA-M05-R33 - Context evidence availability: P2 SHALL make sufficient context/source trace information available to P7 when evaluation, assurance, audit or regulated evidence requires it.
18.8 P2 logical control points
M05 establishes three logical control-point classes.
Context Resource Admission Control Point
Determines whether a candidate resource may enter the context-construction path under applicable:
- purpose/use scope;
- access/privacy constraints;
- source identity;
- classification;
- currentness;
- policy/governance conditions.
Grounding Qualification Control Point
Determines whether a resource may play the Grounding Source Role for the applicable context/use and records the applicable basis/traceability.
Context Release / Provision Control Point
Determines whether an assembled context is sufficiently bounded, policy-compatible and internally consistent to be provided to P4 under the declared use conditions.
These are logical controls. They may be embedded, federated, automated, human-reviewed or combined.
GEAIRA-M05-R34 - Control-point non-inflation: P2 logical control points SHALL NOT automatically become standalone policy engines, gateways or services.
GEAIRA-M05-R35 - Admission fail-safe: Where resource eligibility is materially indeterminate under applicable controls, P2 SHALL NOT silently treat the resource as eligible.
GEAIRA-M05-R36 - Grounding fail-safe: Where Grounding Source qualification is required but cannot be established, P2 SHALL preserve the distinction rather than silently labelling the material as grounded.
GEAIRA-M05-R37 - Release fail-safe: Where context conflict, policy incompatibility or insufficiency is material to the governed use, P2 SHALL use an explicit failure, degradation or escalation behavior rather than silently presenting the context as fully valid.
18.9 Cross-cutting overlay application to P2
All M03 overlays remain applicable. M05 specializes them for P2 as follows.
X1 — Security, Privacy & Trust Boundaries
Applies to resource access, sensitive context, source trust boundaries, data minimization and context leakage prevention.
X2 — Identity, Authority & Policy Enforcement
Applies to source/resource identity, calling identity, access decisions and enforcement of applicable P1 governing constraints.
X3 — Evidence, Provenance & Traceability
Applies strongly to source identity, retrieval trace, context composition and grounding basis.
X4 — Observability, Telemetry & Auditability
Applies to context/retrieval operational events where material, while avoiding capture of sensitive content merely for observability convenience.
X5 — Resilience, Safety & Failure Management
Applies to unavailable sources, degraded retrieval, stale context, conflicting resources and fallback behavior.
X6 — Lifecycle, Version, Currentness & Change
Applies to source versions, supersession, memory retention, cached contexts and invalidation/refresh semantics.
X7 — Interoperability, Contracts & Semantic Compatibility
Applies to heterogeneous resource authorities, schemas, content types, context contracts and cross-model alignment.
X8 — Governance, Human Accountability & Compliance
Applies to sensitive/regulated context, source approvals, exception handling and human review where required.
GEAIRA-M05-R38 - Overlay orthogonality: P2 implementations SHALL NOT collapse the X1-X8 overlays into P2-specific platform services merely because P2 consumes those obligations.
GEAIRA-M05-R39 - Privacy-aware context minimization: P2 SHALL support minimization of context to what is materially required/allowed for the governed use rather than maximizing accessible data by default.
GEAIRA-M05-R40 - Trace/privacy balance: Traceability obligations SHALL preserve sufficient evidence without requiring unrestricted duplication of sensitive source content into observability/evidence stores.
18.10 P2 / P4 / P7 boundary
The primary plane boundary is:
P2
identifies / retrieves / qualifies / assembles context
preserves source and grounding traceability
provisions bounded AI Context
P4
consumes applicable context in AI Execution / Agent runtime
determines runtime participation and execution behavior
P7
evaluates / observes / assures
assesses evidence, quality, currentness or fitness
feeds findings/currentness concerns back to P2
A P4 execution may use only a subset of the material supplied by P2. The semantic relation includesContextResource does not establish causal influence.
P7 may evaluate retrieval quality, grounding evidence or context fitness, but evaluation does not rewrite the historical context actually supplied to a past execution.
GEAIRA-M05-R41 - P2/P4 separation: P2 SHALL own context construction/provision responsibilities while P4 SHALL own execution-time consumption/use; neither plane SHALL silently assume the other's responsibilities.
GEAIRA-M05-R42 - No causal-use inference: P2 provisioning a resource in AI Context SHALL NOT imply that P4 materially used that resource or that it caused the resulting effect.
GEAIRA-M05-R43 - P7 independence: P7 evaluation/assurance findings SHALL be able to assess P2 context/grounding behavior independently of the P2 mechanism that produced it.
GEAIRA-M05-R44 - Historical non-mutation: Later source/currentness/evaluation changes SHALL NOT mutate the historical fact of which context was supplied to a completed execution.
18.11 Source identity, version, currentness and caching
P2 may use cached, replicated, precomputed or federated context/resource representations. GEAIRA does not prohibit such realization choices.
However, where currentness is material, the architecture must retain enough information to distinguish:
source identity
source/version identifier where applicable
retrieval/selection time or applicable observation
context assembly version/identity where material
currentness assessment / concern
supersession or invalidation event
A cached representation does not become the authoritative source merely by existing closer to execution.
GEAIRA-M05-R45 - Cache non-authority: A cache, index, embedding, replica or context snapshot SHALL NOT inherit source authority merely because it is used by P2.
GEAIRA-M05-R46 - Currentness-aware caching: Where currentness materially affects safe or governed use, cached/preassembled context SHALL have an explicit validity/currentness strategy.
GEAIRA-M05-R47 - Supersession non-deletion: Source or context supersession SHALL NOT require deletion of historical evidence needed to reconstruct prior executions, subject to lawful retention/privacy obligations.
18.12 Generic P2 failure and degradation classes
M05 establishes the following generic failure/exception classes:
- Source unavailable — required source cannot be accessed.
- Source ineligible — accessible source fails policy/scope/identity eligibility.
- Source stale / superseded / currentness-indeterminate — currentness requirement is not met or cannot be established.
- Retrieval empty / insufficient — retrieval returns no or insufficient candidate material.
- Resource conflict — eligible sources materially contradict one another.
- Grounding qualification missing — resources are available but required grounding basis is not established.
- Context capacity / bounding failure — candidate context cannot be safely/meaningfully bounded for the target execution.
- Memory isolation / leakage risk — retained state crosses an impermissible user, purpose, tenant, Product or policy boundary.
- Instruction conflict — context carries incompatible instructions requiring precedence, rejection or escalation.
- Context provision failure — a valid context cannot be provided to P4 under required timing/availability constraints.
Allowed architecture responses include:
- fail closed;
- omit non-essential material with explicit degraded-state signaling;
- use an approved fallback source/snapshot;
- require human review/escalation;
- retry or defer;
- continue only where a profile explicitly permits bounded degradation.
GEAIRA-M05-R48 - No silent least-restrictive conflict resolution: Conflicting sources/instructions/constraints SHALL NOT silently resolve to the least restrictive or most convenient interpretation where the conflict is material.
GEAIRA-M05-R49 - Memory isolation: P2 memory contribution SHALL preserve applicable isolation boundaries and SHALL NOT expose retained state across governed scopes without explicit authority.
GEAIRA-M05-R50 - Degradation transparency: Material fallback, omission, stale-source use or degraded context SHALL be visible to the consuming/control/evidence path where it can affect interpretation, decision, action or assurance.
18.13 Canonical P2 reference view
The following is the canonical logical P2 view. It is a responsibility view, not a deployment topology.
P1 GOVERNING ENVELOPE
|
v
purpose / policy / authority constraints
|
+----------------+----------------+
| |
v v
+-------------------------+ +----------------------------+
| P2-ABB-02 | | P2-ABB-03 |
| Knowledge & Retrieval |------>| Grounding & Source |
| Mediation Boundary | | Qualification Boundary |
| | | |
| access / discovery | | grounding role |
| retrieval | | source identity/version |
| resource eligibility | | currentness / traceability |
+------------+------------+ +-------------+--------------+
| |
| retrieved candidates | qualification
+----------------+-----------------+
|
v
+-----------------------------+
| P2-ABB-01 |
| Context Assembly & |
| Provision Boundary |
| |
| bounded AI Context |
| prompt/instruction content |
| memory contribution |
+--------------+--------------+
|
Context Provision
|
v
P4
AI Execution / Agentic
trace / evidence |
v
P7
evaluation / assurance
|
currentness/findings
+-----> P2
External resource authorities / GEDRA / GEARA / P5 integrations feed P2-ABB-02.
The three ABBs may be realized together, separately, embedded in a consuming runtime or federated across enterprise/platform boundaries.
18.14 Representative P2 scenarios
Scenario A — Retrieval-augmented generation with governed grounding
P4 requests context for a generation task.
P2-ABB-02 retrieves candidate resources from approved authorities. P2-ABB-03 qualifies selected resources as Grounding Sources based on source identity/currentness and applicable governance. P2-ABB-01 constructs the bounded context and supplies it to P4.
The architecture does not claim that retrieved sources are correct merely because they were grounded.
Scenario B — Retrieved resource is not grounding
A search/retrieval step obtains several documents to help an Agent navigate a task. Some are supplied as optional context but are not treated as authoritative/evidential grounding.
P2 records them as context resources without assigning Grounding Source Role. This is valid and preserves Retrieval != Grounding.
Scenario C — Memory contribution to context
An assistant uses retained interaction state to maintain continuity.
P2 governs memory scope, isolation, retention and contribution to the next context. No model retraining or parameter update is implied. Cross-user or cross-purpose leakage is prohibited unless independently authorized.
Scenario D — Prompt carries instructions but not authority
A Prompt includes formatting and task instructions. P2 preserves prompt/instruction integrity and supplies them as context.
If an instruction asks the Agent to perform an action outside the P1/P6 authority envelope, P2 does not transform that instruction into permission. P6/P4 control remains authoritative for action.
Scenario E — Source superseded after historical execution
An execution was grounded using version 3 of a policy knowledge resource. Version 4 later supersedes it.
P2 marks future context selection/currentness appropriately while preserving historical evidence that version 3 was supplied to the earlier execution. Historical truth is not rewritten.
Scenario F — Conflicting authoritative resources
Two eligible sources provide materially conflicting instructions or facts.
P2 does not silently select the least restrictive source. It applies declared precedence, confidence/evidence rules, bounded degradation or escalation; P7 may evaluate the conflict-resolution evidence.
18.15 P2 anti-patterns
M05 defines the following P2-specific anti-patterns:
- Vector-database architecture — treating P2 as synonymous with embeddings + vector search.
- Retrieval-equals-grounding — assigning Grounding Source status to everything retrieved.
- Grounding-equals-truth — treating grounded material as guaranteed correct or authoritative.
- Context-equals-environment — collapsing AI Context and operational/physical/digital Environment.
- Prompt-equals-purpose — using Prompt content as the governing intended-purpose declaration.
- Instruction-equals-authority — treating instructions as permission to decide or act.
- Memory-equals-learning — treating retained state as model learning/retraining.
- Knowledge-resource-equals-AI — classifying a knowledge base/graph/document corpus as an AI mechanism merely because AI consumes it.
- Everything-in-context — maximizing context size/access rather than establishing bounded relevance and minimization.
- Central knowledge monolith — assuming P2 requires one enterprise knowledge repository.
- Context-service reflex — turning
hasAIContextdirectly into a mandatory Context Service. - Grounding-service reflex — turning
groundsAIContextdirectly into a mandatory Grounding Service. - Prompt-registry reflex — making prompt discoverability/versioning imply a canonical enterprise Prompt Registry ABB.
- Stale-cache invisibility — using cached/preassembled context without an explicit currentness/validity strategy where currentness matters.
- Provenance-by-copy — duplicating sensitive source content solely to prove traceability instead of preserving appropriate references/evidence.
- Latest-wins rewriting — replacing historical source/context identity with the newest version and losing execution reconstruction.
18.16 M05 frozen decisions
M05 freezes the following P2 architecture decisions for GEAIRA 0.1:
P2 owns knowledge/context/grounding architecture responsibilities.
P2 context != P4 environment.
Retrieval != Grounding.
Grounding != correctness / truth / authority.
Memory != learning / retraining.
Prompt / Instruction != purpose / authority / Authorization.
Context resources retain their external intrinsic identity.
Context must be bounded and eligibility-governed.
Source identity/version/currentness must be preserved where material.
Historical context truth is preserved when sources later change.
P2 -> P4 context provision does not imply causal use of every resource.
P7 independently evaluates/assures context/grounding behavior.
The following P2 ABBs are canonical:
P2-ABB-01 Context Assembly & Provision Boundary
P2-ABB-02 Knowledge & Retrieval Mediation Boundary
P2-ABB-03 Grounding & Source Qualification Boundary
The following remain realization choices rather than generic canonical ABBs:
Vector Database
Knowledge Graph
Knowledge Base
Enterprise Search Service
Document Store
Embedding Service
Prompt Store / Prompt Registry
Memory Store / Memory Service
Context Cache / Context Registry
RAG Service
Grounding Service
M05 does not freeze:
- a universal context document/schema;
- a universal retrieval ranking algorithm;
- a source-trust scoring model;
- a vector/embedding model;
- a prompt language/template framework;
- a memory persistence technology;
- a universal knowledge graph;
- repository/search/query protocols;
- Product-specific P5 serving contracts;
- P4 execution mechanics;
- P7 evaluation/assurance criteria;
- final M11 inter-plane interaction contracts.
GEAIRA-M05-R51 - Frozen ABB restraint: M05 SHALL freeze only the three P2 logical ABBs above and SHALL NOT promote implementation technologies to canonical ABB status.
GEAIRA-M05-R52 - Downstream preservation: Later GEAIRA slices SHALL preserve the M05 semantic/architecture distinctions unless the 0.1 programme explicitly records and reconciles a justified architecture change without redefining GEAIRM semantics.
19. M05 Completion State
GEAIRA-M05 — COMPLETE.
M05 establishes:
- one canonical P2 purpose and architecture boundary;
- thirteen canonical P2 Architectural Responsibilities;
- three canonical logical P2 ABBs;
- twelve deliberately non-admitted implementation-style ABB candidates;
- nine technology-neutral P2 architectural interactions;
- three logical P2 control-point classes;
- ten generic P2 failure/degradation classes;
- explicit P2 application of X1-X8 cross-cutting overlays;
- explicit P2/P4/P7 responsibility boundaries;
- six representative P2 scenarios;
- sixteen P2 anti-patterns;
- 52 normative M05 rules.
M05 introduces no change to GEAIRM semantics, does not prescribe a deployment topology and does not require a central context, knowledge, retrieval, memory, prompt or grounding platform.
Next slice
GEAIRA-M06 — P3: Intelligence Capability & Primitive
M06 will apply the responsibility/ABB discipline to Intelligence Capability, AI Primitive, AI Primitive Type, mechanism availability, capability realization, model/reasoner/tool distinction, lifecycle compatibility and the P3-to-P4 execution boundary.
20. GEAIRA-M06 - P3: Intelligence Capability & Primitive
20.1 Purpose of this slice
M06 establishes the canonical architecture baseline for P3 - Intelligence Capability & Primitive.
GEAIRM and AIPS already separate four semantic questions:
Intelligence Capability = what kind of intelligence is provided?
AI Primitive = what qualifying computational mechanism contributes AI?
Execution Mode = how is that intelligence invoked or operated?
Intelligence Effect = what does the execution produce or cause?
P3 is concerned principally with the first two questions. It establishes the architectural responsibilities required to connect required Intelligence Capabilities to one or more eligible AI Primitives without collapsing capability into mechanism, mechanism into Product, or mechanism availability into runtime execution.
M06 therefore answers:
- How are required Intelligence Capabilities represented architecturally without redefining the AIPS vocabulary?
- How are candidate AI Primitives identified, evaluated for architectural eligibility and related to required capabilities?
- How are single-primitive and multi-primitive realization patterns handled?
- How are primitive identity, version, currentness, compatibility and evidence carried toward execution?
- What belongs in P3 versus P4 execution, P5 Product/serving integration and P7 evaluation/assurance?
- Which responsibilities justify canonical ABBs and which implementation technologies remain realization choices?
GEAIRA-M06-R01 - AIPS authority preservation: P3 SHALL reuse AIPS authority for IntelligenceCapability, AIPrimitive, AIPrimitiveType and their controlled values and SHALL NOT mint GEAIRA substitutes for those semantic concepts.
GEAIRA-M06-R02 - Capability/mechanism separation: P3 SHALL preserve Intelligence Capability and AI Primitive as orthogonal dimensions; lexical similarity such as Planning and Planner SHALL NOT imply semantic or architectural identity.
GEAIRA-M06-R03 - Primitive breadth: P3 SHALL preserve the mechanism-neutral AIPS AI Primitive concept and SHALL NOT narrow enterprise AI architecture to learned models or machine-learning mechanisms.
20.2 P3 architecture boundary
P3 owns the architecture responsibility for capability-to-mechanism realization.
Its generic boundary is:
required Intelligence Capability
|
v
candidate / available AI Primitives
|
eligibility + evidence
compatibility + currentness
|
v
qualified architectural realization set
|
v
P4 AI Execution
The phrase qualified architectural realization set does not mean AIPS AI Product qualification. It means that, for a declared architecture context, the primitive or primitive set satisfies the P3 eligibility and compatibility conditions required before P4 may use it.
P3 does not own:
- the definition of Intelligence Capability or AI Primitive;
- AIPS AI Product qualification;
- Product Family / Product Version semantics;
- runtime AI Execution;
- execution-mode semantics;
- context construction/grounding;
- Agent identity;
- point-of-use Authorization;
- Intelligence Effect interpretation;
- independent evaluation/assurance;
- vendor/model/runtime topology.
GEAIRA-M06-R04 - P3/P4 boundary: P3 SHALL establish primitive eligibility, capability contribution, composition/compatibility and provision responsibilities; P4 SHALL own actual AI Execution, invocation and runtime participation.
GEAIRA-M06-R05 - No availability/execution equivalence: A primitive being available or eligible in P3 SHALL NOT imply that it was invoked, exercised a capability or produced an Intelligence Effect in P4.
GEAIRA-M06-R06 - No primitive/Product equivalence: AI Primitive status, primitive availability or primitive productization SHALL NOT by itself establish AI Product qualification.
20.3 Source semantic baseline
P3 consumes the following frozen semantic baseline.
AIPS-owned concepts
aips:IntelligenceCapability;- canonical capability values including Perception, Classification, Detection, Prediction, Generation, Recommendation/Ranking, Reasoning, Planning, Optimization and Control;
aips:AIPrimitive;aips:AIPrimitiveType;- primitive-type values including Learned Model, Symbolic Reasoner, Knowledge Reasoner, Planner, Intelligent Search, Optimizer and Hybrid Intelligence Engine;
aips:AIModelas a specialization of AI Primitive where applicable;aips:AIPrimitiveEvidencewhere available;- AIPS Product-version capability/effect/execution declarations where P5 or Product-specific profiles consume them.
GEAIRM-owned relationships
contributesToIntelligenceCapability;classifiedAsPrimitiveType;involvesAIPrimitiveat the P4 occurrence boundary;exercisesIntelligenceCapabilityat the P4 occurrence boundary.
The architecture must preserve the many-to-many nature of capability contribution.
one primitive -> may contribute to many capabilities
one capability -> may depend on many primitives
GEAIRA-M06-R07 - Many-to-many preservation: P3 SHALL NOT impose a one-primitive-to-one-capability cardinality unless a narrower profile or specific realization independently requires it.
GEAIRA-M06-R08 - Contribution is not sufficiency: contributesToIntelligenceCapability SHALL NOT be interpreted architecturally as proof that a primitive independently or exclusively realizes the complete capability.
20.4 Canonical P3 Architectural Responsibilities
M06 establishes thirteen canonical P3 responsibilities.
| ID | Architectural Responsibility | Type | Applicability |
|---|---|---|---|
| P3-RESP-01 | Intelligence Capability Requirement Interpretation | Enablement / Contract | Baseline-mandatory |
| P3-RESP-02 | Primitive Discovery / Availability Determination | Enablement | Baseline-mandatory |
| P3-RESP-03 | Capability-to-Primitive Contribution Mapping | Contract / Interoperability | Baseline-mandatory |
| P3-RESP-04 | Primitive Type & Mechanism Identity Preservation | Contract / Evidence | Baseline-mandatory |
| P3-RESP-05 | Primitive Eligibility Determination | Assurance / Control | Baseline-mandatory |
| P3-RESP-06 | Primitive Version & Currentness Determination | Lifecycle / Evolution | Conditional-mandatory |
| P3-RESP-07 | Primitive Compatibility Determination | Interoperability / Safety | Conditional-mandatory |
| P3-RESP-08 | Multi-Primitive Composition Determination | Coordination | Conditional-mandatory |
| P3-RESP-09 | Primitive Substitution / Fallback Governance | Resilience / Safety | Conditional-mandatory |
| P3-RESP-10 | Primitive Provision to P4 | Enablement / Contract | Baseline-mandatory |
| P3-RESP-11 | Primitive Evidence & Traceability | Observation / Evidence | Baseline-mandatory |
| P3-RESP-12 | Primitive Change / Invalidation Propagation | Lifecycle / Evolution | Conditional-mandatory |
| P3-RESP-13 | Evaluation / Assurance Feedback Intake | Evaluation / Assurance | Conditional-mandatory |
P3-RESP-01 - Intelligence Capability Requirement Interpretation
P3 must be able to receive or derive the applicable Intelligence Capability requirement from Product/solution intent, an architecture profile, P4 demand or another governed source without redefining AIPS capability semantics.
A requirement may be multi-valued, for example:
Reasoning + Generation
or
Perception + Planning + Control
GEAIRA-M06-R09 - Capability requirement traceability: P3 SHALL preserve the authoritative capability identity and source of each requirement used to determine mechanism realization.
P3-RESP-02 - Primitive Discovery / Availability Determination
P3 must determine which AI Primitive instances/versions are available within the applicable enterprise, Product, platform or federated architecture boundary.
Discovery may be realized through configuration, metadata, a catalog, deployment state, Product dependency declarations, runtime advertisements or another mechanism. GEAIRA does not require a central Primitive Registry.
GEAIRA-M06-R10 - Discovery neutrality: Primitive availability SHALL be architecturally discoverable/determinable where required, but GEAIRA SHALL NOT mandate a central registry or catalog implementation.
P3-RESP-03 - Capability-to-Primitive Contribution Mapping
P3 must preserve the relationship between each candidate primitive and the Intelligence Capability or capabilities to which it materially contributes.
This relationship may be declared, evidenced, evaluated or profile-constrained, but must not be inferred merely from a primitive's label.
GEAIRA-M06-R11 - No label inference: A primitive type or implementation name SHALL NOT by itself establish the Intelligence Capability it contributes.
P3-RESP-04 - Primitive Type & Mechanism Identity Preservation
P3 must preserve authoritative primitive identity and primitive-type classification.
For example:
AIModel -> AI Primitive
Symbolic Reasoner -> AI Primitive
Planner -> AI Primitive
Optimizer -> AI Primitive
but:
AI Primitive != AI Model
GEAIRA-M06-R12 - Mechanism identity preservation: P3 SHALL preserve primitive type/identity information required for traceability, compatibility, policy, evaluation and change management.
P3-RESP-05 - Primitive Eligibility Determination
A primitive may exist and be technically callable yet still be ineligible for the intended use because of policy, lifecycle, evidence, compatibility, risk, licensing, geographic, safety, evaluation or other governed constraints.
Eligibility is an architectural decision for the declared context; it is not a new semantic class and is not AIPS AI Product qualification.
GEAIRA-M06-R13 - Eligibility before provision: P3 SHALL NOT provide a primitive to P4 as an approved realization option where a material required eligibility condition is known to be unsatisfied.
GEAIRA-M06-R14 - Eligibility context: Primitive eligibility SHALL be evaluated within the applicable purpose, authority, Product/profile, capability and lifecycle context rather than assumed globally.
P3-RESP-06 - Primitive Version & Currentness Determination
Where primitive versions or mutable mechanism identities exist, P3 must be able to determine which version is current/allowed for the applicable use and preserve the exact version identity supplied to P4.
GEAIRA-M06-R15 - Version-specific traceability: Where primitive version materially affects behavior, evidence or compatibility, P3 SHALL preserve the exact primitive version or equivalent immutable execution identity.
P3-RESP-07 - Primitive Compatibility Determination
P3 must determine relevant compatibility before presenting a realization set to P4.
Compatibility may include:
- input/output contract compatibility;
- context/interface compatibility;
- execution-environment compatibility;
- primitive-to-primitive compatibility;
- version compatibility;
- latency/resource envelopes;
- security/trust-zone compatibility;
- policy/governance constraints.
GEAIRA-M06-R16 - Compatibility explicitness: Material primitive compatibility constraints SHALL be explicit and traceable rather than assumed from co-location, vendor identity or common framework use.
P3-RESP-08 - Multi-Primitive Composition Determination
A capability may require multiple primitives, and a composite AI mechanism may provide multiple capabilities.
P3 must determine the architecture-level mechanism set and dependencies required for the declared capability realization without turning that composition into a Product identity claim.
GEAIRA-M06-R17 - Composition/Product distinction: A multi-primitive composition SHALL NOT by itself imply Compound Product, AI Product or Agent identity.
P3-RESP-09 - Primitive Substitution / Fallback Governance
Where alternative primitives can satisfy an architecture requirement, P3 must establish the substitution/fallback constraints under which a different primitive may be supplied.
Substitution may alter capability quality, effect characteristics, latency, explainability, risk or assurance state and therefore cannot be treated as an invisible implementation detail where those differences are material.
GEAIRA-M06-R18 - Controlled substitution: Material primitive substitution SHALL occur only within declared compatibility, policy and assurance constraints and SHALL remain traceable.
P3-RESP-10 - Primitive Provision to P4
P3 provides P4 with an eligible primitive or primitive set plus the information required for execution binding.
This may be an endpoint, embedded artifact, dependency identifier, runtime capability handle, packaged model, rule set, planner, callable service, local library, federated capability or another realization.
GEAIRA-M06-R19 - Provision/invocation distinction: P3 primitive provision SHALL NOT be treated as evidence that P4 invoked that primitive.
P3-RESP-11 - Primitive Evidence & Traceability
P3 must preserve evidence sufficient to explain why a primitive was considered available/eligible and how it was related to required capabilities where those facts are material to assurance, audit or reproducibility.
GEAIRA-M06-R20 - Evidence without duplication: P3 SHOULD preserve authoritative references/evidence rather than duplicate entire model artifacts or evaluation corpora solely for traceability unless a governed retention requirement demands a copy.
P3-RESP-12 - Primitive Change / Invalidation Propagation
Retirement, supersession, evidence expiry, policy change, security issue, model update or compatibility break may invalidate a previously eligible primitive.
P3 must propagate the change to affected consumers/control paths where continued use may be impacted.
GEAIRA-M06-R21 - Invalidation propagation: Material primitive invalidation or supersession SHALL be propagated to affected P4/P5/P7 responsibilities with sufficient identity to determine impact.
P3-RESP-13 - Evaluation / Assurance Feedback Intake
P7 evaluation/assurance may determine that a primitive or version is no longer fit for a capability/use context or is subject to conditions.
P3 must consume those findings where they affect future eligibility, substitution or provision.
GEAIRA-M06-R22 - Assurance feedback consumption: P3 SHALL incorporate applicable P7 findings into future primitive eligibility/currentness decisions without rewriting historical execution evidence.
20.5 Canonical P3 ABB admission
Applying the M02 ABB admission criteria, M06 admits three logical ABBs.
P3-ABB-01 - Capability Realization Boundary
Purpose: Own the architecture boundary that translates governed Intelligence Capability requirements into explicit capability-to-primitive contribution requirements and realization options.
Principal responsibilities:
- P3-RESP-01 Capability Requirement Interpretation;
- P3-RESP-03 Capability-to-Primitive Contribution Mapping;
- part of P3-RESP-08 Multi-Primitive Composition Determination;
- traceability to Product/profile/solution requirements where applicable.
This ABB does not own the AIPS capability taxonomy and does not execute AI.
P3-ABB-02 - Primitive Eligibility & Compatibility Boundary
Purpose: Own the architecture boundary that determines whether a candidate primitive/version is eligible and compatible for the declared realization context.
Principal responsibilities:
- P3-RESP-02 Primitive Discovery/Availability;
- P3-RESP-04 Primitive Type/Identity Preservation;
- P3-RESP-05 Eligibility Determination;
- P3-RESP-06 Version/Currentness;
- P3-RESP-07 Compatibility;
- P3-RESP-09 Substitution/Fallback Governance;
- P3-RESP-13 Assurance Feedback Intake.
P3-ABB-03 - Primitive Composition & Provision Boundary
Purpose: Own the technology-neutral boundary for assembling the approved primitive set/dependencies and making that realization set available to P4 under explicit contracts.
Principal responsibilities:
- P3-RESP-08 Multi-Primitive Composition Determination;
- P3-RESP-10 Primitive Provision to P4;
- P3-RESP-11 Evidence/Traceability;
- P3-RESP-12 Change/Invalidation Propagation.
The three ABBs may be co-located, distributed, embedded, federated or realized through managed platform capabilities.
GEAIRA-M06-R23 - ABB logicality: P3-ABB-01 through P3-ABB-03 are logical architecture boundaries and SHALL NOT be interpreted as mandatory standalone deployable services.
GEAIRA-M06-R24 - No execution absorption: P3 ABBs SHALL NOT absorb P4 AI Execution solely because a primitive is hosted or provisioned by the same implementation component.
GEAIRA-M06-R25 - No taxonomy ownership transfer: Admission of P3 ABBs SHALL NOT transfer semantic ownership of AIPS capability or primitive taxonomies to GEAIRA.
20.6 Deliberately non-admitted P3 ABB candidates
The following names are common implementation/component choices but are not canonical generic GEAIRA ABBs merely by name:
- Model Registry
- Model Catalog
- Model Gateway
- Inference Server
- LLM Gateway
- Foundation Model Hub
- Model Router
- Prompt/Model Router
- GPU Serving Platform
- Feature Store
- Rule Engine
- Planner Service
- AI Marketplace
- AI Primitive Registry
- Embedding Model Service
Any of these may realize one or more P3/P4/P5 responsibilities in a concrete architecture. They require independent ABB justification before being promoted by a profile.
GEAIRA-M06-R26 - No model-centric default: GEAIRA SHALL NOT make Model the universal architectural synonym for AI Primitive.
GEAIRA-M06-R27 - No registry reflex: Primitive discoverability, identity or versioning SHALL NOT by itself require a canonical Primitive/Model Registry ABB.
GEAIRA-M06-R28 - No gateway reflex: Multi-provider or multi-model access SHALL NOT by itself require a canonical Model/LLM Gateway ABB.
20.7 P3 architectural interactions
M06 establishes nine technology-neutral P3 interactions.
| ID | Interaction | Typical direction | Purpose |
|---|---|---|---|
| P3-INT-01 | Capability Requirement Provision | P1/P5/profile/P4 -> P3 | State required Intelligence Capability and governing constraints. |
| P3-INT-02 | Primitive Availability Discovery | P3 < - > primitive authority/P5/runtime | Determine available primitive identities/versions. |
| P3-INT-03 | Capability Contribution Declaration | primitive authority/evidence -> P3 | State/evidence capability contribution. |
| P3-INT-04 | Primitive Evidence / Metadata Access | P3 < - > P7/AIPS/authority | Obtain evidence, version/currentness and assurance state. |
| P3-INT-05 | Compatibility Determination | P3 < - > P4/P5/primitive dependencies | Establish execution/dependency compatibility. |
| P3-INT-06 | Primitive Realization Provision | P3 -> P4 | Provide approved primitive or primitive set plus binding metadata. |
| P3-INT-07 | Primitive Use / Outcome Evidence | P4 -> P3/P7 | Return actual primitive/version participation evidence where material. |
| P3-INT-08 | Assurance / Fitness Feedback | P7 -> P3 | Update future eligibility/constraints. |
| P3-INT-09 | Primitive Change / Invalidation | authority/P3 -> P4/P5/P7 | Signal retirement, supersession or eligibility change. |
These interactions may be synchronous, asynchronous, configuration-driven, package-based, embedded, federated or human-mediated.
GEAIRA-M06-R29 - Interaction neutrality: M06 interactions SHALL describe architectural intent and SHALL NOT prescribe REST, gRPC, message bus, model protocol, artifact store, service mesh or cloud-specific mechanism.
GEAIRA-M06-R30 - Occurrence evidence boundary: P4 evidence that an execution involved a primitive SHALL remain occurrence-level evidence and SHALL NOT automatically rewrite Product-version declarations or P3 capability mappings.
20.8 P3 control points
M06 establishes three logical control-point classes.
CP-P3-01 - Primitive Eligibility Control Point
Ensures that a primitive/version presented for use satisfies applicable eligibility constraints.
CP-P3-02 - Primitive Compatibility / Composition Control Point
Ensures that the primitive or primitive set is compatible with the declared capability realization and relevant P4/P5 contracts.
CP-P3-03 - Primitive Change / Invalidation Control Point
Ensures that retirement, supersession, revoked evidence, security issues or assurance findings can prevent or constrain future use where required.
A control point may be realized through static validation, deployment admission, runtime policy, signed metadata, configuration, human approval, platform controls or another mechanism.
GEAIRA-M06-R31 - Control-point/component distinction: P3 control points SHALL NOT automatically become standalone policy engines, gateways or registries.
GEAIRA-M06-R32 - Fail-safe eligibility: Where a required primitive eligibility condition is indeterminate and the uncertainty is material, the architecture SHALL NOT silently treat the primitive as eligible.
20.9 P3 relationship to P1, P2, P4, P5, P6 and P7
P1 -> P3
P1 may constrain which primitives, providers, locations, mechanism types or assurance states are permitted for the declared purpose/authority envelope.
P3 consumes those constraints; it does not redefine them.
P2 < - > P3
P2 supplies context/knowledge responsibilities. Some AI Primitives may consume knowledge resources or embeddings, but the knowledge resource itself does not become an AI Primitive merely because P3/P4 uses it.
P3 -> P4
P3 provides an eligible capability-realization set. P4 invokes/operates primitives during AI Execution and records actual participation/exercised capability/effects where required.
P5 < - > P3
P5 may expose Product/serving/deployment dependencies, endpoints, packaged assets or provider integrations used to realize primitives. P3 determines architectural capability/mechanism eligibility; P5 owns serving/Product integration concerns.
P6
P3 does not decide whether a particular decision or action is authorized. P6 owns point-of-use Authorization and human/action control.
P7 < - > P3
P7 evaluates primitive fitness/performance/assurance and feeds findings/currentness back into future P3 eligibility. P3 preserves the evidence link but does not become the independent assurance authority.
GEAIRA-M06-R33 - Knowledge/primitive distinction: Knowledge Resources, Data Products and Information Products SHALL NOT be classified as AI Primitives merely because an AI Primitive consumes them.
GEAIRA-M06-R34 - P3/P6 separation: Primitive eligibility or technical capability SHALL NOT be interpreted as authorization to make a decision or perform an action.
GEAIRA-M06-R35 - P3/P7 separation: P3 MAY consume assurance findings but SHALL NOT self-certify independent fitness/assurance where P7 separation is required.
20.10 Primitive composition and substitution
P3 supports at least four generic realization forms:
A. Single primitive -> one or more capabilities
B. Multiple primitives -> one capability
C. Multiple primitives -> multiple capabilities
D. Alternative primitive sets -> equivalent/acceptable capability envelope
The architecture must distinguish composition from substitution.
Composition means multiple mechanisms participate in the realization.
Substitution means one eligible mechanism/set may replace another under declared constraints.
A fallback primitive may be technically callable but still unacceptable because its capability quality, evidence, risk, latency, licensing, geography, security or effect profile differs materially.
GEAIRA-M06-R36 - Composition traceability: Where multiple primitives jointly realize a material capability, the architecture SHALL preserve enough dependency identity to reconstruct the intended realization set.
GEAIRA-M06-R37 - Substitution non-equivalence: Two primitives SHALL NOT be treated as architecturally interchangeable solely because they share an Intelligence Capability classification.
GEAIRA-M06-R38 - Effect sensitivity: Where primitive substitution can materially alter expected Intelligence Effects or downstream decision/action behavior, that difference SHALL be visible to the applicable P4/P6/P7 control path.
20.11 Primitive lifecycle, version and historical truth
Primitive evolution must preserve both current and historical truth.
Example:
Model/Primitive v3
|
+--> supplied to Execution E1
later:
Model/Primitive v4 supersedes v3
For future executions, P3 may mark v3 ineligible and select v4.
But the historical statement:
Execution E1 involved v3
must remain true.
Likewise, a later evaluation finding does not retroactively alter which primitive participated in a prior execution; it changes the interpretation/currentness/fitness state associated with future or retrospective assurance.
GEAIRA-M06-R39 - Historical primitive identity: Primitive supersession SHALL NOT rewrite historical execution identity or evidence.
GEAIRA-M06-R40 - Currentness versus freshness: P3 SHALL distinguish governed currentness/eligibility from mere artifact recency; the newest primitive version is not automatically the permitted or fit version.
GEAIRA-M06-R41 - Revalidation on material change: Material primitive change SHALL trigger re-evaluation of affected compatibility/eligibility assumptions where the applicable profile requires it.
20.12 Cross-cutting overlay application to P3
All eight M03 overlays apply to P3.
| Overlay | P3 application |
|---|---|
| X1 Security, Privacy & Trust | Provider/artifact trust, supply-chain integrity, sensitive mechanism access, isolation and security constraints. |
| X2 Identity, Authority & Policy | Who may register/select/provision/use primitives; allowed mechanism/provider constraints. |
| X3 Evidence, Provenance & Traceability | Primitive identity/version, capability-contribution evidence, origin and dependency trace. |
| X4 Observability, Telemetry & Auditability | Availability, selection, substitution, version and binding events where material. |
| X5 Resilience, Safety & Failure Management | Fallbacks, incompatible primitives, unavailable providers, unsafe/invalidated mechanisms. |
| X6 Lifecycle, Version, Currentness & Change | Versioning, supersession, retirement, evidence expiry and revalidation. |
| X7 Interoperability, Contracts & Semantic Compatibility | Primitive input/output/dependency contracts and capability mapping compatibility. |
| X8 Governance, Human Accountability & Compliance | Ownership/accountability for primitive admission, change, exceptions and overrides. |
GEAIRA-M06-R42 - Overlay continuity: P3 SHALL apply relevant X1-X8 obligations without converting each cross-cutting concern into a dedicated P3 service.
20.13 Failure and degradation model
M06 establishes ten generic P3 failure/degradation classes:
- No eligible primitive available - no candidate satisfies the required capability/use constraints.
- Capability contribution unsubstantiated - contribution is claimed but required evidence/authority is absent.
- Primitive unavailable - an otherwise eligible primitive cannot be accessed/provisioned.
- Primitive version non-current / revoked - supplied or requested version is no longer permitted.
- Compatibility failure - primitive does not satisfy required interface/dependency/runtime constraints.
- Composition failure - required multi-primitive dependency set cannot be coherently formed.
- Substitution mismatch - fallback changes material capability/effect/assurance characteristics beyond permitted envelope.
- Evidence/currentness indeterminate - required eligibility evidence is missing or stale.
- Primitive change during use - version/provider/configuration changes in a way that threatens traceability or validity.
- Provision failure - approved realization cannot be bound/provided to P4.
Allowed responses include fail closed, bounded fallback, explicit degraded mode, human escalation, deferred execution, retry, or profile-governed alternative realization.
GEAIRA-M06-R43 - No silent capability downgrade: Material capability loss or substitution SHALL NOT be silently hidden from P4/P6/P7 where it can affect execution, decision, action or assurance.
GEAIRA-M06-R44 - No least-evidence fallback: A primitive with weaker or missing required evidence SHALL NOT automatically be selected merely because the preferred primitive is unavailable.
GEAIRA-M06-R45 - Degradation traceability: Material fallback/degradation SHALL preserve the actual primitive identity/version and reason for substitution where required for evidence or assurance.
20.14 Canonical P3 reference view
The canonical logical P3 view is:
P1 GOVERNING ENVELOPE / P5 PRODUCT-PROFILE CONTEXT
|
capability + constraints
|
v
+--------------------------------+
| P3-ABB-01 |
| Capability Realization |
| Boundary |
| |
| capability requirement |
| contribution mapping |
| realization options |
+---------------+----------------+
|
candidate realization
|
v
+--------------------------------+
| P3-ABB-02 |
| Primitive Eligibility & |
| Compatibility Boundary |
| |
| identity / type / version |
| eligibility / currentness |
| compatibility / fallback |
+---------------+----------------+
|
approved primitive set
|
v
+--------------------------------+
| P3-ABB-03 |
| Primitive Composition & |
| Provision Boundary |
| |
| dependency set / binding |
| provision / traceability |
| change propagation |
+---------------+----------------+
|
realization provision
|
v
P4
AI Execution
|
actual-use evidence / outcomes
v
P7
Evaluation / Assurance
|
findings / fitness / currentness
+-----------------> P3
This is a logical responsibility view, not a deployment topology.
20.15 Representative P3 scenarios
Scenario A - Learned model realizes classification and detection
A learned-model primitive contributes to Classification and Detection. P3 preserves the primitive's AIPS type/identity, maps its evidenced capability contribution, validates the required version/compatibility and provides it to P4.
Nothing in the architecture states that every AI Primitive must be a learned model.
Scenario B - Symbolic reasoner for policy reasoning
A symbolic reasoner contributes Reasoning without a trained ML model.
P3 treats it as an AI Primitive according to AIPS semantics and may provide it to P4 if eligible. A learned-model dependency is not required merely to make the architecture look "AI-like".
Scenario C - Multi-primitive reasoning + generation
A solution uses a knowledge reasoner plus a learned generative model. The reasoner contributes Reasoning while the model contributes Generation and may also contribute Reasoning.
P3 records the realization composition rather than inventing a new primitive type or assuming one primitive exclusively owns the capability.
Scenario D - Primitive fallback changes assurance state
A preferred model version becomes unavailable. A fallback model can provide the same high-level Generation capability but has a weaker evaluation basis.
P3 does not treat the fallback as silently equivalent. It applies the declared fallback/assurance constraints and makes degradation visible to P4/P7.
Scenario E - Model Product is still not automatically AI Product
A model is productized and is therefore a Model Product and AI Primitive under AIPS semantics.
P3 may architect its availability and capability contribution, but does not infer AI Product qualification. BPS/AIPS Product qualification remains external.
Scenario F - Primitive superseded after historical execution
Execution E1 used primitive version 3. Version 4 later supersedes it and P7 finds v3 unsuitable for future use.
P3 makes v3 ineligible for new use while preserving that E1 historically involved v3.
20.16 P3 anti-patterns
M06 defines the following P3-specific anti-patterns:
- Model-equals-AI architecture - treating all AI mechanisms as learned models.
- Capability-equals-primitive - treating Reasoning as synonymous with a Reasoner, Planning with Planner, etc.
- Primitive-equals-Product - assuming primitive existence/productization establishes AI Product qualification.
- Registry-equals-architecture - making a Model/Primitive Registry the defining P3 architecture.
- Gateway-equals-interoperability - assuming model-provider diversity requires one universal gateway ABB.
- Newest-is-current - selecting the latest primitive version without governed currentness/eligibility.
- Same-capability-equals-substitutable - treating any two primitives with the same capability label as equivalent.
- Single-primitive assumption - forcing each capability to one primitive and losing composite realization.
- Black-box composition - hiding primitive dependency composition such that execution cannot be reconstructed.
- Execution leakage into P3 - moving runtime invocation/execution logic into P3 because the same platform hosts primitives.
- Knowledge-resource-as-primitive - classifying data/knowledge repositories as AI mechanisms merely because AI uses them.
- Provider-lock semantics - encoding vendor/provider names into generic capability semantics.
- Evidence-free admission - accepting primitive capability claims where the applicable profile requires evidence but none exists.
- Silent fallback - substituting a materially different primitive without traceability or control-path visibility.
- Historical rewrite - replacing historical primitive identity with the current version after supersession.
20.17 M06 frozen decisions
M06 freezes the following P3 architecture decisions for GEAIRA 0.1:
P3 owns capability-to-mechanism realization responsibilities.
Intelligence Capability != AI Primitive.
AI Primitive != AI Model.
AI Primitive != AI Product.
Primitive availability/eligibility != runtime execution.
Primitive eligibility != Authorization.
Capability contribution may be many-to-many.
Contribution != exclusive/sufficient realization.
Primitive composition != Product or Agent identity.
Same capability label != automatic substitutability.
Primitive version/currentness/evidence must be preserved where material.
P4 records actual runtime involvement/exercised capability.
P7 evaluates fitness/assurance independently and feeds findings to P3.
Historical primitive identity survives supersession.
The following P3 ABBs are canonical:
P3-ABB-01 Capability Realization Boundary
P3-ABB-02 Primitive Eligibility & Compatibility Boundary
P3-ABB-03 Primitive Composition & Provision Boundary
The following remain realization choices rather than generic canonical ABBs:
Model Registry / Model Catalog
Model Gateway / LLM Gateway
Inference Server
Foundation Model Hub
Model Router
GPU Serving Platform
Feature Store
Rule Engine
Planner Service
AI Marketplace
AI Primitive Registry
Embedding Model Service
M06 does not freeze:
- vendor/provider selection;
- model serving protocol;
- model packaging format;
- artifact registry technology;
- accelerator/GPU topology;
- inference/runtime implementation;
- primitive evaluation metrics;
- universal model-routing algorithm;
- Product deployment mechanics;
- final P4 execution architecture;
- final P5 serving/integration contracts;
- final P7 assurance criteria;
- final M11 inter-plane interaction contracts.
GEAIRA-M06-R46 - Frozen ABB restraint: M06 SHALL freeze only the three P3 logical ABBs above and SHALL NOT promote model-serving, registry, gateway or accelerator technologies to canonical ABB status.
GEAIRA-M06-R47 - Downstream preservation: Later GEAIRA slices SHALL preserve the M06 capability/mechanism, P3/P4, Product and historical-identity distinctions unless the 0.1 programme explicitly records and reconciles a justified architecture change without redefining GEAIRM/AIPS semantics.
GEAIRA-M06-R48 - Product declaration isolation: P3 occurrence/availability evidence SHALL NOT automatically populate or modify AIPS AI Product Version capability, execution-mode or effect declarations.
GEAIRA-M06-R49 - Primitive type non-inference: P3 SHALL NOT infer a primitive type solely from observed Intelligence Capability or Intelligence Effect.
GEAIRA-M06-R50 - No execution-mode ownership: P3 SHALL NOT treat AIPS Execution Mode as a P3 mechanism classification; runtime execution-mode realization belongs principally to P4/P5 interaction.
GEAIRA-M06-R51 - No effect ownership: P3 SHALL NOT treat expected or observed Intelligence Effects as primitive identity; effects belong to execution/outcome semantics and downstream interpretation.
GEAIRA-M06-R52 - Mechanism neutrality continuity: The P3 architecture SHALL remain valid for learned, symbolic, knowledge-based, planning, intelligent-search, optimization, hybrid and future AIPS-recognized AI Primitive types without structural redesign solely because the mechanism family differs.
21. M06 Completion State
GEAIRA-M06 - COMPLETE.
M06 establishes:
- one canonical P3 purpose and architecture boundary;
- thirteen canonical P3 Architectural Responsibilities;
- three canonical logical P3 ABBs;
- fifteen deliberately non-admitted implementation-style ABB candidates;
- nine technology-neutral P3 architectural interactions;
- three logical P3 control-point classes;
- ten generic P3 failure/degradation classes;
- explicit P3 application of X1-X8 cross-cutting overlays;
- explicit P3/P4/P5/P7 responsibility boundaries;
- six representative P3 scenarios;
- fifteen P3 anti-patterns;
- 52 normative M06 rules.
M06 introduces no change to GEAIRM or AIPS semantics, does not prescribe model-centric architecture, deployment topology, model gateway, model registry, inference server or accelerator platform.
Next slice
GEAIRA-M07 - P4: AI Execution, Agentic & Environment
M07 will apply the responsibility/ABB discipline to AI Execution, execution mode, Agent participation, Tool use, Environment, sensing/actuation, control loops, runtime orchestration and the boundary between execution behavior and P6 decision/action authorization.
22. GEAIRA-M07 - P4: AI Execution, Agentic & Environment
22.1 Purpose
GEAIRA-M07 establishes the canonical architecture baseline for P4 - AI Execution, Agentic & Environment.
P4 is the occurrence-level runtime plane in which eligible AI mechanisms are actually invoked or enacted, bounded AI Context is consumed, Agent participation is coordinated, Tools may be used, Execution Mode is realized, Environment interaction may occur, Control Loops may operate, and Intelligence Effects may be produced or caused.
P4 answers:
How should enterprise AI executions be coordinated and enacted at runtime while preserving authority, context, mechanism, agent, environment, decision/action and evidence boundaries?
P4 does not redefine AI Execution, Agent, Tool, Environment, Sensing, Actuation, Control Loop, Decision Formation, Authorization, Intelligence Capability, Execution Mode or Intelligence Effect. It consumes those semantics from GEAIRM, AIPS and the KivaNura Agent Ontology.
GEAIRA-M07-R01 - Occurrence-level scope: P4 SHALL treat AI Execution as an occurrence-level runtime concern and SHALL NOT reinterpret AI Execution as a deployment, runtime platform, Product, Agent, ABB, Execution Mode or Intelligence Effect.
GEAIRA-M07-R02 - Runtime boundary: P4 SHALL own the architecture responsibilities required to coordinate and enact AI Execution while preserving P1 authority, P2 context, P3 mechanism, P5 serving/integration, P6 decision/action control and P7 assurance boundaries.
22.2 Source semantic baseline
M07 is grounded principally in the following frozen GEAIRM/AIPS/Agent semantics:
GEAIRM AI Execution
GEAIRM Delegated Authority
GEAIRM Tool Role
GEAIRM Environment
GEAIRM Sensing
GEAIRM Actuation
GEAIRM Control Loop
GEAIRM Decision Formation
GEAIRM Authorization
AIPS AI Primitive
AIPS Intelligence Capability
AIPS Execution Mode
AIPS Intelligence Effect
AIPS Embodiment / Agency / Autonomy semantics where applicable
KivaNura Agent Ontology Agent and Agent taxonomy
The principal GEAIRM relationships architecturalized by P4 include:
AI Execution involves AI Primitive
AI Execution exercises Intelligence Capability
AI Execution operates in Execution Mode
AI Execution produces or causes Intelligence Effect
AI Execution has AI Context
Agent participates in AI Execution
Agent operates under Delegated Authority
Agent / AI Execution uses Tool
AI Execution occurs in Environment
Agent / AI Execution senses Environment
Agent / AI Execution acts upon Environment
Agent / AI Execution participates in Control Loop
Agent / AI Execution participates in Decision Formation
AI Execution / invocation / Action may be authorized by Authorization
Authorization is within Delegated Authority
participants may participate in Orchestration
These semantic relationships constrain architecture but do not each require a corresponding service, registry, API or standalone component.
GEAIRA-M07-R03 - Semantic non-invention: P4 SHALL preserve the frozen GEAIRM/AIPS/Agent meanings and non-implications of execution, agent, tool, environment, sensing, actuation, control-loop and authorization relationships.
GEAIRA-M07-R04 - Relationship/component separation: A P4 semantic relationship SHALL NOT automatically create a one-to-one runtime component or service.
22.3 P4 architecture boundary
P4 owns architectural responsibilities for:
- creation and coordination of an AI Execution occurrence;
- binding the execution to governed context and eligible primitive realization information;
- realizing the applicable Execution Mode at runtime;
- coordinating Agent participation where present;
- coordinating Tool use where present;
- coordinating runtime orchestration among execution participants;
- receiving observations / sensing Environment state where applicable;
- technically enacting already-authorized Environment interaction or Actuation where applicable;
- participating in Control Loops where applicable;
- recording occurrence-level primitive participation, capability exercise, effects and runtime evidence;
- maintaining runtime state, correlation and failure/degradation visibility;
- applying runtime restrictions received from P1/P2/P3/P6/P7.
P4 does not own:
- enterprise purpose or governing intent - P1;
- source/resource grounding or context construction - P2;
- primitive eligibility or capability-to-mechanism realization - P3;
- Product serving boundary or external integration contract - P5;
- final point-of-use Authorization, consequential Decision commitment or Action permission - P6;
- independent evaluation, assurance or longitudinal currentness determination - P7.
GEAIRA-M07-R05 - P3/P4 separation: P3 provision of an eligible primitive realization SHALL NOT be treated as actual runtime invocation or proof that the primitive participated in a particular AI Execution.
GEAIRA-M07-R06 - P2/P4 separation: P2 provision of AI Context SHALL NOT by itself prove that every context resource materially influenced the execution.
GEAIRA-M07-R07 - P1/P4 separation: A governing or delegated authority envelope SHALL constrain P4 but SHALL NOT be treated as a point-of-use Authorization for every execution, Tool invocation, decision or action.
GEAIRA-M07-R08 - P6/P4 separation: P4 SHALL NOT convert technical ability to invoke, decide, actuate or call a Tool into permission to do so; applicable point-of-use Authorization and consequential decision/action control belong principally to P6.
GEAIRA-M07-R09 - P7/P4 separation: Runtime telemetry or execution evidence emitted by P4 SHALL NOT be treated as independent assurance, fitness determination or governance approval.
22.4 Canonical P4 Architectural Responsibilities
M07 defines fourteen canonical P4 responsibilities.
| ID | Architectural Responsibility | Primary obligation class | Architectural outcome |
|---|---|---|---|
| P4-RESP-01 | Execution Occurrence Establishment | Baseline-mandatory | Establish a distinct, traceable AI Execution occurrence with stable correlation identity and bounded runtime scope. |
| P4-RESP-02 | Context Binding & Consumption | Baseline-mandatory | Bind the applicable P2-provided AI Context to the execution and preserve which context was made available. |
| P4-RESP-03 | Primitive Realization Binding | Baseline-mandatory | Bind the execution to the P3-approved primitive realization set actually eligible for use. |
| P4-RESP-04 | Execution Mode Realization | Conditional-mandatory | Realize and record the applicable AIPS Execution Mode without treating it as primitive identity or deployment topology. |
| P4-RESP-05 | Runtime Invocation & Coordination | Baseline-mandatory | Invoke/enact qualifying primitives and coordinate execution steps/participants in a bounded occurrence. |
| P4-RESP-06 | Agent Participation Coordination | Conditional-mandatory | Coordinate Agent participation while preserving Agent identity, Agent type, autonomy/control-mode and authority boundaries. |
| P4-RESP-07 | Tool-Use Coordination | Conditional-mandatory | Coordinate use of resources playing the Tool Role with explicit runtime scope and applicable authority/control checks. |
| P4-RESP-08 | Runtime Orchestration | Conditional-mandatory | Coordinate multi-step/multi-participant execution flow where orchestration is required without equating orchestration with Agent or AI Product identity. |
| P4-RESP-09 | Environment Observation / Sensing | Conditional-mandatory | Acquire or receive relevant Environment observations/signals where the execution or Agent senses an Environment. |
| P4-RESP-10 | Environment Interaction / Actuation Execution | Conditional-mandatory | Technically enact an authorized Environment interaction or Actuation while preserving the distinction between capability, command, Authorization and observed effect. |
| P4-RESP-11 | Control-Loop Participation | Conditional-mandatory | Coordinate recurring sense-evaluate-decide/authorize-act-observe participation without inferring autonomy or Control Intelligence Capability. |
| P4-RESP-12 | Occurrence-Level Effect & Outcome Recording | Baseline-mandatory | Record Intelligence Effects or relevant outcomes produced/caused by the execution without treating effect as success, authorization or Product classification. |
| P4-RESP-13 | Runtime Evidence, State & Correlation | Baseline-mandatory | Emit sufficient occurrence-level evidence/state/correlation for traceability, observability, reconstruction and downstream P7 assessment. |
| P4-RESP-14 | Runtime Failure, Interruption & Degradation Handling | Baseline-mandatory | Make execution failure, partial completion, cancellation, timeout, degraded mode and control interruption explicit and governable. |
GEAIRA-M07-R10 - Execution identity: Each materially governed AI Execution SHALL be distinguishable as an occurrence for traceability and reconstruction.
GEAIRA-M07-R11 - Context binding evidence: P4 SHALL preserve which AI Context was made available to the execution where that context is material to governed interpretation.
GEAIRA-M07-R12 - Primitive-use evidence: P4 SHALL distinguish a P3-approved primitive from the primitive version(s) actually invoked or involved in the occurrence.
GEAIRA-M07-R13 - Execution-mode fidelity: Where Execution Mode is material, P4 SHALL preserve the mode actually realized for the occurrence and SHALL NOT infer deployment topology from that classification.
GEAIRA-M07-R14 - Agent identity preservation: Where an Agent participates, P4 SHALL preserve the Agent's authoritative identity/reference and SHALL NOT mint a shadow Agent identity merely for runtime convenience.
GEAIRA-M07-R15 - Authority non-inference: Agent participation, Tool availability, autonomy classification or technical capability SHALL NOT be treated as Authorization.
GEAIRA-M07-R16 - Tool-role contextuality: P4 SHALL preserve that Tool is a contextual role and SHALL NOT change the intrinsic identity or Product classification of the resource playing that role.
GEAIRA-M07-R17 - Sensing/perception distinction: P4 SHALL NOT infer Perception Intelligence Capability merely because Environment sensing or observation occurred.
GEAIRA-M07-R18 - Actuation/authority distinction: P4 SHALL NOT infer permission, successful effect or AIPS Embodiment classification merely because technical Actuation or Environment interaction occurred.
GEAIRA-M07-R19 - Control-loop distinction: Participation in a Control Loop SHALL NOT by itself imply autonomous operation, Control Intelligence Capability or ClosedLoopAutonomous Embodiment.
GEAIRA-M07-R20 - Effect non-equivalence: An Intelligence Effect produced or caused by an execution SHALL NOT be treated as proof of successful business outcome, accepted decision, Authorization or durable artifact.
22.5 Canonical P4 ABB admission
Applying the M02 ABB admission criteria, M07 admits three canonical logical P4 ABBs.
P4-ABB-01 - Execution Coordination Boundary
Purpose:
Establish, bind, coordinate and close a traceable AI Execution occurrence.
Primary responsibilities:
- P4-RESP-01 Execution Occurrence Establishment;
- P4-RESP-02 Context Binding & Consumption;
- P4-RESP-03 Primitive Realization Binding;
- P4-RESP-04 Execution Mode Realization;
- P4-RESP-05 Runtime Invocation & Coordination;
- P4-RESP-08 Runtime Orchestration;
- P4-RESP-12 Occurrence-Level Effect & Outcome Recording;
- P4-RESP-13 Runtime Evidence, State & Correlation;
- P4-RESP-14 Runtime Failure, Interruption & Degradation Handling.
This ABB is not synonymous with an inference server, model runtime, workflow engine, agent framework, serverless function, container platform or universal orchestration service.
P4-ABB-02 - Agent & Tool Participation Boundary
Purpose:
Coordinate Agent participation and contextual Tool use within a bounded AI Execution while preserving identity, authority, control and tool-role distinctions.
Primary responsibilities:
- P4-RESP-06 Agent Participation Coordination;
- P4-RESP-07 Tool-Use Coordination;
- relevant runtime authority/control checks from P1/P6;
- occurrence-level traceability for Agent and Tool participation.
This ABB is conditional: an AI Execution need not involve an Agent or a Tool.
It is not synonymous with an agent framework, tool registry, plugin marketplace, MCP server, function-calling layer, IAM system or autonomous-agent platform.
P4-ABB-03 - Environment & Control Interaction Boundary
Purpose:
Coordinate runtime sensing, observation, authorized Environment interaction, technical Actuation and Control-Loop participation while preserving P6 authorization/action boundaries.
Primary responsibilities:
- P4-RESP-09 Environment Observation / Sensing;
- P4-RESP-10 Environment Interaction / Actuation Execution;
- P4-RESP-11 Control-Loop Participation;
- relevant runtime evidence from P4-RESP-13;
- relevant failure/degradation handling from P4-RESP-14.
This ABB is conditional and need not exist as a distinct realization where the AI has no material Environment/control-loop interaction.
It is not synonymous with a robotics controller, digital-twin platform, IoT gateway, sensor hub, actuator gateway, PLC, ROS deployment, SCADA system or physical-AI platform.
GEAIRA-M07-R21 - Three-ABB baseline: The canonical P4 logical ABB set for GEAIRA 0.1 SHALL be Execution Coordination Boundary, Agent & Tool Participation Boundary, and Environment & Control Interaction Boundary.
GEAIRA-M07-R22 - Conditional ABB applicability: P4-ABB-02 and P4-ABB-03 SHALL be treated as conditional architecture boundaries and SHALL NOT be required for executions that have no applicable Agent/Tool or Environment/control interaction.
GEAIRA-M07-R23 - Co-location neutrality: The three P4 ABBs MAY be co-located or distributed in concrete realizations without losing their logical responsibility distinctions.
GEAIRA-M07-R24 - No runtime-platform equivalence: No P4 ABB SHALL be interpreted as a mandatory generic "AI Runtime Platform" component.
22.6 Deliberately non-admitted P4 ABB candidates
M07 does not canonically admit the following implementation-shaped candidates:
AI Runtime Platform
Inference Server
LLM Runtime
Agent Framework
Agent Runtime
Workflow Engine
Task Queue
Tool Registry
Plugin Marketplace
MCP Gateway / MCP Server
Function-Calling Service
Session Store
Execution State Store
Robot Controller
IoT Gateway
Sensor Hub
Actuator Gateway
Digital Twin Platform
ROS / ROS2 Runtime
SCADA / PLC Platform
These may realize one or more P4 responsibilities in a concrete architecture profile.
GEAIRA-M07-R25 - Technology non-promotion: Runtime, agentic, robotics, workflow and tool-use technologies SHALL NOT become canonical P4 ABBs solely because they are common implementation choices.
22.7 Execution occurrence lifecycle
P4 uses the following generic logical occurrence lifecycle for architectural reasoning:
Established
|
v
Bound
|
v
Ready
|
v
Running
|
+--> Suspended / Waiting
|
+--> Degraded
|
+--> Cancelled
|
+--> Failed
|
v
Completed
These are architecture-level occurrence phases, not new GEAIRM ontology states and not a mandatory runtime state machine.
A concrete profile may specialize the lifecycle while preserving occurrence identity and material transitions.
GEAIRA-M07-R26 - Lifecycle non-ontology: The P4 occurrence lifecycle SHALL be treated as an architecture reasoning model and SHALL NOT mint replacement GEAIRM Lifecycle State semantics.
GEAIRA-M07-R27 - Material transition visibility: Material suspension, degradation, cancellation, failure or completion transitions SHALL be visible where they affect downstream interpretation, safety, assurance or action.
GEAIRA-M07-R28 - Partial completion visibility: P4 SHALL NOT silently report a materially partial execution as fully successful.
22.8 Agentic execution architecture
Agentic execution is a conditional P4 specialization, not the definition of enterprise AI.
Agent
|
| participates in
v
AI Execution
|
+--> may consume AI Context
+--> may involve AI Primitive(s)
+--> may use Tool(s)
+--> may participate in Decision Formation
+--> may interact with Environment
The Agent remains governed by the KivaNura Agent Ontology.
P4 may coordinate:
- Agent session/episode correlation;
- multi-step execution;
- Tool-use requests/results;
- sub-task delegation where authority permits;
- execution continuation/termination;
- Agent-to-Agent or Agent-to-resource coordination;
- runtime state/evidence.
P4 must not infer:
- Agent type from Tool use;
- autonomy from multi-step behavior;
- authority from autonomy;
- AI Product status from Agent participation;
- Agent identity from an AI Primitive;
- Agent accountability from mere execution participation.
GEAIRA-M07-R29 - Agent/primitive separation: P4 SHALL preserve Agent identity separately from AI Primitive identity even where one runtime implementation hosts both.
GEAIRA-M07-R30 - Autonomy/authority separation: P4 SHALL NOT infer Delegated Authority or Authorization from Agent autonomy, Tool Use Pattern, control mode or observed self-directed behavior.
GEAIRA-M07-R31 - Agent/Product separation: Agent participation SHALL NOT imply Product or AI Product identity.
22.9 Tool-use architecture
A Tool is a contextual role. P4 therefore architectures governed use, not a universal Tool object model.
A Tool-use occurrence may require:
Tool identity/reference
Tool capability / operation requested
requesting Agent / AI Execution
applicable governing envelope
point-of-use Authorization where required
input / context boundary
result / error
side-effect classification
correlation / evidence
A Tool may be read-only, effect-producing, internal, external, human-mediated or another Product/service/resource.
GEAIRA-M07-R32 - Tool invocation control: P4 SHALL expose applicable Tool invocation to P6 authorization/control where the Tool can cause consequential side effects or where policy requires point-of-use approval.
GEAIRA-M07-R33 - Tool-result non-trust: A Tool result SHALL NOT automatically be treated as correct, authoritative, grounded or safe merely because the invocation succeeded.
GEAIRA-M07-R34 - Tool failure visibility: Tool timeout, rejection, partial result or side-effect uncertainty SHALL remain visible to the execution and downstream evidence path.
22.10 Environment, sensing, actuation and control loops
P4 distinguishes four concerns:
Environment
surrounding domain/state with which execution/Agent interacts
Sensing
acquisition of observations/signals/state
Actuation
technical application of command/control/intervention
Control Loop
recurring observe/evaluate/decide-authorize/act/observe pattern
The generic reference pattern is:
Environment
|
| observation / sensing
v
P4 execution / Agent participation
|
+--> intelligence processing
|
+--> P6 decision / Authorization when applicable
|
| authorized command / actuation request
v
P4 Environment & Control Interaction Boundary
|
| technical interaction / Actuation
v
Environment
|
+--> changed state -> subsequent observation
The exact placement of low-level control execution may vary by safety, latency, physical-control and operational requirements.
For hard real-time or safety-certified control, a concrete profile may keep deterministic control below or beside generic AI execution while still exposing governed AI influence and evidence.
GEAIRA-M07-R35 - Environment/context distinction: P4 SHALL preserve Environment as distinct from AI Context even when Environment observations become context resources.
GEAIRA-M07-R36 - Observation/source distinction: A sensed Environment observation SHALL NOT automatically become a qualified Grounding Source; P2 source/grounding rules still apply where grounding is claimed.
GEAIRA-M07-R37 - Authorized actuation: Where Actuation or Environment interaction is consequential, P4 SHALL require the applicable P6 decision/Authorization outcome before technical enactment unless an explicitly governed pre-authorized control envelope applies.
GEAIRA-M07-R38 - Pre-authorized loop constraint: A pre-authorized control envelope SHALL remain bounded by scope, conditions, currentness, safety controls and revocation semantics and SHALL NOT become unrestricted future permission.
GEAIRA-M07-R39 - Hard-real-time neutrality: GEAIRA SHALL NOT require consequential hard-real-time control decisions to traverse a centralized P4 or P6 network service; profiles MAY realize control responsibilities locally/embedded while preserving logical responsibilities and evidence.
22.11 P4/P6 decision and action boundary
P4 may participate in Decision Formation and may produce an Intelligence Effect classified as a recommendation, score, plan or decision-like effect.
That does not establish final authority.
P4 execution output / Intelligence Effect
|
v
Decision Formation
|
v
P6 point-of-use decision / Authorization / human oversight
|
v
authorized Action / command
|
v
P4 technical enactment where applicable
Examples:
Recommendation != Decision authority
Decision Intelligence Effect != organizational Decision authority
Instruction != Authorization
Delegated Authority != occurrence Authorization
technical actuation != permission to actuate
GEAIRA-M07-R40 - Decision-formation non-authority: P4 participation in Decision Formation SHALL NOT establish final decision authority or organizational accountability.
GEAIRA-M07-R41 - Effect/decision separation: A decision-like Intelligence Effect SHALL NOT bypass P6 authorization/decision controls where those controls are applicable.
GEAIRA-M07-R42 - Authorization scope: An Authorization SHALL be treated as context-bounded and SHALL NOT be generalized into unrestricted future runtime authority.
22.12 Canonical P4 architectural interactions
M07 defines eleven technology-neutral P4 interactions.
| ID | Interaction | Direction | Purpose |
|---|---|---|---|
| P4-INT-01 | Governing Envelope Provision | P1 -> P4 | Provide applicable purpose, authority and governing constraints. |
| P4-INT-02 | AI Context Provision | P2 -> P4 | Provide bounded context/reference and relevant currentness information. |
| P4-INT-03 | Primitive Realization Provision | P3 -> P4 | Provide eligible primitive/composition/version realization information. |
| P4-INT-04 | Execution Invocation / Serving Request | P5 -> P4 | Request or initiate an execution through a serving/integration boundary. |
| P4-INT-05 | Agent / Tool Participation | Agent / Tool participant < - > P4 | Coordinate Agent participation, Tool invocation and results. |
| P4-INT-06 | Authorization / Decision Control Request | P4 -> P6 | Request point-of-use authorization or decision/action control where applicable. |
| P4-INT-07 | Authorization / Action Control Outcome | P6 -> P4 | Return approval, rejection, conditions, human decision or action command. |
| P4-INT-08 | Environment Observation | Environment -> P4 | Provide sensed/observed state or event information. |
| P4-INT-09 | Authorized Environment Interaction | P4 -> Environment | Technically enact authorized interaction/Actuation. |
| P4-INT-10 | Runtime Evidence / Outcome | P4 -> P7 | Provide execution trace, primitive involvement, context binding, effects, failures and relevant telemetry. |
| P4-INT-11 | Assurance / Currentness / Safety Feedback | P7 -> P4 | Provide findings or status that constrain, suspend or alter future/ongoing execution where applicable. |
These interactions may be synchronous, asynchronous, local, embedded, event-driven, human-mediated or distributed.
GEAIRA-M07-R43 - Protocol neutrality: P4 interactions SHALL remain technology/protocol neutral in the generic reference architecture.
GEAIRA-M07-R44 - No synchronous-central assumption: GEAIRA SHALL NOT require every P4 control or dependency interaction to be a synchronous call to a centralized service.
22.13 P4 control points
M07 defines three logical P4 control-point classes.
CP-P4-01 - Execution Admission & Binding Control Point
Verifies that the execution has the required governing envelope, context and eligible primitive realization before material invocation begins.
CP-P4-02 - Runtime Participation & Side-Effect Control Point
Applies runtime constraints to Agent participation, Tool use, orchestration steps and side-effecting operations and exposes applicable consequential operations to P6.
CP-P4-03 - Runtime Continuation / Termination Control Point
Supports interruption, suspension, degradation, cancellation or termination based on failure, revocation, assurance/safety feedback or violated runtime conditions.
Control points may be embedded, distributed or human-mediated and do not imply standalone services.
GEAIRA-M07-R45 - Fail-safe control: Where a material runtime precondition or control outcome is indeterminate, P4 SHALL NOT silently default to unrestricted continuation.
GEAIRA-M07-R46 - Revocation propagation: Material revocation or currentness feedback affecting an ongoing or future execution SHALL be propagated with sufficient timeliness for the applicable risk/safety profile.
22.14 Failure and degradation model
M07 defines twelve generic P4 failure/degradation classes:
- execution admission failure;
- context binding failure;
- primitive binding / invocation failure;
- execution-mode realization mismatch;
- Agent participation failure;
- Tool invocation / side-effect uncertainty;
- orchestration deadlock / step failure;
- Environment observation/sensing failure;
- Authorization rejection/timeout/indeterminacy;
- Actuation / Environment interaction failure;
- control-loop instability/interruption;
- runtime evidence/correlation failure.
A concrete profile may add domain-specific failure classes.
GEAIRA-M07-R47 - Failure visibility: Material P4 failure or degradation SHALL be observable and SHALL NOT be silently converted into a successful execution outcome.
GEAIRA-M07-R48 - Side-effect uncertainty: Where the state of a consequential Tool or Environment side effect is uncertain, P4 SHALL preserve that uncertainty rather than assuming success or safe rollback.
GEAIRA-M07-R49 - Correlation integrity: Runtime evidence SHALL preserve sufficient correlation to distinguish concurrent or repeated execution occurrences where that distinction is material.
22.15 Cross-cutting overlays applied to P4
All eight M03 cross-cutting overlays apply to P4.
| Overlay | P4 application |
|---|---|
| X1 Security, Privacy & Trust Boundaries | execution isolation, input/output protection, Tool/Environment trust boundaries, secret handling, side-effect protection |
| X2 Identity, Authority & Policy Enforcement | Agent identity, delegated authority context, point-of-use authorization integration, Tool/operation constraints |
| X3 Evidence, Provenance & Traceability | execution identity, context/primitive bindings, Tool/Agent participation, effects, side-effect evidence |
| X4 Observability, Telemetry & Auditability | execution state, latency, failures, retries, orchestration steps, resource/runtime signals |
| X5 Resilience, Safety & Failure Management | admission fail-safe behavior, cancellation, circuit breaking, degraded mode, control-loop safety |
| X6 Lifecycle, Version, Currentness & Change | primitive/context/currentness feedback, runtime version identity, interruption after revocation/change |
| X7 Interoperability, Contracts & Semantic Compatibility | invocation contracts, Tool contracts, Agent interaction boundaries, Environment adapters, effect/result contracts |
| X8 Governance, Human Accountability & Compliance | human oversight hooks, decision/action escalation, accountability evidence, policy-constrained execution |
GEAIRA-M07-R50 - Overlay non-localization: P4 SHALL apply X1-X8 where material but SHALL NOT treat any overlay as wholly owned by one P4 ABB.
22.16 Canonical P4 reference view
The P4 logical reference view is:
P1 Governing Envelope
|
v
P2 AI Context -----> +---------------------------+ <----- P3 Primitive Realization
| P4-ABB-01 |
P5 Invocation -----> | Execution Coordination |
| Boundary |
+-------------+-------------+
|
+-------------+-------------+
| |
v v
+----------------------+ +---------------------------+
| P4-ABB-02 | | P4-ABB-03 |
| Agent & Tool | | Environment & Control |
| Participation | | Interaction |
+----------+-----------+ +-------------+-------------+
| |
+--------------+---------------+
|
consequential
decision/action requests
|
v
P6
Authorization / Oversight
|
approved action/control
|
+-------> P4 / Environment
P4 runtime evidence / effects / failures -------------------------> P7
P7 assurance/currentness/safety feedback ------------------------> P4
This is a logical responsibility view, not a deployment topology.
22.17 Representative P4 scenarios
Scenario A - Stateless model inference
P5 invokes an execution. P2 provides bounded context and P3 provides an eligible learned-model primitive. P4 establishes an AI Execution, records the realized Execution Mode, invokes the primitive, records exercised capability/effect evidence and returns the result through P5.
No Agent ABB or Environment ABB is required simply because AI was executed.
Scenario B - Agent uses a read-only Tool
An Agent participates in an AI Execution and invokes a Tool that retrieves account-reference information without consequential side effect. P4 coordinates Tool use and records the result. Tool use does not change the Tool resource's intrinsic identity and does not establish unrestricted authority.
Scenario C - Agent proposes consequential payment action
An Agent forms a proposed action through an AI Execution. P4 may participate in Decision Formation, but the consequential payment action is sent to P6 for applicable Authorization/human oversight. Only an authorized outcome may proceed to the action-enactment path.
Scenario D - Physical AI control loop
An AI Execution receives Environment observations, exercises applicable capabilities, participates in a Control Loop and proposes an actuator command. P6 supplies the applicable Authorization/control outcome or a governed pre-authorized envelope applies. P4 then technically enacts the allowed Actuation and observes subsequent Environment state.
Control-loop participation does not by itself establish autonomous authority or ClosedLoopAutonomous Embodiment.
Scenario E - Primitive revoked during execution
P7/P3 reports that a primitive version is no longer eligible while a long-running execution is active. P4 applies the applicable continuation policy: terminate, suspend, switch only through a governed substitution path, or complete under an explicitly permitted exception. The event remains traceable.
Scenario F - Tool side effect times out ambiguously
P4 invokes a side-effecting Tool and times out before learning whether the operation committed. P4 records indeterminate side-effect state and does not blindly retry if duplication could be harmful. Recovery/escalation follows the applicable contract/control policy.
22.18 P4 anti-patterns
M07 defines the following P4-specific anti-patterns:
- AI-runtime-equals-P4 - treating one runtime platform as the reference architecture.
- Agent-equals-execution - treating Agent identity and AI Execution occurrence as the same thing.
- Agent-equals-primitive - collapsing Agent identity into model/primitive identity.
- Autonomy-equals-authority - assuming self-directed behavior grants permission.
- Tool-availability-equals-permission - assuming an invocable Tool is authorized for any use.
- Tool-registry-equals-governance - treating Tool registration as sufficient authority/safety control.
- Orchestrator-equals-agent - assuming workflow/orchestration software is necessarily an Agent.
- Context-equals-environment - collapsing runtime context with the external Environment.
- Sensing-equals-perception - treating any sensor input as Perception Intelligence Capability.
- Actuation-equals-action-authority - treating technical control ability as permission to act.
- Control-loop-equals-autonomy - inferring autonomous authority from recurring feedback control.
- Decision-effect-equals-decision-authority - allowing decision-like model output to bypass P6.
- Central-control-everything - forcing all runtime controls through a remote central service regardless of latency/safety constraints.
- Silent-side-effect-retry - retrying uncertain consequential Tool/Environment actions without idempotency or reconciliation safeguards.
- Execution-without-identity - losing occurrence correlation so evidence cannot be reconstructed.
- Historical-runtime-rewrite - replacing actual historical primitive/context/Agent/Tool involvement with later current-state declarations.
22.19 M07 frozen decisions
M07 freezes the following P4 architecture decisions for GEAIRA 0.1:
P4 is the occurrence-level AI runtime plane.
AI Execution != runtime platform / Product / Agent / ABB.
P3 provision != actual primitive invocation.
P2 context provision != proof of material context use.
Agent participation != authority / accountability / Product identity.
Tool use != entitlement / Authorization / ownership.
Environment != AI Context.
Sensing != Perception capability.
Actuation != authority to act.
Control Loop != autonomy / Control capability / ClosedLoopAutonomous Embodiment.
Decision-like effect != final decision authority.
Delegated Authority != point-of-use Authorization.
P4 may technically enact authorized action; P6 controls applicable Authorization/decision/action permission.
Runtime evidence != independent assurance.
Side-effect uncertainty must remain explicit.
Historical occurrence identity/bindings survive later supersession or currentness changes.
The following P4 ABBs are canonical:
P4-ABB-01 Execution Coordination Boundary
P4-ABB-02 Agent & Tool Participation Boundary
P4-ABB-03 Environment & Control Interaction Boundary
The following remain realization choices rather than generic canonical ABBs:
AI Runtime Platform
Inference Server / LLM Runtime
Agent Framework / Agent Runtime
Workflow Engine / Task Queue
Tool Registry / Plugin Marketplace
MCP Gateway / MCP Server
Function-Calling Service
Session / Execution State Store
Robot Controller / ROS Runtime
IoT Gateway / Sensor Hub / Actuator Gateway
Digital Twin Platform
SCADA / PLC Platform
M07 does not freeze:
- a runtime technology stack;
- an Agent framework;
- an Agent communication protocol;
- Tool protocol or marketplace;
- model/inference serving protocol;
- workflow/orchestration engine;
- robotics middleware;
- sensor/actuator protocol;
- hard-real-time control implementation;
- P5 Product serving architecture;
- final P6 decision/action architecture;
- final P7 assurance/observability architecture;
- final M11 inter-plane interaction contracts.
GEAIRA-M07-R51 - Frozen ABB restraint: M07 SHALL freeze only the three P4 logical ABBs above and SHALL NOT promote runtime platforms, agent frameworks, orchestration engines, Tool registries, robotics middleware or Environment gateways to canonical ABB status.
GEAIRA-M07-R52 - Historical occurrence preservation: Later lifecycle/currentness changes SHALL NOT rewrite which context, primitive version, Agent, Tool, Execution Mode or material Environment interaction actually participated in a historical AI Execution.
GEAIRA-M07-R53 - Runtime/Product isolation: Occurrence-level execution evidence SHALL NOT automatically populate or modify AIPS Product-version capability, execution-mode, effect, Agency/Autonomy or Embodiment declarations.
GEAIRA-M07-R54 - Execution-effect non-inference: P4 SHALL NOT infer primitive type, Agent type, Product type or architecture topology solely from an observed Intelligence Effect.
GEAIRA-M07-R55 - Downstream preservation: Later GEAIRA slices SHALL preserve the M07 runtime/authority, Agent/primitive, context/environment, sensing/perception, actuation/authorization and effect/decision distinctions unless the 0.1 programme explicitly records and reconciles a justified architecture change without redefining GEAIRM/AIPS/Agent semantics.
GEAIRA-M07-R56 - Runtime diversity continuity: The P4 architecture SHALL remain valid for non-agentic, agentic, software-only, physical, mixed-environment, synchronous, asynchronous, batch, streaming, interactive and long-running AI executions without structural redesign solely because the runtime style differs.
23. M07 Completion State
GEAIRA-M07 - COMPLETE.
M07 establishes:
- one canonical P4 purpose and occurrence-level architecture boundary;
- fourteen canonical P4 Architectural Responsibilities;
- three canonical logical P4 ABBs;
- twenty deliberately non-admitted implementation-style ABB candidates;
- eleven technology-neutral P4 architectural interactions;
- three logical P4 control-point classes;
- twelve generic P4 failure/degradation classes;
- explicit P4 application of X1-X8 cross-cutting overlays;
- explicit P1/P2/P3/P5/P6/P7 responsibility boundaries;
- six representative P4 scenarios;
- sixteen P4 anti-patterns;
- 56 normative M07 rules.
M07 introduces no change to GEAIRM, AIPS or Agent Ontology semantics, does not prescribe an AI runtime platform, Agent framework, Tool registry, orchestration engine, robotics middleware, Environment gateway or centralized control topology.
24. GEAIRA-M08 - P5: AI Serving, Product & Integration
24.1 Purpose
GEAIRA-M08 defines the canonical architecture responsibilities of P5 - AI Serving, Product & Integration.
P5 answers:
How are governed AI capabilities and executions exposed, contracted, integrated and consumed across enterprise boundaries without confusing serving with Product identity or AI Product qualification?
P5 sits between consumer-facing/integration concerns and the occurrence-level execution responsibilities of P4. It also provides the architecture boundary through which BPS/AIPS Product semantics, where applicable, become serving and integration concerns without being redefined by GEAIRA.
M08 does not assume that every served AI capability is a Product, every Product is an AI Product, every AI Product is exposed through an API, or every consumer interaction requires a standalone gateway/service.
GEAIRA-M08-R01 - Serving boundary: P5 SHALL own generic architectural responsibilities for exposing, contracting, integrating and governing consumption of enterprise AI capabilities/executions while preserving P1-P4, P6-P7 and external Product-authority boundaries.
GEAIRA-M08-R02 - Serving/Product distinction: The existence of a serving or integration boundary SHALL NOT by itself establish Productness, AI Product qualification, Product Family, Product Kind or Product-version classification.
24.2 Source semantic and architecture baseline
M08 consumes the frozen GEAIRM/AIPS/BPS distinctions and the GEAIRA M01-M07 architecture baseline.
Relevant semantic/architecture inputs include:
- BPS Product / Product Version / Product Intrinsic Nature semantics where Productness applies;
- AIPS AI Product and AI Product Version semantics where AI Product qualification applies;
- GEAIRM AI Execution and relevant Product/dependency/composition relationships;
- GEAIRM hard distinctions between Productness, AI Product qualification, Model Product, AI-enabled Product and AI mechanisms;
- P1 governing intent/authority envelopes;
- P2 context provision contracts;
- P3 primitive realization/provision contracts;
- P4 AI Execution invocation and result/effect boundaries;
- P6 decision/action control where consumption triggers consequential action;
- P7 evidence, assurance, observability and lifecycle feedback;
- X1-X8 cross-cutting overlays.
M08 preserves:
serving boundary != Product
Product != AI Product
AI Product != AI Primitive
AI Product != AI Execution
Model Product != AI Product automatically
AI-enabled Product != AI Product automatically
API endpoint != Product
integration adapter != Product
consumer request != Authorization
GEAIRA-M08-R03 - External Product authority: P5 SHALL consume BPS/AIPS Product and AI Product semantics where applicable and SHALL NOT create substitute Product qualification rules.
GEAIRA-M08-R04 - Non-Product applicability: P5 serving/integration responsibilities MAY apply to non-Product enterprise AI capabilities where governed consumption or integration is required.
24.3 P5 architecture boundary
P5 owns the architectural concerns required to make governed AI capability/execution available to consumers and adjacent systems through stable, explicit and lifecycle-aware contracts.
P5 principally owns:
- consumption boundary definition;
- serving contract exposure;
- request/response or equivalent interaction mediation;
- Product/AI Product realization association where externally established;
- contract and compatibility management;
- consumer/integration boundary mediation;
- provider/dependency mediation where external or shared AI capabilities are integrated;
- invocation routing into P4;
- delivery of execution results/effects back to consumers;
- serving lifecycle, version and deprecation management;
- integration evidence and traceability.
P5 does not principally own:
- Productness or AI Product qualification - BPS/AIPS;
- intended purpose and governing authority - P1;
- context construction - P2;
- capability-to-primitive eligibility - P3;
- occurrence-level AI Execution - P4;
- consequential decision/action Authorization - P6;
- independent evaluation/assurance - P7.
GEAIRA-M08-R05 - Execution separation: P5 SHALL NOT treat a serving request, endpoint call or integration event as equivalent to the resulting P4 AI Execution occurrence.
GEAIRA-M08-R06 - Product qualification isolation: P5 SHALL NOT infer AI Product qualification solely because an AI capability is packaged, exposed, monetized, versioned or consumed through a Product-like interface.
24.4 Canonical P5 Architectural Responsibilities
M08 establishes fourteen canonical P5 Architectural Responsibilities.
| ID | Responsibility | Description | Applicability |
|---|---|---|---|
| P5-RESP-01 | Consumption Boundary Definition | Define the governed boundary through which a consumer/system accesses an AI capability or Product realization. | Baseline-mandatory where P5 applies |
| P5-RESP-02 | Serving Contract Definition | Define technology-neutral request/input, response/output, interaction, error and semantic expectations. | Baseline-mandatory |
| P5-RESP-03 | Product/Offering Association | Associate the serving boundary with externally established Product/Product Version/AI Product identity where applicable. | Conditional-mandatory |
| P5-RESP-04 | Consumer Compatibility | Determine whether a consumer/integration contract is compatible with the exposed serving contract/version. | Baseline-mandatory |
| P5-RESP-05 | Consumption Mode Realization | Support the applicable interaction mode without making one protocol mandatory. | Conditional-mandatory |
| P5-RESP-06 | Invocation Admission & Validation | Validate that a consumption request satisfies the applicable contract and governing preconditions before P4 invocation. | Baseline-mandatory |
| P5-RESP-07 | Invocation Routing | Route/mediate an admitted serving request toward the appropriate P4 execution path without taking over P4 execution ownership. | Baseline-mandatory |
| P5-RESP-08 | Result / Effect Delivery | Deliver eligible P4 outputs/effects to the appropriate consumer boundary under the serving contract. | Baseline-mandatory |
| P5-RESP-09 | Integration / Provider Mediation | Mediate external/shared provider, platform, Product or service dependencies where needed. | Conditional-mandatory |
| P5-RESP-10 | Contract Version & Currentness | Maintain serving-contract version, compatibility, currentness, supersession and deprecation state. | Baseline-mandatory |
| P5-RESP-11 | Product/Serving Configuration Projection | Project externally governed Product-version or offering-specific configuration into serving behavior without redefining source semantics. | Conditional-mandatory |
| P5-RESP-12 | Consumption Evidence & Traceability | Preserve consumer, contract, Product/version, invocation and integration trace needed for audit/reconstruction. | Baseline-mandatory |
| P5-RESP-13 | Usage / Outcome Feedback | Provide relevant serving/integration usage and outcome evidence to P7 and applicable Product/governance authorities. | Conditional-mandatory |
| P5-RESP-14 | Change, Deprecation & Migration Coordination | Coordinate consumer-impacting contract/provider/Product-serving changes and migration windows. | Conditional-mandatory |
GEAIRA-M08-R07 - Contract before protocol: P5 SHALL define the semantic/architectural serving contract independently of a specific wire protocol, SDK, gateway or vendor interface.
GEAIRA-M08-R08 - Consumer traceability: Material P5 consumption SHALL preserve sufficient correlation to identify the applicable consumer, serving contract/version and resulting P4 execution where such traceability is required.
GEAIRA-M08-R09 - Product association provenance: Where P5 exposes a Product/AI Product realization, Product and Product Version identity SHALL originate from their authoritative Product semantics rather than being minted ad hoc by the serving layer.
GEAIRA-M08-R10 - Mode neutrality: P5 SHALL remain valid for synchronous, asynchronous, event-driven, streaming, batch, interactive, agent-to-agent and human-mediated consumption without making any one interaction mode universally mandatory.
24.5 Canonical P5 ABB admission
Applying the M02 ABB admission criteria results in three canonical logical P5 ABBs.
P5-ABB-01 - AI Serving & Consumption Boundary
Purpose:
Establish the governed consumer-facing boundary through which an enterprise AI capability or Product realization is invoked/consumed and results are returned.
Primary responsibilities:
- P5-RESP-01 Consumption Boundary Definition;
- P5-RESP-02 Serving Contract Definition;
- P5-RESP-04 Consumer Compatibility;
- P5-RESP-05 Consumption Mode Realization;
- P5-RESP-06 Invocation Admission & Validation;
- P5-RESP-07 Invocation Routing;
- P5-RESP-08 Result / Effect Delivery;
- P5-RESP-12 Consumption Evidence & Traceability.
This ABB does not imply an API Gateway, model gateway, service mesh, SDK, UI or standalone network service.
P5-ABB-02 - Product Realization & Contract Boundary
Purpose:
Bind externally established Product/AI Product identity and Product-version semantics to the serving/integration contract without deciding Productness or AI Product qualification.
Primary responsibilities:
- P5-RESP-03 Product/Offering Association;
- P5-RESP-10 Contract Version & Currentness;
- P5-RESP-11 Product/Serving Configuration Projection;
- P5-RESP-14 Change, Deprecation & Migration Coordination.
This ABB is conditionally applicable. A non-Product AI serving boundary MAY legitimately use P5-ABB-01 and/or P5-ABB-03 without P5-ABB-02.
P5-ABB-03 - Integration & Provider Mediation Boundary
Purpose:
Mediate external/shared providers, services, platforms or adjacent Products while preserving contract, identity, compatibility, authority and evidence boundaries.
Primary responsibilities:
- P5-RESP-04 Consumer Compatibility where integration-facing;
- P5-RESP-09 Integration / Provider Mediation;
- P5-RESP-10 Contract Version & Currentness;
- P5-RESP-12 Consumption Evidence & Traceability;
- P5-RESP-13 Usage / Outcome Feedback;
- P5-RESP-14 Change, Deprecation & Migration Coordination.
This ABB does not imply a vendor connector service, enterprise service bus, API gateway, service mesh or integration platform.
GEAIRA-M08-R11 - ABB logical status: P5-ABB-01 through P5-ABB-03 are logical architecture responsibility boundaries and SHALL NOT be interpreted as mandatory standalone deployable services.
GEAIRA-M08-R12 - Conditional Product ABB: P5-ABB-02 SHALL be applied only where authoritative Product/Offering semantics exist; GEAIRA SHALL NOT manufacture Product identity merely to make the P5 architecture uniform.
GEAIRA-M08-R13 - Co-location independence: Co-locating P5 ABB responsibilities in one component SHALL NOT erase their independently traceable architectural responsibilities.
24.6 Deliberately non-admitted P5 ABB candidates
M08 deliberately does not promote the following implementation choices into generic canonical ABBs:
API Gateway
AI Gateway / Model Gateway
LLM Gateway
Service Mesh
Enterprise Service Bus
Message Broker / Event Bus
SDK / Client Library
Backend-for-Frontend
Web / Mobile UI
Developer Portal
API Marketplace
AI Marketplace
Integration Platform / iPaaS
Vendor Connector Service
Provider Adapter Service
Protocol Translator
Load Balancer
Ingress Controller
Service Registry / Discovery Platform
Contract Registry / API Catalog
Any of these MAY realize one or more P5 responsibilities in a concrete architecture.
GEAIRA-M08-R14 - Gateway restraint: The existence of a serving responsibility SHALL NOT by itself justify a canonical API/AI/model gateway ABB.
GEAIRA-M08-R15 - Registry restraint: Contract discovery/currentness requirements SHALL NOT automatically require a standalone registry or catalog ABB.
24.7 Serving contract model
A P5 serving contract is an architectural interaction contract, not merely an API specification.
At minimum, where applicable, the contract SHOULD identify:
servingBoundaryId
contractId
contractVersion
consumerClass / consumer identity expectations
Product / Product Version identity if applicable
capability / outcome semantics exposed
request/input expectations
response/output/effect expectations
interaction mode
compatibility rules
error / indeterminate outcomes
timeout / retry / idempotency expectations where relevant
applicable governing envelope references
P4 invocation correlation expectations
evidence / observability expectations
deprecation / supersession state
GEAIRA-M08-R16 - Contract semantic stability: A protocol-compatible change that materially changes the meaning, capability, effect or governing expectation of a serving contract SHALL NOT be treated as semantically equivalent merely because the wire schema remains compatible.
GEAIRA-M08-R17 - Explicit indeterminacy: Where a serving/integration outcome cannot determine whether a consequential downstream side effect occurred, P5 SHALL preserve an explicit indeterminate/reconciliation state rather than blindly retrying.
24.8 Consumption modes
P5 supports multiple consumption modes without defining them as Product kinds.
Representative modes include:
request / response
asynchronous job
publish / subscribe
stream
batch
interactive session
human-facing application interaction
agent-to-agent interaction
embedded invocation
file / artifact exchange
A Product or capability MAY expose more than one mode.
GEAIRA-M08-R18 - Mode/Product distinction: Consumption mode SHALL NOT be used as a substitute for Product classification, Execution Mode or Intelligence Capability classification.
GEAIRA-M08-R19 - Multi-mode consistency: Where multiple serving modes expose the same governed capability/Product realization, material semantic differences between those modes SHALL be explicit rather than assumed equivalent.
24.9 Product realization boundary
P5 is where externally governed Product semantics may become architecturally visible to consumers, but P5 does not own the Product ontology.
The generic flow is:
BPS / AIPS Product authority
|
| Product / Product Version / AI Product semantics
v
+-----------------------------------+
| P5-ABB-02 |
| Product Realization & Contract |
| Boundary |
+----------------+------------------+
|
serving realization
v
+-----------------------------------+
| P5-ABB-01 |
| AI Serving & Consumption Boundary |
+----------------+------------------+
|
v
Consumer
For a non-Product internal AI capability:
P5-ABB-01 serving boundary
|
v
internal governed consumer
is valid without inventing Product semantics.
GEAIRA-M08-R20 - No packaging inference: Packaging an AI model/capability behind a reusable endpoint SHALL NOT by itself establish BPS Productness or AIPS AI Product qualification.
GEAIRA-M08-R21 - Product-version isolation: P4 occurrence-level execution evidence SHALL NOT automatically mutate P5/AIPS Product Version declarations; governed Product changes require the applicable Product lifecycle process.
24.10 External/provider integration
P5 may integrate externally provided AI capabilities, Products, models, services or adjacent enterprise capabilities.
Provider mediation responsibilities may include:
- provider identity and contract association;
- compatibility and version mapping;
- request/response transformation where semantically safe;
- provider currentness/deprecation handling;
- governing constraint projection;
- traceability across enterprise/provider boundaries;
- fault/timeout/retry reconciliation;
- provider substitution under governed compatibility rules.
Provider substitution is not automatically semantics-preserving.
GEAIRA-M08-R22 - Provider substitution discipline: P5 SHALL NOT substitute one provider/integration target for another merely because interfaces are syntactically compatible; material capability, effect, risk, governance and assurance differences SHALL be respected.
GEAIRA-M08-R23 - External evidence preservation: External-provider invocation evidence SHALL preserve sufficient provider/version/contract correlation for applicable audit, assurance and incident reconstruction.
GEAIRA-M08-R24 - Boundary control continuity: External integration SHALL NOT bypass P1 governing envelopes, P6 Authorization/action controls or P7 evidence/assurance obligations merely because the provider is outside the enterprise boundary.
24.11 P5 relationship to P1-P4, P6 and P7
P1 -> P5
P1 supplies governing constraints relevant to serving availability, permitted consumers, prohibited use, Product/Offering scope and applicable authority.
P5 does not define those governing semantics.
P2 < - > P5
P5 MAY collect or mediate consumer-provided contextual material, but P2 owns governed context construction/qualification where AI Context responsibilities apply.
A request payload is not automatically the canonical AI Context.
P3 < - > P5
P5 MAY expose provider/deployment/serving dependencies relevant to primitive realization. P3 owns capability-to-primitive eligibility and compatibility.
A model endpoint exposed by P5 does not make P5 the owner of primitive semantics.
P4 < - > P5
P5 accepts/mediates a serving request and invokes/routes to P4. P4 establishes and enacts the actual AI Execution occurrence.
P4 returns occurrence output/effects/evidence to P5 for contract-governed delivery.
P6 < - > P5
Where consumer-facing activity itself requires point-of-use Authorization or consequential decision/action control, P5 SHALL preserve the P6 boundary rather than treating successful request admission as sufficient authority.
P7 < - > P5
P5 provides serving/integration evidence and receives assurance/currentness/deprecation findings that may alter exposure eligibility or migration state.
GEAIRA-M08-R25 - Request/context distinction: P5 SHALL NOT equate a consumer request payload with a fully governed P2 AI Context unless the applicable architecture explicitly establishes that equivalence.
GEAIRA-M08-R26 - Admission/Authorization distinction: P5 contract admission SHALL NOT be treated as equivalent to P6 Authorization for consequential decisions/actions.
GEAIRA-M08-R27 - Serving/execution trace: P5 and P4 SHALL preserve a traceable distinction between serving request identity and AI Execution occurrence identity.
24.12 Canonical P5 architectural interactions
M08 establishes ten technology-neutral P5 interactions.
| ID | Interaction | Direction | Architectural intent |
|---|---|---|---|
| P5-INT-01 | Governing Serving Constraints | P1 -> P5 | Supply purpose/authority/policy constraints relevant to exposure and consumption. |
| P5-INT-02 | Consumer Consumption Request | Consumer/peer -> P5 | Request use of a governed AI capability/Product realization. |
| P5-INT-03 | Product / Offering Association | Product authority -> P5 | Supply authoritative Product/Product Version/AI Product identity where applicable. |
| P5-INT-04 | Execution Invocation / Serving Request | P5 -> P4 | Initiate/request an AI Execution through the serving boundary. |
| P5-INT-05 | Execution Result / Effect Return | P4 -> P5 | Return applicable execution outputs/effects/status for delivery. |
| P5-INT-06 | Provider / Integration Invocation | P5 -> external/peer provider | Invoke or exchange with an integrated provider/dependency. |
| P5-INT-07 | Provider / Integration Response | external/peer provider -> P5 | Return provider result/status/evidence. |
| P5-INT-08 | Serving / Integration Evidence | P5 -> P7 | Supply trace, usage, contract and integration evidence. |
| P5-INT-09 | Assurance / Currentness Feedback | P7/authority -> P5 | Alter exposure eligibility, compatibility or deprecation state. |
| P5-INT-10 | Contract Change / Deprecation Notice | P5 -> consumers/peers | Signal version, migration, supersession or retirement changes. |
GEAIRA-M08-R28 - Interaction neutrality: P5 interactions SHALL remain technology/protocol neutral at generic GEAIRA level.
GEAIRA-M08-R29 - Contract-bound invocation: P5 SHALL correlate an execution invocation with the serving contract/version that admitted it where material to reconstruction, compatibility or governance.
24.13 P5 control points
M08 establishes three logical control-point classes.
CP-P5-01 - Consumption Admission Control
Determines whether a consumption request satisfies the applicable serving contract, compatibility and governing preconditions.
It does not replace P6 Authorization where consequential authority is required.
CP-P5-02 - Integration Boundary Control
Controls external/peer provider selection, compatibility, transformation, currentness and boundary policy obligations.
CP-P5-03 - Contract Lifecycle Control
Controls contract publication/currentness, supersession, deprecation, compatibility windows and migration state.
These control points MAY be embedded, distributed, platform-provided, federated or human-governed.
GEAIRA-M08-R30 - Control-point restraint: P5 control points SHALL NOT automatically be promoted to standalone gateways, registries or lifecycle services.
24.14 Failure and degradation model
M08 defines twelve generic P5 failure/degradation classes.
| Class | Example condition | Required architecture behavior |
|---|---|---|
| Contract violation | Request does not satisfy declared contract | Reject or explicitly degrade; do not silently reinterpret. |
| Consumer incompatibility | Consumer depends on incompatible contract/version | Fail explicitly or use governed compatibility path. |
| Serving unavailable | Consumption boundary unavailable | Surface unavailable/degraded state. |
| P4 invocation failure | Serving accepted but execution cannot start | Distinguish admission from execution failure. |
| Indeterminate execution outcome | Timeout leaves execution/result state uncertain | Preserve indeterminate/reconciliation state. |
| Provider unavailable | External/peer dependency unavailable | Apply governed fallback/degradation if allowed. |
| Provider incompatibility | Provider version/semantics no longer compatible | Prevent silent substitution. |
| Contract superseded | Consumer invokes retired/deprecated contract | Enforce declared migration/deprecation policy. |
| Product/serving mismatch | Serving configuration conflicts with authoritative Product/version | Block or escalate; do not mutate Product semantics locally. |
| Governing restriction | P1/P7 state makes exposure ineligible | Stop/restrict serving according to applicable policy. |
| Response delivery failure | Execution completed but result cannot be delivered | Preserve execution completion separately from delivery status. |
| Duplicate/retry risk | Retry may duplicate consequential downstream activity | Apply idempotency/reconciliation semantics where required. |
GEAIRA-M08-R31 - Admission/execution/result separation: P5 SHALL represent serving admission, P4 execution status and consumer delivery status as distinguishable concerns where failure semantics matter.
GEAIRA-M08-R32 - No silent semantic fallback: P5 SHALL NOT silently fallback to a materially different Product, provider, capability or contract when that changes governed semantics or assurance expectations.
GEAIRA-M08-R33 - Delivery/history preservation: Failure to deliver a completed execution result SHALL NOT rewrite the historical fact that the P4 execution occurred.
24.15 Cross-cutting overlays applied to P5
The M03 overlays apply to P5 as follows.
| Overlay | P5 application |
|---|---|
| X1 Security, Privacy & Trust Boundaries | Protect consumption boundaries, payloads, provider crossings and tenant/data separation. |
| X2 Identity, Authority & Policy Enforcement | Bind consumer/provider identity and governing policy without conflating request admission with P6 Authorization. |
| X3 Evidence, Provenance & Traceability | Trace consumer, contract, Product/version, provider and P4 execution correlation. |
| X4 Observability, Telemetry & Auditability | Observe serving availability, latency, errors, provider behavior and contract use. |
| X5 Resilience, Safety & Failure Management | Govern fallback, retries, idempotency, provider outages and degraded modes. |
| X6 Lifecycle, Version, Currentness & Change | Manage serving contract/version, provider version, deprecation and migration. |
| X7 Interoperability, Contracts & Semantic Compatibility | Core P5 concern: maintain explicit semantic/contract compatibility across consumers and providers. |
| X8 Governance, Human Accountability & Compliance | Preserve accountable ownership of exposure, Product realization and migration decisions. |
GEAIRA-M08-R34 - Cross-cutting non-centralization: Applying X1-X8 to P5 SHALL NOT imply that all serving/integration controls must be centralized in one enterprise gateway or platform.
24.16 Canonical P5 reference view
The canonical logical P5 view is:
BPS / AIPS PRODUCT AUTHORITIES
|
Product / Version semantics
|
v
+-----------------------------+
| P5-ABB-02 |
| Product Realization & |
| Contract Boundary |
+--------------+--------------+
|
v
CONSUMER -----> +-----------------------------+ -----> P4 AI EXECUTION
| P5-ABB-01 | invocation
<---- RESULT ---| AI Serving & Consumption | <----- P4 result/effect
| Boundary |
+--------------+--------------+
|
v
+-----------------------------+
| P5-ABB-03 |
| Integration & Provider |
| Mediation Boundary |
+--------------+--------------+
|
v
EXTERNAL / PEER PROVIDERS
P1 governs exposure constraints.
P6 retains applicable point-of-use Authorization/action control.
P7 receives serving/integration evidence and returns assurance/currentness feedback.
This is a logical responsibility view, not a required deployment topology.
GEAIRA-M08-R35 - Reference-view restraint: The P5 reference view SHALL NOT be interpreted as requiring separate network hops or deployable services for each ABB.
24.17 Representative P5 scenarios
Scenario A - Internal non-Product AI capability
An internal team exposes a governed summarization capability to another internal system. No BPS Product has been established.
P5-ABB-01 defines the serving contract and invokes P4. P5-ABB-02 is not applied merely for architectural symmetry.
Scenario B - Qualified AI Product Version
A Product Version already qualified by AIPS is exposed to consumers.
P5-ABB-02 binds authoritative Product/Version semantics to the serving contract. P5 does not perform AI Product qualification.
Scenario C - Multi-mode serving
The same governed capability is available through synchronous request/response and asynchronous batch interfaces.
P5 records whether the two modes are semantically equivalent or materially different; protocol differences alone do not establish Product differences.
Scenario D - External model/provider integration
P5 integrates an external AI provider through P5-ABB-03. P3 still determines primitive eligibility where the provider realizes an AI Primitive; P4 owns the actual execution occurrence boundary; externality does not bypass P1/P6/P7 controls.
Scenario E - Contract deprecation
Contract v1 is superseded by v2. Existing consumers receive an explicit migration window. New executions correlate to the contract version actually used; historical v1 consumption records remain intact after v1 retirement.
Scenario F - Timeout with uncertain downstream state
P5 invokes P4 and times out. The architecture cannot yet determine whether P4 completed or whether a consequential action subsequently occurred.
P5 records an indeterminate/reconciliation state rather than issuing an unsafe blind retry.
24.18 P5 anti-patterns
M08 prohibits or warns against the following generic anti-patterns:
- API-equals-Product - treating an endpoint as proof of Productness.
- served-model-equals-AI-Product - treating a model endpoint as automatic AI Product qualification.
- gateway-everything - centralizing every P5 responsibility into one enterprise AI/API gateway.
- protocol-equals-contract - assuming an OpenAPI/schema/protocol description fully captures semantic serving expectations.
- request-equals-context - treating raw consumer request data as governed P2 AI Context automatically.
- admission-equals-authorization - treating a valid request as P6 permission for consequential action.
- provider-equals-primitive - assuming every external service maps one-to-one to an AIPS AI Primitive.
- syntactic-substitutability - swapping providers because interface schemas match.
- latest-equals-current - exposing newest contract/provider version without governed currentness determination.
- silent-contract-drift - changing capability/effect meaning under an unchanged contract identity.
- marketplace-equals-governance - assuming listing/discovery provides sufficient qualification or control.
- retry-everything - blindly retrying indeterminate consequential operations.
- serving-equals-execution - collapsing request identity with P4 occurrence identity.
- historical-contract-rewrite - rewriting historical consumption as if it used the latest contract/Product version.
- integration-bypass - letting external provider integration bypass P1, P6 or P7 obligations.
- Product-uniformity-forcing - inventing Product identity for internal/non-Product AI solely to fit the P5 model.
24.19 M08 frozen decisions
M08 freezes the following P5 architecture decisions for GEAIRA 0.1:
P5 is the serving, Product-realization and integration plane.
Serving boundary != Product.
Product != AI Product.
AI Product qualification remains AIPS authority.
Non-Product AI may legitimately use P5 serving/integration responsibilities.
Consumer request != AI Execution.
Consumer request != governed AI Context automatically.
Contract admission != consequential Authorization.
P4 owns occurrence-level execution.
P5 owns exposure, consumption contracts, integration and delivery boundaries.
Protocol compatibility != semantic compatibility.
Provider substitution != semantic equivalence.
Contract/version currentness must be explicit.
Historical serving request/contract/Product-version correlation survives later supersession.
The following P5 ABBs are canonical:
P5-ABB-01 AI Serving & Consumption Boundary
P5-ABB-02 Product Realization & Contract Boundary
P5-ABB-03 Integration & Provider Mediation Boundary
The following remain realization choices rather than generic canonical ABBs:
API / AI / Model / LLM Gateway
Service Mesh / ESB / Message Broker / Event Bus
SDK / Client Library / UI / BFF
Developer Portal / API Marketplace / AI Marketplace
Integration Platform / iPaaS
Vendor Connector / Provider Adapter
Protocol Translator / Load Balancer / Ingress Controller
Service Registry / Contract Registry / API Catalog
M08 does not freeze:
- a mandatory serving protocol;
- API style or schema language;
- message/event technology;
- service-mesh/gateway topology;
- consumer authentication technology;
- external provider vendor;
- Product distribution channel;
- commercial Product model;
- AI Product qualification criteria;
- P6 decision/action architecture;
- final P7 assurance/observability architecture;
- final M11 inter-plane interaction contracts.
GEAIRA-M08-R36 - Frozen ABB restraint: M08 SHALL freeze only the three P5 logical ABBs above and SHALL NOT promote gateways, brokers, service meshes, marketplaces, registries or vendor adapters to canonical ABB status.
GEAIRA-M08-R37 - Historical serving preservation: Later contract/Product/provider lifecycle changes SHALL NOT rewrite which serving contract/version, Product Version or provider boundary actually applied to a historical consumption request/execution.
GEAIRA-M08-R38 - Product authority continuity: Later GEAIRA slices SHALL preserve BPS/AIPS authority for Productness and AI Product qualification and SHALL NOT infer those statuses from serving architecture.
GEAIRA-M08-R39 - Downstream boundary preservation: Later slices SHALL preserve the P5/P2 context, P5/P3 primitive, P5/P4 execution, P5/P6 Authorization and P5/P7 evidence distinctions unless a reconciled 0.1 architecture change explicitly requires otherwise.
GEAIRA-M08-R40 - Serving diversity continuity: The P5 architecture SHALL remain valid across internal/external, Product/non-Product, synchronous/asynchronous, event/stream/batch, human/system/agent and embedded/federated consumption styles without structural redesign solely because the serving style differs.
25. M08 Completion State
GEAIRA-M08 - COMPLETE.
M08 establishes:
- one canonical P5 purpose and serving/Product/integration architecture boundary;
- fourteen canonical P5 Architectural Responsibilities;
- three canonical logical P5 ABBs;
- twenty deliberately non-admitted implementation-style ABB candidates;
- ten technology-neutral P5 architectural interactions;
- three logical P5 control-point classes;
- twelve generic P5 failure/degradation classes;
- explicit P5 application of X1-X8 cross-cutting overlays;
- explicit P1/P2/P3/P4/P6/P7 and BPS/AIPS Product-authority boundaries;
- six representative P5 scenarios;
- sixteen P5 anti-patterns;
- 40 normative M08 rules.
M08 introduces no change to GEAIRM, BPS or AIPS semantics, does not perform Product/AI Product qualification, and does not prescribe an API gateway, AI/model gateway, service mesh, message broker, marketplace, integration platform, vendor connector or serving protocol.
Next slice
GEAIRA-M09 - P6: Decision, Action & Human Oversight
M09 will define the architecture responsibilities for Decision Formation, point-of-use Authorization, consequential action control, human oversight, escalation, approval/override patterns and the boundary between intelligence effects/recommendations and governed enterprise decisions/actions.
26. GEAIRA-M09 - P6: Decision, Action & Human Oversight
26.1 Purpose of this slice
GEAIRA-M09 establishes the canonical architecture baseline for P6 - Decision, Action & Human Oversight.
P6 answers:
How should enterprise AI participate in consequential decision and action paths while preserving explicit decision authority, context-bounded Authorization, human oversight, escalation and safe action control?
P6 consumes the frozen GEAIRM distinctions among Recommendation, Plan, Decision Formation, AIPS Decision Intelligence Effect, Authorization, Delegated Authority, DigitalAction and PhysicalAction. It converts those semantics into technology-neutral architectural responsibilities without inventing a universal decision workflow, a generic legal-accountability ontology or a mandatory approval service.
The principal P6 boundary is:
P1
establishes governing intent, authority source,
accountability basis and policy envelope
P4
performs AI Execution and may produce recommendations,
plans, scores, Decision effects or proposed actions
P6
coordinates governed Decision Formation,
point-of-use Authorization,
human oversight and consequential action commitment
P4 / P5 / external system
technically enacts or delivers an authorized outcome where applicable
P7
evaluates, observes and assures the decision/action path
GEAIRA-M09-R01 - P6 purpose: P6 SHALL provide the architecture responsibilities required to govern Decision Formation, operative decision commitment, context-specific Authorization, consequential action control, human oversight, escalation and intervention without redefining GEAIRM or AIPS semantics.
GEAIRA-M09-R02 - No universal pipeline: P6 SHALL NOT require Recommendation, Plan, Decision Formation, Decision effect, Authorization and Action to occur as one mandatory sequential pipeline.
26.2 Semantic and authority baseline
P6 is grounded in the following frozen semantic authorities.
| Semantic concern | Authority | P6 treatment |
|---|---|---|
| RecommendationEffect, Decision, Plan, DigitalAction, PhysicalAction | AIPS Intelligence Effect | REUSED; P6 does not redefine effect identity |
| Agency / Autonomy values | AIPS | REUSED; informs architecture patterns but does not replace occurrence authorization |
| Decision Formation | GEAIRM | REUSED |
| Authorization | GEAIRM | REUSED |
| Delegated Authority | GEAIRM | REUSED as a broader mandate/envelope |
participatesInDecisionFormation | GEAIRM | REUSED |
informsDecisionFormation | GEAIRM | REUSED |
authorizedBy | GEAIRM | REUSED |
withinDelegatedAuthority | GEAIRM | REUSED |
| Agent identity/taxonomy/control-mode semantics | KivaNura Agent Ontology | REFERENCED |
| Productness / Product identity | BPS | REFERENCED |
| AI Product qualification / Product-version risk-governance semantics | AIPS | REFERENCED |
| organizational/legal decision authority and accountability | applicable enterprise/legal/regulatory authority | EXTERNAL / REFERENCED |
| decision-support analytics/information semantics | GEARM/GEARA | CROSS-DOMAIN REFERENCE |
| approval workflow, IAM, policy enforcement and action technology | implementation/profile | REALIZATION CHOICE |
P6 preserves these hard distinctions:
RecommendationEffect != operative Decision
Plan != Decision
Plan != Action
Decision Formation != Decision Intelligence Effect
Decision Intelligence Effect != organizational decision authority
Decision authority != Authorization
Authorization != Delegated Authority
Authorization != successful Action
Authorization != safety / correctness / low risk
Human participation != proof of review / approval
Agent autonomy != authority
Action effect != proof of intended business outcome
GEAIRA-M09-R03 - Semantic authority preservation: P6 SHALL consume AIPS, GEAIRM, Agent Ontology, BPS and applicable enterprise/legal authority semantics without minting replacement meanings for Recommendation, Plan, Decision, Authorization, Delegated Authority, Agent autonomy, Product status or accountability.
GEAIRA-M09-R04 - Effect/authority separation: An AIPS Decision Intelligence Effect SHALL NOT by itself establish operative organizational decision authority, Authorization, accountability or permission to act.
GEAIRA-M09-R05 - Human-participation restraint: The presence of a Human Agent in a Decision Formation or Control Loop SHALL NOT by itself prove review, approval, final decision authority, accountability or informed consent.
26.3 P6 architecture boundary
P6 is the primary architecture plane for:
- Decision Formation coordination where enterprise control is required;
- determining and applying the applicable decision/action control pattern;
- point-of-use Authorization where required;
- binding Authorization to scope, conditions, identity, target and current authority basis;
- human review, approval, intervention, escalation and override where required;
- committing/rejecting/deferring an operative decision or action outcome;
- safe hand-off of an authorized action to its technical enactment boundary;
- recording occurrence-level decision/authorization/oversight evidence;
- handling indeterminate, conflicting, expired or revoked authority/control state;
- responding to P7 assurance/safety/currentness feedback.
P6 does not own:
- enterprise purpose or standing governing mandate - P1;
- AI Context construction - P2;
- Intelligence Capability / AI Primitive realization - P3;
- AI Execution, Agent runtime coordination or technical Actuation - P4;
- serving/Product exposure and consumer integration - P5;
- evaluation/assurance and systemic observability - P7;
- Productness or AI Product qualification - BPS/AIPS;
- a universal organizational/legal decision-authority ontology - applicable enterprise/legal authority.
GEAIRA-M09-R06 - P1/P6 separation: P1 SHALL establish the governing authority and policy envelope; P6 SHALL apply the principal context-specific decision, Authorization, action-control and human-oversight responsibilities within that envelope.
GEAIRA-M09-R07 - P4/P6 separation: P4 technical capability to invoke a Tool, actuate an Environment or produce a decision-like effect SHALL NOT establish permission to commit a consequential decision or action; applicable control belongs principally to P6.
GEAIRA-M09-R08 - P5/P6 separation: Valid admission through a P5 serving contract SHALL NOT by itself authorize a consequential Decision or Action.
26.4 Canonical P6 Architectural Responsibilities
M09 freezes the following fourteen canonical P6 Architectural Responsibilities.
| ID | Responsibility | Applicability | Required architectural outcome |
|---|---|---|---|
| P6-RESP-01 | Decision-Control Context Determination | Baseline-mandatory | Determine which decision/action control pattern applies to the occurrence, including whether P6 intervention is required at all. |
| P6-RESP-02 | Decision Input & Evidence Binding | Conditional-mandatory | Bind relevant recommendations, plans, context, evidence, human input and governing constraints to the Decision Formation occurrence. |
| P6-RESP-03 | Decision Formation Coordination | Conditional-mandatory | Coordinate the governed formation of a determination where AI, human or system participants contribute. |
| P6-RESP-04 | Operative Decision Commitment | Conditional-mandatory | Record/commit the operative determination where a governed decision is required, preserving the distinction from AI-produced Decision effects. |
| P6-RESP-05 | Authorization Requirement Determination | Baseline-mandatory | Determine whether a specific Decision, invocation or Action requires context-specific Authorization. |
| P6-RESP-06 | Point-of-Use Authorization | Conditional-mandatory | Establish, consume or verify the context-bounded Authorization that permits the governed occurrence to proceed. |
| P6-RESP-07 | Authorization Scope & Condition Binding | Conditional-mandatory | Bind Authorization to subject/actor, target, action/decision scope, conditions, time/currentness and applicable Delegated Authority/policy basis. |
| P6-RESP-08 | Human Oversight Requirement Determination | Baseline-mandatory | Determine whether review, approval, supervision, intervention, escalation or post-hoc oversight is required and at what control point. |
| P6-RESP-09 | Human Review / Approval / Intervention | Conditional-mandatory | Enable required human participation with sufficient context, time, authority and ability to approve, reject, intervene, defer or escalate. |
| P6-RESP-10 | Consequential Action Commitment & Control | Conditional-mandatory | Commit, reject, constrain, defer or cancel a consequential Action before technical enactment where applicable. |
| P6-RESP-11 | Escalation, Exception & Override Handling | Conditional-mandatory | Route unresolved, conflicting, exceptional or safety-sensitive cases to an applicable governed escalation/override path. |
| P6-RESP-12 | Revocation, Intervention & Stop Propagation | Conditional-mandatory | Propagate authority withdrawal, human intervention, emergency stop or equivalent control to affected execution/action boundaries. |
| P6-RESP-13 | Decision / Authorization / Oversight Evidence | Baseline-mandatory | Preserve traceable evidence of inputs, authority basis, decisions, approvals/rejections, conditions, human participation and action outcomes. |
| P6-RESP-14 | Outcome / Assurance Feedback Intake | Conditional-mandatory | Consume P7/P4/P5 outcome, assurance, safety and currentness feedback that may affect continued decision/action permission. |
GEAIRA-M09-R09 - Responsibility completeness: A conforming P6 realization SHALL address each applicable P6 responsibility explicitly; co-location or delegation to another component SHALL NOT erase responsibility traceability.
GEAIRA-M09-R10 - Control-pattern determination: P6-RESP-01 SHALL permit low-consequence/advisory cases to bypass unnecessary decision/action control while requiring stronger control where policy, authority, safety, risk or consequence demands it.
GEAIRA-M09-R11 - Decision-input non-acceptance: Binding a Recommendation, Plan, Decision effect, analytical output or Evidence item into Decision Formation SHALL NOT imply that the input is accepted, correct or determinative.
GEAIRA-M09-R12 - Operative-decision distinction: P6-RESP-04 SHALL preserve a distinguishable operative decision record/occurrence where governance requires one and SHALL NOT silently equate it with an AIPS Decision Intelligence Effect.
GEAIRA-M09-R13 - Authorization conditionality: P6-RESP-06 SHALL apply only where the governed context requires a specific Authorization; GEAIRA SHALL NOT invent an Authorization occurrence for every AI Execution or every informational result.
GEAIRA-M09-R14 - Human-oversight explicitness: Where human oversight is required, P6-RESP-08/09 SHALL make its timing, purpose, authority, available information and permitted intervention explicit rather than relying on an undocumented operational assumption.
26.5 Decision Formation and operative Decision
P6 distinguishes Decision Formation from the eventual operative enterprise determination.
Score / Recommendation / Plan / Decision effect / evidence / human judgment
|
v
Decision Formation
|
+------------+------------+
| |
v v
operative determination unresolved / defer /
where one is required escalate / terminate
Decision Formation may be:
- human-led;
- AI-assisted;
- Agent-coordinated;
- rule-mediated;
- multi-party;
- asynchronous;
- embedded in a Control Loop;
- omitted entirely for purely informational use.
The architecture does not require a distinct persisted Decision object unless the applicable profile/governance context requires one. Where an operative determination must be evidenced, P6 SHALL preserve its identity and relationship to the inputs and authority basis.
GEAIRA-M09-R15 - Decision-formation diversity: P6 SHALL support human-led, AI-assisted, automated, distributed and exception-based Decision Formation without assuming one universal workflow or participant model.
GEAIRA-M09-R16 - No forced AI Decision effect: A human or organizational Decision MAY be formed without an AI Decision Intelligence Effect, and P6 SHALL NOT invent such an effect merely to normalize the architecture.
GEAIRA-M09-R17 - No reverse inference: Presence of a later Action or Authorization SHALL NOT be used to infer that Recommendation, Plan or AI Decision effect necessarily occurred beforehand.
26.6 Point-of-use Authorization model
GEAIRM Authorization is context-bounded. P6 therefore treats an Authorization as a governed permission basis for a specific scope rather than a permanent entitlement.
A P6 Authorization binding SHOULD be capable of identifying, where applicable:
authorizationId
authority source / approver identity
authorized subject / participant
authorized decision / invocation / action scope
target / resource / environment scope
conditions and constraints
applicable Delegated Authority / policy basis
valid-from / valid-until / currentness
revocation state
required human involvement
risk / safety conditions where externally governed
related Decision Formation / execution / action occurrence
supporting evidence / reason
Not every realization requires all fields, but the applicable scope and authority basis must be determinable.
Delegated Authority
broad bounded mandate
|
v
Authorization
context-specific grant/permission
|
v
Decision / invocation / Action occurrence
GEAIRA-M09-R18 - Authorization scope: A P6 Authorization SHALL be bounded to an identifiable scope/context and SHALL NOT be generalized into unrestricted future authority.
GEAIRA-M09-R19 - Delegated-authority containment: Where an Authorization is governed by a Delegated Authority, P6 SHALL NOT expand the Authorization beyond the applicable delegated mandate.
GEAIRA-M09-R20 - Authorization non-success: Authorization SHALL NOT be treated as proof that the authorized operation succeeded, was safe, was correct, achieved the intended business outcome or remained within scope after execution.
GEAIRA-M09-R21 - Currentness: Expired, revoked, superseded or materially indeterminate Authorization SHALL NOT silently remain valid for new governed occurrences.
26.7 Authorization patterns
P6 supports multiple valid authorization patterns.
Pattern A - No occurrence Authorization required
AI Execution -> informational effect -> consumer
Applicable where the result is advisory/informational and no governed consequential commitment requires occurrence-specific permission.
Pattern B - Synchronous point-of-use approval
proposed consequential action
|
v
P6 authorization / human approval
|
approve
v
technical enactment
Pattern C - Pre-authorized bounded envelope
P1 Delegated Authority / policy envelope
|
v
P6 establishes bounded pre-authorization conditions
|
v
repeated P4 actions while conditions/currentness remain valid
This is useful for low-latency or control-loop cases but does not create unlimited authority.
Pattern D - Local embedded authorization
A safety-critical or hard-real-time environment MAY realize P6 logic locally/embedded rather than through a centralized network service.
Pattern E - External authority decision
P6 MAY consume a context-specific Authorization established by an external authority if provenance, scope and currentness are sufficiently trustworthy.
GEAIRA-M09-R22 - Authorization realization neutrality: GEAIRA SHALL NOT require centralized synchronous authorization for every governed occurrence; authorization responsibilities MAY be embedded, local, federated, human-mediated, cached or externally established where their validity and scope are preserved.
GEAIRA-M09-R23 - Pre-authorization restraint: A pre-authorized envelope SHALL have explicit bounds, currentness/revocation semantics and evidence sufficient to determine whether a particular occurrence remained within the permitted scope.
26.8 Human oversight architecture
Human oversight is an explicit architecture responsibility, not a decorative label.
M09 recognizes several architecture patterns without redefining Agent Ontology control-mode semantics:
| Pattern | Architectural purpose |
|---|---|
| Pre-decision review | Human examines inputs/recommendation before an operative decision is committed. |
| Approval before action | Human Authorization/approval is required before consequential enactment. |
| Supervisory monitoring | Human monitors ongoing operation and can intervene under defined conditions. |
| Exception review | Only uncertain/high-risk/out-of-policy cases are routed for human disposition. |
| Escalation decision | Human or designated authority resolves conflicts/indeterminacy beyond automated scope. |
| Override / stop | Human or authorized mechanism can pause, reject, reverse where possible, or stop further action. |
| Post-hoc review | Human reviews completed occurrences for governance/assurance without pretending to have approved them beforehand. |
A genuine human oversight responsibility requires more than putting a person in a workflow. The human must have, as applicable:
- relevant decision/action context;
- sufficient explanation/evidence for the required task;
- known role and authority;
- enough time to act for the intended control purpose;
- an effective mechanism to approve, reject, intervene, stop or escalate;
- visibility of material uncertainty/indeterminacy;
- evidence that the oversight event actually occurred.
GEAIRA-M09-R24 - Meaningful human control: Where human oversight is a control requirement, the architecture SHALL provide the human with sufficient information, authority, timing and intervention capability for that control to be meaningful.
GEAIRA-M09-R25 - No rubber-stamp inference: A human approval event SHALL NOT be treated as meaningful oversight merely because a UI click or workflow transition occurred; profiles SHALL preserve the required review/evidence expectations appropriate to the control purpose.
GEAIRA-M09-R26 - Post-hoc honesty: Post-hoc human review SHALL NOT be represented as prior approval or pre-action oversight.
GEAIRA-M09-R27 - Oversight evidence: P6 SHALL distinguish actual recorded human review/approval/intervention from merely configured or theoretically available human participation.
26.9 Decision authority, Authorization and accountability
P6 consumes but does not own a universal theory of organizational/legal decision authority or accountability.
AI Decision effect
what the AI produced
operative Decision
determination used by the enterprise process
Decision authority
who/what is legitimately empowered to make that determination
Authorization
permission for a specific occurrence to proceed
Accountability
organizational/legal responsibility for the governed outcome
These may involve the same participant in a concrete enterprise, but GEAIRA does not collapse them.
GEAIRA-M09-R28 - Authority/accountability separation: P6 SHALL preserve distinguishable architectural references to operative decision authority, Authorization and accountability where the applicable governance context differentiates them.
GEAIRA-M09-R29 - No accountability inference: P6 SHALL NOT infer legal/organizational accountability solely from AI participation, human approval, Agent identity, AIPS Agency/Autonomy classification or occurrence Authorization.
26.10 Consequential Action control and enactment
P6 controls whether/how a consequential action may be committed; the technical side effect may be enacted elsewhere.
proposed action / Decision outcome
|
v
P6 action commitment / Authorization
|
+------+------+
| |
v v
approve reject / defer /
| escalate / condition
v
P4 Tool / Actuation
or P5/external integration
|
v
observed outcome / evidence
This supports:
- digital actions;
- physical actions;
- business-process commitments;
- external-provider commands;
- Tool calls with side effects;
- irreversible or partially reversible operations.
GEAIRA-M09-R30 - Commit/enact separation: P6 action commitment/Authorization SHALL remain distinguishable from P4/P5/external technical enactment where the realization separates them.
GEAIRA-M09-R31 - Consequence-sensitive control: P6 SHALL permit stronger authorization, human-oversight, safety and evidence requirements for materially consequential actions without requiring the same control intensity for every AI output.
GEAIRA-M09-R32 - Indeterminate side effect: Where an action request times out or fails ambiguously and the system cannot determine whether the side effect occurred, P6 SHALL treat the outcome as indeterminate/reconciliation-required rather than automatically safe to retry.
GEAIRA-M09-R33 - Reversibility awareness: Where reversibility materially affects action risk/control, P6 SHOULD make irreversible, compensatable and readily reversible action paths distinguishable to applicable control logic and human oversight.
26.11 Agent autonomy and P6 control
AIPS Agency/Autonomy classification can influence which P6 pattern is appropriate but does not replace P6 responsibilities.
Inform
may require no action Authorization for the informational output itself
Recommend
recommendation remains advisory until an applicable Decision/Action path accepts it
Assist
may participate actively while preserving applicable human/authority controls
ActWithApproval
commonly aligns with a P6 approval/Authorization pattern
but is not semantically identical to Authorization
AutonomousAction
may operate within a governed delegated/pre-authorized envelope
but is not unlimited authority
GEAIRA-M09-R34 - AR/Authorization orthogonality: AIPS Agency/Autonomy classification SHALL NOT substitute for context-specific Authorization or human-oversight requirements where those are independently applicable.
GEAIRA-M09-R35 - AutonomousAction restraint: AutonomousAction SHALL NOT be interpreted as permission to exceed P1 governing purpose, Delegated Authority, P6 action controls, safety constraints or revocation state.
26.12 Canonical P6 ABBs
Applying the M02 ABB admission criteria yields three canonical logical P6 ABBs.
P6-ABB-01 - Decision Formation & Commitment Boundary
Purpose:
Coordinate governed Decision Formation and preserve the transition from advisory/informational inputs to an operative enterprise determination where one is required.
Primary responsibilities:
- P6-RESP-01 Decision-Control Context Determination;
- P6-RESP-02 Decision Input & Evidence Binding;
- P6-RESP-03 Decision Formation Coordination;
- P6-RESP-04 Operative Decision Commitment;
- portions of P6-RESP-13 Decision/Authorization/Oversight Evidence.
This ABB is justified because the boundary:
- separates advisory AI output from operative commitment;
- creates an independently governable decision-control boundary;
- may coordinate human, AI and non-AI participants;
- has distinct failure/escalation and evidence obligations;
- is technology-neutral.
It does not imply a Decision Engine, BPM engine or universal persisted decision object.
P6-ABB-02 - Authorization & Action Control Boundary
Purpose:
Determine and apply context-bounded point-of-use Authorization and consequential action-control conditions before governed commitment/enactment.
Primary responsibilities:
- P6-RESP-05 Authorization Requirement Determination;
- P6-RESP-06 Point-of-Use Authorization;
- P6-RESP-07 Authorization Scope & Condition Binding;
- P6-RESP-10 Consequential Action Commitment & Control;
- P6-RESP-12 Revocation, Intervention & Stop Propagation;
- portions of P6-RESP-13/14.
This ABB does not imply a standalone Authorization Service, policy engine, IAM service, entitlement service, action gateway or command bus.
P6-ABB-03 - Human Oversight & Escalation Boundary
Purpose:
Provide explicit, evidence-bearing human review, approval, intervention, escalation and override responsibilities where the governing context requires human participation.
Primary responsibilities:
- P6-RESP-08 Human Oversight Requirement Determination;
- P6-RESP-09 Human Review / Approval / Intervention;
- P6-RESP-11 Escalation, Exception & Override Handling;
- portions of P6-RESP-12/13/14.
This ABB is conditional: P6 does not force human involvement into every AI path.
It does not imply a human-review queue, case-management system, approval portal or supervisor console.
GEAIRA-M09-R36 - ABB restraint: M09 SHALL freeze only the three logical P6 ABBs above and SHALL NOT promote concrete workflow, IAM, policy, approval, case-management, queue, UI or action-gateway technologies to canonical ABB status.
GEAIRA-M09-R37 - ABB co-location: One realization MAY co-locate all three P6 ABBs, distribute one ABB across multiple components, or embed a P6 ABB in a runtime/control environment while preserving logical responsibility and evidence boundaries.
26.13 Candidates deliberately not promoted to canonical P6 ABBs
The following are implementation/profile choices rather than generic canonical ABBs:
| Candidate | M09 disposition |
|---|---|
| Decision Engine | NOT ADMITTED |
| Rules Engine | NOT ADMITTED |
| BPM / Workflow Engine | NOT ADMITTED |
| Approval Service | NOT ADMITTED |
| Authorization Service | NOT ADMITTED |
| Policy Engine / PDP | NOT ADMITTED |
| IAM / Entitlement / PAM Service | NOT ADMITTED |
| Human Review Queue | NOT ADMITTED |
| Case Management Platform | NOT ADMITTED |
| Supervisor Console / Approval UI | NOT ADMITTED |
| Action Gateway / Command Gateway | NOT ADMITTED |
| Command Bus / Task Queue | NOT ADMITTED |
| Kill Switch Service | NOT ADMITTED |
| Risk Engine | NOT ADMITTED |
| Governance Workflow Platform | NOT ADMITTED |
| Ticketing / Work Management System | NOT ADMITTED |
| RPA / Automation Orchestrator | NOT ADMITTED |
| Robotic / PLC / SCADA Controller | NOT ADMITTED |
A concrete architecture may use any of these where justified.
26.14 Canonical P6 architectural interactions
M09 establishes twelve technology-neutral P6 interactions.
| ID | Interaction | Direction | Architectural meaning |
|---|---|---|---|
| P6-INT-01 | Governing Authority / Policy Basis | P1 -> P6 | Provide governing purpose, authority, accountability and policy basis. |
| P6-INT-02 | Decision Context / Evidence Provision | P2/P7 -> P6 | Provide applicable context, evidence, provenance, currentness or assurance information. |
| P6-INT-03 | Decision / Action Control Request | P4/P5 -> P6 | Present a proposed decision, invocation or action requiring P6 control. |
| P6-INT-04 | Decision Input / Intelligence Effect | P4/P5/GEARA -> P6 | Provide recommendation, plan, score, analytical/information output or Decision effect that may inform Decision Formation. |
| P6-INT-05 | Human Review / Approval Request | P6 -> Human / external authority | Request review, approval, judgment, intervention or escalation. |
| P6-INT-06 | Human / Authority Outcome | Human / external authority -> P6 | Return approval, rejection, conditions, determination, escalation or intervention. |
| P6-INT-07 | Authorization / Decision Outcome | P6 -> P4/P5 | Return permitted/rejected/conditional/deferred/indeterminate control outcome. |
| P6-INT-08 | Authorized Action Commitment | P6 -> P4/P5/external system | Hand off an authorized/committed action for technical enactment. |
| P6-INT-09 | Action / Decision Outcome Evidence | P4/P5/external system -> P6 | Return success/failure/indeterminate/observed outcome information for reconciliation. |
| P6-INT-10 | Decision / Authorization Evidence | P6 -> P7 | Expose decision, authorization, human-oversight and action-control evidence. |
| P6-INT-11 | Assurance / Safety / Currentness Feedback | P7/P1 -> P6 | Constrain, suspend, revoke or modify continued control behavior. |
| P6-INT-12 | Intervention / Revocation / Stop | P6 -> P4/P5/external system | Propagate stop, revoke, override or intervention control where applicable. |
These interactions may be synchronous, asynchronous, event-driven, local/embedded, human-mediated, federated or batch-oriented.
GEAIRA-M09-R38 - Interaction neutrality: P6 interactions SHALL remain technology-neutral and SHALL NOT require REST, workflow APIs, IAM protocols, queues, message buses, case-management products or centralized control services.
GEAIRA-M09-R39 - Evidence correlation: A governed P6 interaction SHALL preserve sufficient correlation among Decision Formation, Authorization, human-oversight event, action commitment, P4 execution and P5 serving occurrence where those elements materially participate in the path.
26.15 Logical P6 control points
M09 establishes three logical control-point classes.
CP6-01 - Decision Commitment Control Point
Determines whether an advisory/informational formation process may be committed as an operative determination where governance requires explicit decision control.
CP6-02 - Authorization & Action Commitment Control Point
Determines whether a Decision, invocation or Action may proceed under the applicable authority, conditions, currentness and safety constraints.
CP6-03 - Human Intervention & Escalation Control Point
Provides required human review, escalation, intervention, override or stop capability.
These are logical control responsibilities. They may be embedded in one component, distributed, human-operated or federated.
GEAIRA-M09-R40 - Control-point non-inflation: P6 control points SHALL NOT automatically be modelled as standalone deployed services or network hops.
26.16 Failure, indeterminacy and safe degradation
P6 must treat decision/action control failures explicitly.
Generic failure/degradation classes include:
- governing authority/policy basis unavailable or stale;
- Decision Formation input/evidence incomplete or materially conflicting;
- required human reviewer/approver unavailable;
- reviewer lacks authority or sufficient information;
- Authorization rejected;
- Authorization expired/revoked/superseded;
- Authorization outcome indeterminate;
- requested action exceeds Delegated Authority or policy scope;
- safety/interlock condition violated;
- action-commit handoff failure;
- action outcome unknown / ambiguous side effect;
- human intervention/stop propagation failure;
- assurance/currentness feedback requires suspension;
- conflicting authoritative approvals/constraints.
For material cases P6 SHALL choose an explicit governed response such as:
reject
fail closed
defer
escalate
enter safe state
continue only within a narrower pre-authorized envelope
require reconciliation
require human disposition
GEAIRA-M09-R41 - Fail-safe control: Where required Authorization, authority currentness, safety state or human approval is unavailable or materially indeterminate, P6 SHALL NOT default to permissive consequential action unless an explicit governing rule authorizes a defined degraded mode.
GEAIRA-M09-R42 - Conflict handling: Conflicting authoritative constraints/approvals SHALL be surfaced and resolved/escalated according to governed precedence rather than silently selecting the least restrictive outcome.
GEAIRA-M09-R43 - Stop propagation: Where a valid intervention/revocation/stop condition applies, P6 SHALL propagate that control to affected downstream enactment boundaries with priority appropriate to the governing/safety context.
26.17 Cross-cutting overlays applied to P6
All eight M03 cross-cutting overlays apply to P6.
| Overlay | P6 architectural implication |
|---|---|
| X1 Security, Privacy & Trust | Protect sensitive decision inputs, approvals, credentials and action controls; prevent unauthorized manipulation. |
| X2 Identity, Authority & Policy Enforcement | Establish approver/actor identity, authority source, Authorization scope and policy-currentness evidence. |
| X3 Evidence, Provenance & Traceability | Preserve lineage from inputs/effects through decision, Authorization, human participation and action outcome. |
| X4 Observability, Telemetry & Auditability | Observe control decisions, approvals/rejections, intervention latency and downstream outcomes without equating logs with assurance. |
| X5 Resilience, Safety & Failure Management | Provide fail-safe/degraded behavior, intervention, emergency stop and indeterminate-outcome handling. |
| X6 Lifecycle, Version, Currentness & Change | Track current authority/policy/approval state while preserving historical occurrence truth. |
| X7 Interoperability, Contracts & Semantic Compatibility | Maintain clear decision/action-control contracts across P1/P4/P5/P7 and external authorities. |
| X8 Governance, Human Accountability & Compliance | Make required human roles, oversight obligations, escalation and accountable governance evidence explicit. |
GEAIRA-M09-R44 - Overlay continuity: P6 SHALL apply X1-X8 according to applicability and SHALL NOT localize cross-cutting obligations solely inside the P6 components that happen to implement them.
26.18 Representative P6 scenarios
Scenario 1 - Advisory recommendation, human final Decision
P4 -> RecommendationEffect
|
v
P6 Decision Formation
|
Human review
|
v
operative human Decision
The Recommendation informs but does not become the operative Decision automatically. No action Authorization is created unless a downstream action requires one.
Scenario 2 - ActWithApproval Product Version
P4 forms a proposed action. P6 determines that human approval is required, records the approved/rejected Authorization, and only an approved occurrence is handed to P4/P5 for enactment.
ActWithApproval remains AIPS classification and is not semantically identical to the P6 Authorization event.
Scenario 3 - Autonomous bounded action
An AutonomousAction Product/Agent operates within a valid P1 Delegated Authority and P6 pre-authorized action envelope. P6 does not require a human approval for each action, but scope/currentness/safety controls and evidence remain explicit. Exceeding the envelope triggers rejection/escalation.
Scenario 4 - Hard-real-time physical safety loop
P6 Authorization/action-control responsibility is embedded locally near P4 actuation/control logic. Network loss to a central enterprise control plane does not force unsafe latency; the local governed envelope and safety interlock determine permitted operation and evidence is reconciled later.
Scenario 5 - Indeterminate payment/tool side effect
P6 authorizes a consequential Tool action. P4 reports a timeout and cannot determine whether the side effect occurred. P6 records the state as indeterminate/reconciliation-required and does not blindly retry.
Scenario 6 - Human oversight unavailable
A case requires human approval by policy. The designated reviewer is unavailable before the validity window expires. P6 defers/escalates or fails closed according to the governed policy rather than treating non-response as approval.
Scenario 7 - P7 safety finding revokes continued action
P7 emits a current safety/assurance finding that invalidates continued autonomous action. P6 narrows/revokes the action envelope and propagates the stop/control outcome to P4 without rewriting previous valid historical Authorizations.
26.19 P6 anti-patterns
M09 prohibits or warns against the following generic anti-patterns:
- recommendation-equals-decision - treating advisory AI output as the operative enterprise Decision.
- AI-decision-equals-authority - treating an AIPS Decision effect as proof of legitimate decision authority.
- authorization-equals-delegation - treating occurrence permission and broad delegated mandate as identical.
- autonomy-equals-permission - treating
AutonomousActionor Agent autonomy as unrestricted authority. - human-presence-equals-oversight - assuming a human participant proves meaningful review/control.
- rubber-stamp-HITL - implementing nominal approval without meaningful context, time or intervention capability.
- workflow-equals-governance - assuming a BPM/approval workflow itself establishes semantic authority or accountability.
- central-authorizer-everything - forcing every decision/action through one network Authorization service regardless of latency/safety context.
- policy-equals-authorization - treating applicable policy/risk obligation as a specific occurrence Authorization.
- decision-engine-everything - collapsing human, AI, rule and external authority paths into one generic Decision Engine ABB.
- action-commit-equals-execution - failing to distinguish permission/commitment from technical enactment.
- timeout-equals-failure - assuming an ambiguous side effect did not occur and blindly retrying.
- latest-authority-equals-historical-authority - rewriting historical decisions/approvals as though later authority state applied.
- silent-override - allowing overrides/interventions without evidence, scope or authority traceability.
- post-hoc-equals-prior-approval - representing after-the-fact review as pre-action approval.
- risk-tier-equals-action-permission - treating AIPS risk classification as direct permission to decide or act.
- P1/P6-collapse - conflating standing purpose/authority envelope with occurrence-level Authorization.
- P4/P6-collapse - treating technical ability to execute/act as permission to do so.
26.20 P6 lifecycle and historical truth
P6 preserves historical occurrence truth when authority, policy, approval or oversight state changes later.
Example:
Authorization A17 valid at T1
|
+--> permits ActionOccurrence X at T1
later at T2:
Authorization A17 revoked / policy superseded
The later change affects future/current eligibility. It does not rewrite the historical fact that X was authorized under A17 at T1 if that was true at the time.
Similarly:
- a later reviewer-role change does not rewrite who approved an earlier action;
- a later policy version does not replace the version that governed an earlier occurrence;
- a later P7 finding may invalidate future use without fabricating that earlier Authorization never existed;
- remediation/compensation may be required without historical mutation.
GEAIRA-M09-R45 - Historical decision preservation: Later policy, authority, Authorization, human-role or assurance changes SHALL NOT rewrite which control state actually applied to a historical decision/action occurrence.
GEAIRA-M09-R46 - Future-currentness separation: P6 SHALL distinguish current/future permission from historical Authorization evidence and SHALL NOT use currentness updates to erase valid historical traceability.
26.21 M09 frozen decisions
M09 freezes the following P6 architecture decisions for GEAIRA 0.1:
P6 is the primary Decision, Action & Human Oversight plane.
RecommendationEffect != operative Decision.
Decision Intelligence Effect != organizational decision authority.
Decision authority != Authorization.
Authorization != Delegated Authority.
Authorization != success / safety / correctness.
Human participation != proof of review / approval.
Agent autonomy / AIPS AR != occurrence Authorization.
P1 owns standing governing intent/authority envelope.
P6 owns principal point-of-use decision/action control.
P4 owns technical AI Execution and applicable enactment.
P5 serving admission != consequential Authorization.
Human oversight must be explicit and evidence-bearing where required.
No universal Recommendation -> Plan -> Decision -> Authorization -> Action pipeline exists.
Hard-real-time/local embedded control is permitted.
Historical decision/authorization truth survives later supersession/revocation.
The following P6 ABBs are canonical:
P6-ABB-01 Decision Formation & Commitment Boundary
P6-ABB-02 Authorization & Action Control Boundary
P6-ABB-03 Human Oversight & Escalation Boundary
M09 does not freeze:
- a universal Decision ontology beyond the consumed semantic authorities;
- a universal legal/accountability model;
- one approval workflow;
- one human-in-the-loop pattern;
- centralized Authorization infrastructure;
- IAM/PAM technology;
- a policy engine;
- a Decision Engine;
- BPM/case-management technology;
- an action gateway/command bus;
- a risk engine;
- final P7 assurance/observability architecture;
- final M11 inter-plane contracts.
GEAIRA-M09-R47 - Downstream boundary preservation: Later slices SHALL preserve the P1/P6 authority distinction, P4/P6 execution/action distinction, P5/P6 admission/authorization distinction and P6/P7 control/assurance distinction unless a reconciled 0.1 architecture change explicitly requires otherwise.
GEAIRA-M09-R48 - Human-oversight profileability: Later profiles MAY strengthen human-review, approval, explanation, dual-control, separation-of-duty, escalation or intervention requirements but SHALL NOT weaken mandatory P6 invariants applicable to the profiled context.
GEAIRA-M09-R49 - No Product inference: Presence of decision/action control, human approval or Authorization architecture SHALL NOT establish Productness or AI Product qualification.
GEAIRA-M09-R50 - Architecture diversity continuity: P6 SHALL remain valid across advisory-only, human-final, AI-assisted, approval-required, bounded autonomous, supervisory and embedded safety-control patterns without structural redesign solely because the decision/action-control style differs.
27. M09 Completion State
GEAIRA-M09 - COMPLETE.
M09 establishes:
- one canonical P6 purpose and Decision/Action/Human-Oversight boundary;
- fourteen canonical P6 Architectural Responsibilities;
- three canonical logical P6 ABBs;
- eighteen deliberately non-admitted implementation-style ABB candidates;
- twelve technology-neutral P6 architectural interactions;
- three logical P6 control-point classes;
- fourteen generic P6 failure/degradation classes;
- explicit application of all eight X1-X8 cross-cutting overlays;
- explicit P1/P4/P5/P7 and external decision-authority/accountability boundaries;
- seven representative P6 scenarios;
- eighteen P6 anti-patterns;
- 50 normative M09 rules.
M09 introduces no change to GEAIRM, BPS, AIPS or Agent Ontology semantics, does not define a universal legal/accountability model, and does not prescribe a Decision Engine, Authorization Service, BPM/workflow engine, human-review queue, policy engine, IAM/PAM system, action gateway or centralized control topology.
Next slice
GEAIRA-M10 - P7: Evaluation, Assurance, Observability & Evolution
M10 will define the architecture responsibilities for evaluation, evidence, assurance, observability, currentness, lifecycle/evolution feedback and how those responsibilities constrain continued enterprise-AI operation without turning telemetry into assurance or later findings into historical semantic mutation.
28. GEAIRA-M10 - P7: Evaluation, Assurance, Observability & Evolution
28.1 Purpose
GEAIRA-M10 defines the canonical architecture responsibilities for P7 - Evaluation, Assurance, Observability & Evolution.
P7 answers:
How is enterprise AI made observable, evaluable, evidentially assessable, current, and governable over time?
P7 closes the principal feedback loop across GEAIRA. It receives material evidence, telemetry, results, lifecycle/currentness signals and semantic-change information from P1-P6 and relevant external authorities; coordinates or supports Evaluation and Assurance; and returns findings, currentness conclusions and change signals that may constrain future operation.
P7 is not a generic monitoring platform. It does not equate telemetry with Evidence, Evaluation with Assurance, an Assurance Finding with approval, or currentness with simple timestamp freshness.
GEAIRA-M10-R01 - P7 purpose: P7 SHALL provide the architecture responsibilities required to observe material AI behavior/state, preserve usable evidence, coordinate evaluation and assurance, assess currentness, and propagate evolution/change feedback without redefining the underlying semantic authorities.
GEAIRA-M10-R02 - Feedback-loop restraint: P7 feedback SHALL influence future eligibility, constraints, remediation or governance where applicable but SHALL NOT retroactively rewrite historical execution, decision, evidence, version or authority facts.
28.2 Semantic authority baseline
P7 is grounded in the following frozen semantic authorities.
| Semantic concern | Authority | P7 treatment |
|---|---|---|
| Evaluation | GEAIRM | REUSED |
| Evaluation Criterion | GEAIRM | REUSED |
| Evaluation Result | GEAIRM | REUSED |
| Evaluation Target role | GEAIRM | REUSED |
| Evidence role | GEAIRM | REUSED |
| Assurance Assessment | GEAIRM | REUSED |
| Assurance Finding | GEAIRM | REUSED |
| Version Identifier | GEAIRM | REUSED |
| Lifecycle State | GEAIRM | REUSED |
| Currentness Assessment | GEAIRM | REUSED |
| Semantic Change Event | GEAIRM | REUSED |
evaluatesTarget, usesEvaluationCriterion, producesEvaluationResult | GEAIRM | REUSED |
recordedInAIEvaluationReport, usesEvidence | GEAIRM/AIPS bridge | REUSED |
assessesAssuranceOf, producesAssuranceFinding | GEAIRM | REUSED |
identifiedByVersion, hasLifecycleState | GEAIRM | REUSED |
assessesCurrentnessOf, hasCurrentnessConcern, hasCurrentnessStatus | GEAIRM | REUSED |
supersedes, affectsSemanticTarget | GEAIRM | REUSED |
| AI Evaluation Report, AI Product Assurance, Fitness for Purpose, Trust Currentness | AIPS | REFERENCED; P7 does not redefine AIPS Product assurance semantics |
| Effective Risk Tier, Risk Driver Assessment, Governance Obligation Profile | AIPS | REFERENCED; not replaced by P7 findings |
| Provenance | W3C PROV | REFERENCED |
| Product identity/version semantics | BPS/AIPS | REFERENCED |
| Data / Information quality and lineage semantics | GEDRM/GEARM and adjacent architectures | CROSS-DOMAIN REFERENCE |
| telemetry stores, SIEM/APM/ML monitoring tools, dashboards, data lakes | implementation/profile | REALIZATION CHOICE |
P7 preserves these hard distinctions:
Observation != Evaluation
Telemetry != Evidence automatically
Log record != provenance automatically
Evaluation != Assurance
Evaluation Result != Assurance Finding
Evidence != proof
Assurance != approval / Authorization
Assurance != Risk Tier
Risk Tier != Governance Obligation
Currentness != freshness
Lifecycle State != Currentness Assessment
latest version != governed current version
supersedes != deletes
Semantic Change Event != historical mutation
drift signal != Semantic Change Event automatically
monitoring threshold != governance decision
GEAIRA-M10-R03 - Semantic authority preservation: P7 SHALL consume GEAIRM, AIPS, BPS, PROV and adjacent-domain semantics without minting replacement meanings for Evaluation, Evidence, Assurance, Currentness, Version, Lifecycle, Risk Tier, Governance Obligation or Provenance.
GEAIRA-M10-R04 - Evaluation/assurance separation: An Evaluation or Evaluation Result SHALL NOT by itself establish an Assurance Finding, governance approval, Authorization, Product qualification or Risk Tier.
GEAIRA-M10-R05 - Evidence restraint: Presence of telemetry, logs, traces, test outputs or reports SHALL NOT by itself establish that those resources are sufficient Evidence for a declared Evaluation or Assurance scope.
28.3 P7 architecture boundary
P7 is the primary architecture plane for:
- defining material observability/evidence requirements for enterprise-AI behavior and state;
- correlating occurrence-level evidence across P1-P6;
- coordinating Evaluation against explicit criteria;
- preserving Evaluation Results and report linkage;
- coordinating Assurance Assessments against declared assurance claims/scope;
- producing/propagating Assurance Findings;
- assessing currentness of semantics, evidence, dependencies, assurance and governance applicability;
- observing lifecycle/version state relevant to continued use;
- consuming Semantic Change Events and material change signals;
- identifying when re-evaluation, re-assurance, suspension or remediation is required;
- propagating findings/currentness/change feedback to affected planes;
- preserving historical evidence and traceability through evolution.
P7 does not own:
- standing purpose/authority - P1;
- context construction - P2;
- primitive eligibility/composition - P3;
- runtime AI Execution - P4;
- serving/Product exposure - P5;
- point-of-use Authorization/action control - P6;
- Productness or AI Product qualification - BPS/AIPS;
- the enterprise risk taxonomy or Governance Obligation Profile - AIPS/applicable governance authority;
- generic provenance semantics - W3C PROV;
- every telemetry agent, log collector, dashboard or monitoring technology.
GEAIRA-M10-R06 - Plane boundary: P7 SHALL own the principal architecture responsibilities for evaluation, assurance support, observability/evidence correlation, currentness and change feedback while leaving operational execution and governance authorities in their respective planes/specifications.
GEAIRA-M10-R07 - No centralized-observability mandate: P7 SHALL NOT require all telemetry/evidence to be physically centralized; evidence MAY remain distributed provided identity, correlation, accessibility, integrity and applicable retention requirements are satisfied.
GEAIRA-M10-R08 - No governance substitution: P7 findings SHALL NOT directly substitute for P1 governance decisions or P6 Authorization unless an applicable governing rule explicitly binds the finding to that consequence.
28.4 Canonical P7 Architectural Responsibilities
M10 freezes the following sixteen canonical P7 Architectural Responsibilities.
| ID | Responsibility | Applicability | Required architectural outcome |
|---|---|---|---|
| P7-RESP-01 | Observability Requirement Determination | Baseline-mandatory | Determine which behavior, state, decisions, effects, dependencies and changes must be observable for the applicable context. |
| P7-RESP-02 | Evidence Capture & Correlation | Baseline-mandatory | Preserve/refer to sufficient material evidence with identity and correlation across relevant P1-P6 occurrences. |
| P7-RESP-03 | Evidence Integrity & Provenance Association | Conditional-mandatory | Preserve applicable origin, integrity, custody/provenance and transformation references needed to assess evidence fitness. |
| P7-RESP-04 | Evaluation Target & Criterion Binding | Conditional-mandatory | Bind each Evaluation to an explicit target, criteria, scope and context. |
| P7-RESP-05 | Evaluation Coordination | Conditional-mandatory | Coordinate or consume an Evaluation using declared methods/evidence without conflating it with Assurance. |
| P7-RESP-06 | Evaluation Result Management | Conditional-mandatory | Preserve Evaluation Results, their scope, criteria, method/context and report linkage where applicable. |
| P7-RESP-07 | Assurance Scope & Claim Binding | Conditional-mandatory | Define the target, property/claim, conditions and evidence expectations of an Assurance Assessment. |
| P7-RESP-08 | Assurance Assessment | Conditional-mandatory | Assess whether the declared assurance scope is sufficiently supported by applicable evidence. |
| P7-RESP-09 | Assurance Finding Management | Conditional-mandatory | Preserve and propagate findings including sufficient/insufficient/conditional/indeterminate outcomes. |
| P7-RESP-10 | Currentness Assessment | Baseline-mandatory where currentness materially affects use | Assess whether semantics, evidence, dependencies, governance or assurance remain applicable for the stated current purpose. |
| P7-RESP-11 | Lifecycle & Version State Observation | Baseline-mandatory | Observe relevant version/lifecycle/supersession state without equating latest with current. |
| P7-RESP-12 | Semantic Change Intake & Impact Triggering | Conditional-mandatory | Consume Semantic Change Events/material changes and identify potentially affected architecture targets. |
| P7-RESP-13 | Operational Drift / Performance / Safety Signal Intake | Conditional-mandatory | Consume operational signals that may justify Evaluation, Assurance or Currentness reassessment. |
| P7-RESP-14 | Re-evaluation / Re-assurance Triggering | Conditional-mandatory | Trigger or request reassessment when change, drift, expiry or findings invalidate prior assumptions. |
| P7-RESP-15 | Constraint / Finding Feedback Propagation | Conditional-mandatory | Propagate current findings/status to affected P1-P6 responsibilities with scope and effective time. |
| P7-RESP-16 | Historical Evidence & Assessment Preservation | Baseline-mandatory | Preserve what was observed, evaluated, assured and considered current at the relevant historical time without mutation by later findings. |
GEAIRA-M10-R09 - Responsibility completeness: A conforming P7 realization SHALL address each applicable P7 responsibility explicitly; delegation to a shared platform or external assurance function SHALL NOT erase traceability.
GEAIRA-M10-R10 - Observability proportionality: P7-RESP-01 SHALL establish observability proportionate to materiality, consequence, risk, assurance and governance needs rather than requiring indiscriminate collection of all possible telemetry.
GEAIRA-M10-R11 - Correlation integrity: P7-RESP-02 SHALL preserve sufficient identifiers/correlation to reconstruct material relationships among serving request, AI Execution, context/primitive bindings, Decision/Authorization/action, results/effects and relevant governance state where those elements participated.
GEAIRA-M10-R12 - Evaluation explicitness: P7-RESP-04/05 SHALL make Evaluation target, criteria and scope explicit enough that an Evaluation Result is not detached from the conditions under which it was produced.
GEAIRA-M10-R13 - Assurance scope explicitness: P7-RESP-07/08 SHALL preserve the declared assurance scope and SHALL NOT generalize a scoped Assurance Finding beyond the target, property, evidence and conditions actually assessed.
GEAIRA-M10-R14 - Currentness concern specificity: P7-RESP-10 SHALL identify the concern/category of currentness being assessed; a target MAY be current for one concern and non-current/indeterminate for another.
GEAIRA-M10-R15 - Change-trigger non-equivalence: P7-RESP-12/13 SHALL treat change/drift signals as triggers for assessment where applicable and SHALL NOT automatically classify every signal as a GEAIRM Semantic Change Event.
GEAIRA-M10-R16 - Historical preservation: P7-RESP-16 SHALL prevent later supersession, currentness or assurance conclusions from rewriting earlier evidence/assessment facts.
28.5 P7 observability and evidence model
P7 distinguishes observation, telemetry, Evidence, and proof.
runtime / serving / decision / context / primitive state
|
v
observation / telemetry
|
applicability + integrity + scope
|
v
resource playing
Evidence role
|
v
Evaluation / Assurance
A telemetry record becomes usable Evidence only in a declared context when its identity, relevance, integrity, provenance/currentness and relationship to the target are sufficient for the assessment being performed.
Evidence MAY include, where applicable:
- execution traces;
- model/primitive identifiers and versions;
- context/source bindings;
- serving-contract/request correlation;
- Decision/Authorization/human-oversight records;
- output/effect observations;
- evaluation datasets/results;
- safety/intervention events;
- source/version/currentness records;
- external reports/certifications;
- governed human observations.
GEAIRA-M10-R17 - Telemetry/evidence distinction: P7 SHALL NOT classify every log, metric, trace or event as Evidence without an applicable assessment context.
GEAIRA-M10-R18 - Evidence/proof distinction: Evidence SHALL support, challenge, qualify or justify a claim/assessment but SHALL NOT be treated as automatic proof of correctness, safety, compliance or fitness.
GEAIRA-M10-R19 - Distributed evidence: P7 MAY reference evidence held by authoritative external systems rather than copy it, provided the evidence can be resolved and its integrity/currentness can be assessed when required.
GEAIRA-M10-R20 - Observation gap visibility: Where required evidence or observability is absent, delayed or uncorrelatable, P7 SHALL expose the gap/indeterminacy rather than fabricate a successful assessment state.
28.6 Evaluation model
GEAIRM defines Evaluation as a context-bounded assessment against explicit criteria. P7 preserves that discipline.
Evaluation Target
+
Evaluation Criterion(s)
+
method / context
+
Evidence
|
v
Evaluation
|
v
Evaluation Result(s)
|
+--> AIPS AI Evaluation Report where applicable
Evaluation MAY concern, for example:
- capability performance;
- primitive behavior;
- context quality/grounding behavior;
- serving/integration behavior;
- decision/action outcomes;
- safety properties;
- robustness;
- fairness or bias under an applicable governance method;
- latency/reliability;
- human-oversight effectiveness;
- a declared Product/AI Product property.
GEAIRA does not define universal evaluation criteria. Criteria remain profile/domain/governance dependent.
GEAIRA-M10-R21 - Criterion binding: Every governed Evaluation SHALL identify the criteria actually applied and SHALL NOT infer evaluation coverage from the target name alone.
GEAIRA-M10-R22 - Result context: An Evaluation Result SHALL remain bound to its target/version, criteria, method/context and relevant evidence/currentness where material.
GEAIRA-M10-R23 - No universal threshold: GEAIRA SHALL NOT define a universal score/threshold that establishes fitness, safety, assurance or AI Product qualification across all enterprise-AI contexts.
GEAIRA-M10-R24 - Reproducibility proportionality: Where reproducibility is a material requirement, P7 SHALL preserve sufficient evaluation configuration/evidence references to permit governed reconstruction or justified equivalence; GEAIRA SHALL NOT require identical reruns where the method/context inherently prevents them.
28.7 Assurance model
Assurance asks a different question from Evaluation:
Is there sufficient, current and applicable evidence to support the declared assurance claim/scope?
Evaluation Results
operational evidence
provenance/currentness
control evidence
human/governance evidence
external evidence
|
v
Assurance Assessment
|
v
Assurance Finding
An Assurance Finding MAY be:
- sufficient;
- insufficient;
- conditional;
- partially supported;
- not applicable;
- indeterminate;
- expired/non-current.
The exact controlled vocabulary remains profile/authority specific unless externally standardized.
GEAIRA-M10-R25 - Evaluation/assurance non-collapse: P7 SHALL NOT treat a passing Evaluation Result as automatic assurance; Assurance Assessment SHALL consider the declared assurance scope and applicable evidence sufficiency/currentness.
GEAIRA-M10-R26 - Assurance non-authorization: An Assurance Finding SHALL NOT by itself create P1 authority, P6 Authorization, legal approval or permission to act.
GEAIRA-M10-R27 - Assurance/risk separation: P7 SHALL NOT infer AIPS Effective Risk Tier or Governance Obligation Profile directly from an Assurance Finding unless the authoritative AIPS/governance process explicitly defines that derivation.
GEAIRA-M10-R28 - Negative/indeterminate findings: Insufficient or indeterminate assurance SHALL be representable explicitly and SHALL NOT be coerced into a binary PASS solely for operational convenience.
28.8 Currentness, lifecycle and version model
P7 treats currentness as time- and concern-qualified applicability, not as simple freshness.
Examples:
Model/primitive version may be technically available
but assurance evidence is non-current.
Policy may be recently published
but not yet applicable to this Product Version.
Dataset may be fresh
while its provenance claim is indeterminate.
Evaluation Result may remain historically valid
but no longer sufficient for current deployment conditions.
The canonical distinction is:
Version identity
!= lifecycle state
!= currentness status
!= freshness timestamp
GEAIRA-M10-R29 - Latest/current distinction: P7 SHALL NOT equate the latest identifiable version with the governed current version for every purpose.
GEAIRA-M10-R30 - Lifecycle/currentness separation: A Lifecycle State SHALL NOT by itself establish currentness; P7 SHALL assess currentness concerns independently where material.
GEAIRA-M10-R31 - Concern-qualified currentness: Currentness conclusions SHALL identify the applicable concern/scope where ambiguity would affect governance or operation.
GEAIRA-M10-R32 - Supersession preservation: supersedes SHALL identify replacement for a declared current purpose without implying deletion, invalidation of historical use, or mutation of the superseded resource.
28.9 Semantic change and evolution model
P7 consumes Semantic Change Events and other change signals that may invalidate architectural assumptions.
Examples include changes to:
- Product/AI Product constitution or qualification;
- primitive dependencies;
- capability contribution claims;
- context/source meaning;
- policy/authority basis;
- serving contract meaning;
- agent/tool behavior;
- risk/governance obligations;
- evaluation criteria/methods;
- assurance evidence;
- lifecycle/currentness state.
The generic evolution path is:
change observed / declared
|
v
impact candidate identification
|
v
currentness assessment
|
+----+------------------+
| |
v v
still applicable re-evaluation /
re-assurance /
constrain / suspend /
migrate / remediate
GEAIRA-M10-R33 - Change impact restraint: A Semantic Change Event SHALL identify potentially affected targets without assuming that every dependent target is automatically invalid.
GEAIRA-M10-R34 - Reassessment trigger: Where a material change invalidates assumptions supporting an Evaluation, Assurance Finding, currentness conclusion or operational eligibility, P7 SHALL trigger or request applicable reassessment.
GEAIRA-M10-R35 - No historical mutation: Evolution handling SHALL preserve historical versions, evidence, evaluations and findings even when they are no longer current for future use.
GEAIRA-M10-R36 - Dependency-aware evolution: P7 SHOULD use available dependency/traceability information to identify potentially affected downstream targets without inferring identity inheritance or transitive semantic equivalence.
28.10 Canonical P7 ABBs
Applying the M02 ABB admission criteria, M10 admits four canonical logical P7 ABBs.
P7-ABB-01 - Observability & Evidence Boundary
Purpose: provide a coherent architecture boundary for material observability requirements, evidence identity/correlation, evidence integrity/provenance association and historical traceability.
Principal responsibilities:
- P7-RESP-01 Observability Requirement Determination;
- P7-RESP-02 Evidence Capture & Correlation;
- P7-RESP-03 Evidence Integrity & Provenance Association;
- portions of P7-RESP-16 Historical Evidence Preservation.
This ABB is not a mandatory log platform, SIEM, telemetry lake or observability vendor stack.
P7-ABB-02 - Evaluation Boundary
Purpose: provide the independently governable boundary around Evaluation target/criterion binding, evaluation coordination and Evaluation Results.
Principal responsibilities:
- P7-RESP-04 Evaluation Target & Criterion Binding;
- P7-RESP-05 Evaluation Coordination;
- P7-RESP-06 Evaluation Result Management.
This ABB is not a mandatory test harness, benchmark service or model-evaluation platform.
P7-ABB-03 - Assurance & Currentness Boundary
Purpose: provide the independently governable boundary around assurance scope, evidence sufficiency, Assurance Findings and currentness assessment.
Principal responsibilities:
- P7-RESP-07 Assurance Scope & Claim Binding;
- P7-RESP-08 Assurance Assessment;
- P7-RESP-09 Assurance Finding Management;
- P7-RESP-10 Currentness Assessment;
- P7-RESP-11 Lifecycle & Version State Observation.
This ABB is not a mandatory assurance service, governance dashboard or certification engine.
P7-ABB-04 - Evolution & Change Feedback Boundary
Purpose: provide a coherent boundary for material change intake, drift/safety/performance signals, reassessment triggers and propagation of current findings/constraints.
Principal responsibilities:
- P7-RESP-12 Semantic Change Intake & Impact Triggering;
- P7-RESP-13 Operational Signal Intake;
- P7-RESP-14 Re-evaluation / Re-assurance Triggering;
- P7-RESP-15 Constraint / Finding Feedback Propagation;
- portions of P7-RESP-16 historical preservation.
This ABB is not a mandatory model-drift service, change-data-capture platform or release manager.
GEAIRA-M10-R37 - ABB admission: The four P7 ABBs above are canonical logical responsibility boundaries because each has a cohesive independently governable architecture purpose and survives technology substitution.
GEAIRA-M10-R38 - ABB realization independence: P7 ABBs MAY be co-located, distributed, embedded in domain platforms or partly human-operated; co-location SHALL NOT erase their responsibility boundaries.
GEAIRA-M10-R39 - No tool-name promotion: A monitoring, logging, dashboard, model-evaluation, assurance, governance, provenance or drift-management product SHALL NOT become a canonical P7 ABB merely because it can implement part of a P7 responsibility.
28.11 Deliberately non-admitted P7 ABB candidates
The following are implementation patterns/technologies and are not generic canonical GEAIRA ABBs:
- Observability Platform;
- Logging Platform;
- Metrics Platform;
- Tracing Platform;
- SIEM;
- APM platform;
- ML Monitoring Platform;
- Model Monitoring Service;
- Drift Detection Service;
- Dashboard;
- Evidence Store;
- Evidence Lake;
- Audit Log Store;
- Provenance Database;
- Evaluation Harness;
- Benchmark Service;
- Model Evaluation Platform;
- Assurance Service;
- Certification Engine;
- Governance Dashboard;
- Currentness Service;
- Version Registry;
- Change Data Capture platform;
- Release Manager;
- Incident Management platform.
Any of these MAY realize one or more P7 responsibilities in a concrete implementation/profile.
GEAIRA-M10-R40 - Implementation non-normativity: GEAIRA SHALL NOT require the non-admitted candidates above unless a profile explicitly specializes P7 for a concrete environment.
28.12 P7 interaction model
M10 establishes the following technology-neutral P7 interactions.
| ID | Interaction | Source | Target | Purpose |
|---|---|---|---|---|
| P7-INT-01 | Governing Evaluation / Assurance Requirements | P1 / governance authority | P7 | Provide applicable criteria, obligations, claims or assurance scope. |
| P7-INT-02 | Context / Source Evidence | P2 | P7 | Provide source/context construction, grounding and currentness evidence. |
| P7-INT-03 | Primitive / Capability Evidence | P3 | P7 | Provide primitive/version, capability-contribution, eligibility and dependency evidence. |
| P7-INT-04 | Runtime Evidence / Outcome | P4 | P7 | Provide execution, context/primitive binding, Agent/Tool, Environment, effect and failure evidence. |
| P7-INT-05 | Serving / Integration Evidence | P5 | P7 | Provide consumer, contract, Product/version, provider and delivery evidence. |
| P7-INT-06 | Decision / Authorization / Oversight Evidence | P6 | P7 | Provide decision, authority, human-oversight and action-control evidence. |
| P7-INT-07 | External Evaluation / Assurance Evidence | external authority/evaluator | P7 | Consume authoritative external reports, certifications or findings. |
| P7-INT-08 | Governing / Assurance Feedback | P7 | P1 | Inform continued purpose/authority/governance decisions. |
| P7-INT-09 | Currentness / Eligibility Feedback | P7 | P2/P3/P5 | Constrain resource/primitive/contract eligibility or trigger refresh/migration. |
| P7-INT-10 | Safety / Continuation Feedback | P7 | P4/P6 | Constrain, suspend, stop or require additional control for future/ongoing operation where governed. |
| P7-INT-11 | Change / Reassessment Notification | P7 | affected P1-P6/external authority | Trigger re-evaluation, re-assurance, remediation or migration. |
Interactions MAY be synchronous, asynchronous, batch, event-driven, federated, embedded or human-mediated.
GEAIRA-M10-R41 - Interaction neutrality: P7 interactions SHALL remain independent of a mandatory protocol, observability stack, event bus, data lake or centralized assurance service.
GEAIRA-M10-R42 - Feedback scope: P7 feedback SHALL include sufficient target/scope/effective-time information to prevent a finding for one version/context from being indiscriminately applied to unrelated targets.
GEAIRA-M10-R43 - Control-path priority: Where a P7 finding indicates an applicable urgent safety, authority or assurance condition, the implementation SHALL support propagation with priority appropriate to the governing context, including local/embedded paths where central-network latency would be unsafe.
28.13 P7 control points
M10 establishes three logical P7 control-point classes.
CP7-01 - Evidence Sufficiency & Correlation Control Point
Determines whether required evidence is available, identifiable, current enough and sufficiently correlated to permit the intended Evaluation/Assurance activity.
CP7-02 - Evaluation / Assurance Disposition Control Point
Determines the governed disposition of Evaluation Results or Assurance Findings, including sufficient, insufficient, conditional, indeterminate, expired/non-current or escalation-required outcomes.
CP7-03 - Currentness / Change Response Control Point
Determines whether change/currentness findings require continued use, restricted use, re-evaluation, re-assurance, remediation, migration, suspension or escalation.
These are logical control responsibilities and may be implemented centrally, locally, distributedly, through human governance or within another component.
GEAIRA-M10-R44 - Control-point semantics: A P7 control point SHALL NOT be assumed to be a standalone service or network hop.
GEAIRA-M10-R45 - Fail-safe assessment state: Where required evidence/currentness is materially indeterminate, the control outcome SHALL preserve that indeterminacy and SHALL NOT fabricate a successful PASS/current state.
28.14 Failure and degradation model
P7 recognizes at least the following generic failure/degradation classes:
- required observability absent;
- evidence uncorrelatable to the target/occurrence;
- evidence integrity/provenance indeterminate;
- Evaluation Criterion ambiguous/non-current;
- evaluation method unavailable/inapplicable;
- Evaluation Result stale/out-of-scope;
- assurance scope mismatch;
- insufficient/contradictory evidence;
- currentness indeterminate;
- lifecycle/version state conflict;
- material change not yet reassessed;
- drift/performance/safety threshold breach requiring assessment;
- feedback propagation failure;
- historical evidence/assessment mutation attempt;
- external assurance dependency unavailable;
- observation blind spot causing unknown operational state.
Possible governed outcomes include:
continue with current assurance
continue conditionally
narrow scope
require additional evidence
re-evaluate
re-assure
remediate / migrate
suspend new use
stop / intervene via P6/P4
escalate
mark indeterminate
reconcile
GEAIRA-M10-R46 - Failure visibility: Material P7 failures/degradation SHALL be observable to the applicable governing/control boundary and SHALL NOT be silently translated into successful assurance/currentness.
GEAIRA-M10-R47 - Contradictory evidence: Materially contradictory evidence SHALL be preserved and surfaced for governed disposition rather than silently selecting the most favorable observation.
GEAIRA-M10-R48 - Feedback failure: Failure to deliver a material P7 finding to an affected control plane SHALL itself be treated as an observable condition where the finding could affect safety, authority, eligibility or assurance.
28.15 Cross-cutting overlays applied to P7
All eight M03 cross-cutting overlays apply to P7.
| Overlay | P7 architectural implication |
|---|---|
| X1 Security, Privacy & Trust | Protect sensitive telemetry/evidence, assessment integrity and trust boundaries; minimize unnecessary data capture. |
| X2 Identity, Authority & Policy Enforcement | Establish evaluator/assessor authority, evidence-access controls and current governance basis. |
| X3 Evidence, Provenance & Traceability | Preserve evidence origin, relationships, transformations, assessment lineage and historical truth. |
| X4 Observability, Telemetry & Auditability | Define proportionate observability and correlation without equating collection volume with assurance. |
| X5 Resilience, Safety & Failure Management | Detect/propagate safety and failure findings, support degraded/safe states and retain indeterminate outcomes. |
| X6 Lifecycle, Version, Currentness & Change | Track version/lifecycle/currentness and trigger reassessment after material change. |
| X7 Interoperability, Contracts & Semantic Compatibility | Preserve assessment/evidence semantics across tools/domains and prevent version/scope mismatches. |
| X8 Governance, Human Accountability & Compliance | Make assessment authority, human assurance decisions, escalation and compliance evidence explicit. |
GEAIRA-M10-R49 - Overlay continuity: P7 SHALL apply X1-X8 according to applicability and SHALL NOT absorb cross-cutting responsibilities from P1-P6 merely because P7 is the primary evaluation/observability plane.
28.16 Representative P7 scenarios
Scenario 1 - Primitive evaluation affects future eligibility
A P3 primitive version is evaluated against declared performance/safety criteria. Evaluation Results contribute to an Assurance Assessment. The resulting conditional finding causes P3 to restrict future eligibility while historical P4 executions using that version remain unchanged.
Scenario 2 - Runtime anomaly triggers reassessment
P4 reports an unexpected effect/failure pattern. P7 correlates the runtime evidence, records the signal, performs or requests Evaluation, and may mark prior assurance/currentness as needing reassessment. P6 may receive a safety/continuation constraint if required.
Scenario 3 - Context source supersession
A Knowledge Resource used in earlier P2 contexts is superseded. P7 assesses whether the change affects current grounding/evidence assumptions. Future P2 eligibility may change; earlier executions still truthfully reference the older source version.
Scenario 4 - Serving contract semantic change
P5 introduces a new contract version with materially different output semantics. P7 treats the change as a reassessment trigger for applicable evaluation/assurance rather than assuming protocol compatibility proves semantic compatibility.
Scenario 5 - Human oversight evidence contributes to assurance
P6 provides evidence that required human approvals occurred with suitable authority/context. That evidence may support an Assurance Assessment, but its presence does not prove every decision was correct or safe.
Scenario 6 - Non-Product internal AI capability
An internal AI Execution that is not a BPS Product still emits runtime/context/primitive evidence. P7 evaluates and assures relevant enterprise properties without inventing AI Product assurance semantics or Risk Tier.
Scenario 7 - Physical-AI safety degradation
P7 receives local P4 control-loop safety signals indicating material degradation. A currentness/assurance finding triggers a high-priority constraint to P6/P4. The architecture permits local safety enforcement rather than requiring a remote central P7 call before stopping unsafe operation.
28.17 P7 anti-patterns
M10 prohibits or warns against the following generic anti-patterns:
- telemetry-equals-evidence - treating every collected signal as usable Evidence.
- evidence-equals-proof - treating evidence presence as proof of correctness/compliance/safety.
- evaluation-equals-assurance - treating a test result as a complete assurance conclusion.
- assurance-equals-approval - converting an Assurance Finding into permission/authority without governance rule.
- risk-tier-equals-assurance - treating classification and assurance as interchangeable.
- dashboard-equals-observability - assuming a visualization proves required state is observable/correlated.
- log-equals-provenance - treating logging as complete provenance.
- monitor-everything - indiscriminate telemetry collection regardless of materiality/privacy/cost.
- central-observability-everything - forcing all evidence through one centralized platform.
- pass-threshold-universality - using one generic threshold to establish enterprise-AI fitness across contexts.
- latest-equals-current - selecting latest version as governed current by default.
- freshness-equals-currentness - using age alone as currentness.
- supersession-equals-deletion - erasing prior evidence/resources after replacement.
- drift-equals-semantic-change - treating any metric drift as a Semantic Change Event automatically.
- currentness-rewrites-history - altering historical evaluation/assurance records after later change.
- assurance-service-everything - collapsing evaluation, evidence, assurance, currentness and change into one opaque component.
- feedback-equals-mutation - directly mutating source-plane truth instead of propagating governed findings/constraints.
- post-hoc-coverage-inflation - claiming an assessment covered targets/criteria that were not actually in scope.
28.18 P7 lifecycle and historical truth
P7 preserves assessment history as a temporal record.
Example:
Evaluation EVAL-17 at T1
target: Primitive v3
result: acceptable under Criterion Set C2
Assurance Assessment AS-09 at T1
finding: sufficient for Scope S1
later at T2:
Primitive v4 supersedes v3
Criterion Set C3 replaces C2
new evidence shows a broader risk
At T2, P7 may conclude that:
- v3 is no longer current for new use;
- AS-09 is no longer sufficient for the current purpose;
- re-evaluation/re-assurance is required.
But P7 must not rewrite the historical facts that EVAL-17 and AS-09 occurred with their stated scope/evidence at T1.
GEAIRA-M10-R50 - Historical assessment truth: Later evidence, version, criterion, policy or assurance changes SHALL NOT rewrite what an earlier Evaluation/Assurance Assessment concluded under its actual historical scope and evidence.
GEAIRA-M10-R51 - Present applicability distinction: P7 SHALL distinguish historical validity of an assessment occurrence from whether that assessment remains sufficient/current for a present purpose.
28.19 M10 frozen decisions
M10 freezes the following P7 architecture decisions for GEAIRA 0.1:
P7 is the primary Evaluation, Assurance, Observability & Evolution plane.
Observation != Evaluation.
Telemetry != Evidence automatically.
Evidence != proof.
Evaluation != Assurance.
Evaluation Result != Assurance Finding.
Assurance != approval / Authorization / Risk Tier.
Currentness != freshness.
Lifecycle != currentness.
latest != governed current.
supersedes != deletion.
Semantic Change != historical mutation.
Drift/change signals may trigger reassessment without automatically becoming Semantic Change Events.
P7 feedback constrains future/current operation; it does not rewrite historical source-plane truth.
P7 observability/evidence may be distributed.
P7 does not require one centralized monitoring/assurance platform.
The following P7 ABBs are canonical:
P7-ABB-01 Observability & Evidence Boundary
P7-ABB-02 Evaluation Boundary
P7-ABB-03 Assurance & Currentness Boundary
P7-ABB-04 Evolution & Change Feedback Boundary
M10 does not freeze:
- universal evaluation criteria;
- universal assurance thresholds;
- one evidence schema for all enterprises;
- one telemetry/observability platform;
- one Evaluation or model-monitoring product;
- one Assurance Service;
- one provenance database;
- one currentness service;
- one drift algorithm;
- Productness / AI Product qualification;
- AIPS Risk Tier / Governance Obligation semantics;
- final M11 inter-plane interaction contracts.
GEAIRA-M10-R52 - Downstream boundary preservation: Later slices SHALL preserve the Evaluation/Assurance, telemetry/Evidence, currentness/freshness and feedback/historical-truth distinctions unless a reconciled 0.1 architecture change explicitly requires otherwise.
GEAIRA-M10-R53 - Profileability: Later profiles MAY strengthen evaluation frequency, evidence integrity, observability, assurance, retention, currentness, safety or human-review requirements but SHALL NOT weaken mandatory P7 invariants applicable to the profiled context.
GEAIRA-M10-R54 - No Product inference: Presence of evaluation, assurance, observability, lifecycle or change architecture SHALL NOT establish Productness or AI Product qualification.
GEAIRA-M10-R55 - Architecture diversity continuity: P7 SHALL remain valid across Product/non-Product AI, learned/symbolic/hybrid mechanisms, centralized/distributed systems, human-led assurance and embedded/physical-AI environments without structural redesign solely because the technology differs.
GEAIRA-M10-R56 - M11 hand-off readiness: M11 SHALL formalize inter-plane interaction/contract semantics using the P1-P7 responsibilities and interactions frozen through M10 without collapsing the planes into a mandatory linear runtime pipeline.
29. M10 Completion State
GEAIRA-M10 - COMPLETE.
M10 establishes:
- one canonical P7 purpose and Evaluation/Assurance/Observability/Evolution boundary;
- sixteen canonical P7 Architectural Responsibilities;
- four canonical logical P7 ABBs;
- twenty-five deliberately non-admitted implementation-style ABB candidates;
- eleven technology-neutral P7 architectural interactions;
- three logical P7 control-point classes;
- sixteen generic P7 failure/degradation classes;
- explicit application of all eight X1-X8 cross-cutting overlays;
- explicit GEAIRM/AIPS/BPS/PROV and adjacent-architecture boundaries;
- seven representative P7 scenarios;
- eighteen P7 anti-patterns;
- 56 normative M10 rules.
M10 introduces no change to GEAIRM, BPS, AIPS, Agent Ontology or PROV semantics, does not define universal evaluation/assurance thresholds, and does not prescribe a telemetry platform, evidence store, model-monitoring system, assurance service, provenance database or centralized control topology.
Next slice
GEAIRA-M11 - Inter-Plane Interaction & Contract Model
M11 will consolidate P1-P7 hand-offs into a canonical technology-neutral interaction/contract model, defining contract content, identity/correlation, authority/evidence propagation, synchronous/asynchronous semantics, failure/indeterminate outcomes and cross-plane evolution without turning the seven planes into a mandatory sequence.
30. GEAIRA-M11 - Inter-Plane Interaction & Contract Model
30.1 Purpose and architectural boundary
M04-M10 froze the seven GEAIRA planes and their plane-specific interaction responsibilities. M11 reconciles those declarations into one canonical cross-plane model.
M11 answers:
How do GEAIRA planes exchange governing intent, context, mechanism realization, execution requests/results, decision/action controls, evidence and change signals without becoming a mandatory linear pipeline or a protocol-specific integration architecture?
M11 therefore freezes:
- canonical inter-plane interaction families and interaction identities;
- a technology-neutral Interaction Contract model;
- identity, causality and correlation rules;
- authority and policy propagation rules;
- evidence/provenance propagation rules;
- synchronous, asynchronous, embedded and human-mediated interaction semantics;
- outcome, failure, timeout and indeterminate-state rules;
- version/currentness and compatibility rules;
- cross-plane change/revocation/reassessment behavior;
- boundary control-point responsibilities;
- interaction anti-patterns.
M11 does not define:
- REST, gRPC, GraphQL, Kafka, MQTT, MCP, AMQP or any other mandatory protocol;
- one enterprise event bus, API gateway or service mesh;
- one canonical payload schema for every plane;
- a mandatory P1 -> P2 -> P3 -> P4 -> P5 -> P6 -> P7 runtime sequence;
- external peer-architecture contracts with GEDRA/GEARA/AIPDS/HDIP in detail; these are addressed in M12;
- deployment topology, addressed in M13;
- semantic meaning owned by GEAIRM, BPS, AIPS, Agent Ontology or PROV.
GEAIRA-M11-R01 - Contract-model scope: M11 SHALL define technology-neutral architectural interaction contracts and SHALL NOT prescribe a transport, deployment topology or vendor integration stack.
GEAIRA-M11-R02 - No linear-pipeline inference: The P1-P7 plane numbering SHALL NOT be interpreted as a mandatory runtime processing sequence.
GEAIRA-M11-R03 - Source-plane preservation: M11 SHALL reconcile the plane-specific interaction responsibilities frozen in M04-M10 without silently deleting their architectural intent.
30.2 Interaction Contract definition
An Interaction Contract is the technology-neutral architectural agreement governing a material hand-off between architectural responsibilities.
An Interaction Contract specifies the meaning and obligations of the hand-off independently of whether the realization is:
- in-process;
- synchronous remote invocation;
- asynchronous event/message exchange;
- stream;
- batch;
- shared governed state;
- file/artifact exchange;
- embedded/local control;
- human-mediated workflow;
- federated exchange;
- a combination of these.
An Interaction Contract is not automatically an API, topic, queue, table, event schema, service, endpoint or product.
GEAIRA-M11-R04 - Logical contract independence: A material architectural hand-off SHALL retain an explicit logical contract even when producer and consumer responsibilities are co-located in the same deployable component.
GEAIRA-M11-R05 - API non-equivalence: An API or message schema MAY realize part of an Interaction Contract but SHALL NOT be treated as the complete architectural contract solely because its syntax is documented.
GEAIRA-M11-R06 - Shared-state discipline: Shared-state realization MAY be used, but the architecture SHALL preserve ownership, meaning, version/currentness, access/control and failure semantics rather than relying on undocumented database coupling.
30.3 Canonical contract-content model
A GEAIRA Interaction Contract SHOULD define the following dimensions where material:
| Dimension | Architectural question |
|---|---|
| Contract identity | Which logical interaction contract and contract version applies? |
| Purpose | What architectural outcome does the hand-off enable? |
| Producer responsibility | Which plane/responsibility is authoritative for producing the hand-off? |
| Consumer responsibility | Which plane/responsibility consumes or acts on it? |
| Semantic references | Which GEAIRM/AIPS/BPS/Agent/PROV concepts give the exchanged information meaning? |
| Applicability | Under which use, Product, profile, risk, environment or execution conditions does the contract apply? |
| Governing envelope | Which purpose, authority, policy, constraint or obligation references govern the exchange? |
| Identity/correlation | Which request, execution, context, primitive, Product/version, decision, authorization, action or evidence identities must remain traceable? |
| Content obligations | Which information is required, conditional or optional? |
| Temporal semantics | Effective time, event/occurrence time, expiry, currentness and ordering expectations. |
| Interaction mode | Synchronous, asynchronous, stream, batch, embedded, human-mediated or other allowed form. |
| Delivery/processing semantics | What acknowledgement, acceptance, completion, duplicate or replay behavior is meaningful? |
| Outcome semantics | How success, rejection, deferment, partial completion, failure or indeterminate state is represented. |
| Idempotency/side-effect semantics | Whether repetition is safe and how duplicate/unknown side effects are reconciled. |
| Evidence/provenance | Which trace/evidence is retained and how source identity is preserved. |
| Security/privacy | Authentication, confidentiality, minimization, trust-boundary and access obligations where applicable. |
| Compatibility/evolution | How contract versions, semantic changes, deprecation and migration are governed. |
| Failure/degradation | What happens if the producer, consumer, governing state or evidence is unavailable/indeterminate? |
| Human obligations | Where human judgment, approval, notification or escalation forms part of the contract. |
The table is a conceptual contract model, not a mandatory serialized document format.
GEAIRA-M11-R07 - Contract completeness: A material Interaction Contract SHALL state enough information to distinguish semantic meaning, authority, identity/correlation, outcome and change behavior for its architecture context.
GEAIRA-M11-R08 - Minimality: Profiles MAY omit dimensions that are demonstrably not applicable, but omission SHALL NOT erase a material control, identity, evidence, currentness or failure obligation.
GEAIRA-M11-R09 - Semantic-reference preservation: Interaction payloads SHALL reference or preserve authoritative semantic identities where material rather than invent locally equivalent shadow concepts.
30.4 Canonical interaction families
M11 organizes the canonical catalogue into six families.
| Family | Purpose |
|---|---|
| F1 Governance & Authority | Propagate governing purpose, authority, policy, feedback and revocation. |
| F2 Context & Knowledge | Request/provide bounded context and communicate context evidence/currentness. |
| F3 Capability & Primitive | Express capability needs, provide primitive realization and communicate eligibility/change. |
| F4 Serving & Execution | Correlate consumer-facing serving with occurrence-level AI Execution. |
| F5 Decision & Action Control | Request/control consequential decisions/actions and propagate outcomes/interventions. |
| F6 Evidence, Assurance & Evolution | Supply plane evidence and propagate assurance/currentness/change feedback. |
These families organize interaction responsibilities only; they are not deployable buses or integration layers.
GEAIRA-M11-R10 - Family non-deployment: An interaction family SHALL NOT be interpreted as requiring a dedicated integration product, gateway, broker or service.
30.5 Canonical inter-plane interaction catalogue
The following 33 interactions are canonical for the GEAIRA 0.1 human-readable architecture baseline.
F1 - Governance & Authority
| ID | Interaction | Direction | Principal purpose |
|---|---|---|---|
| M11-INT-01 | Governing Context Constraints | P1 -> P2 | Constrain eligible context/resources and context construction. |
| M11-INT-02 | Governing Capability Constraints | P1 -> P3 | Constrain capability/primitive eligibility and realization. |
| M11-INT-03 | Governing Execution Envelope | P1 -> P4 | Supply purpose, authority and execution constraints. |
| M11-INT-04 | Governing Serving Constraints | P1 -> P5 | Constrain exposure, consumer use and integration. |
| M11-INT-05 | Governing Authority / Policy Basis | P1 -> P6 | Supply authority/delegation/policy basis for decision/action control. |
| M11-INT-06 | Governing Evaluation / Assurance Requirements | P1 -> P7 | Supply applicable evaluation, evidence, assurance and governance requirements. |
| M11-INT-07 | Product / Risk Governance Association | P5 -> P1 | Associate authoritative Product-Version/AIPS governance context where valid. |
| M11-INT-08 | Governing / Assurance Feedback | P7 -> P1 | Inform continued purpose, authority and governance decisions. |
| M11-INT-09 | Governing Revocation / Supersession | P1 -> affected P2-P7 | Propagate material change in governing purpose, authority or constraints. |
F2 - Context & Knowledge
| ID | Interaction | Direction | Principal purpose |
|---|---|---|---|
| M11-INT-10 | Context Requirement | P4/P5 or governed caller -> P2 | State bounded context requirements for an execution/consumption path. |
| M11-INT-11 | AI Context Provision | P2 -> P4 | Provide bounded AI Context and relevant source/currentness metadata. |
| M11-INT-12 | Decision Context Contribution | P2 -> P6 | Supply context/resources relevant to Decision Formation or control. |
| M11-INT-13 | Context / Source Evidence | P2 -> P7 | Supply context-construction, grounding/source and currentness evidence. |
| M11-INT-14 | Context Currentness Feedback | P7 -> P2 | Constrain future source/context eligibility or trigger refresh. |
| M11-INT-15 | Context Invalidation / Refresh | P2 -> P4/P5 | Notify affected execution/serving responsibilities of invalidated prepared context. |
F3 - Capability & Primitive
| ID | Interaction | Direction | Principal purpose |
|---|---|---|---|
| M11-INT-16 | Capability Requirement | P1/P4/P5 -> P3 | State required Intelligence Capability and governing constraints. |
| M11-INT-17 | Primitive Realization Provision | P3 -> P4 | Provide eligible primitive/composition/version realization information. |
| M11-INT-18 | Primitive / Capability Evidence | P3/P4 -> P7 | Supply primitive/version, capability-contribution and actual-use evidence. |
| M11-INT-19 | Primitive Assurance / Fitness Feedback | P7 -> P3 | Alter future primitive eligibility, constraints or reassessment requirements. |
| M11-INT-20 | Primitive Change / Invalidation | P3 -> P4/P5/P7 | Propagate retirement, supersession or eligibility change. |
F4 - Serving & Execution
| ID | Interaction | Direction | Principal purpose |
|---|---|---|---|
| M11-INT-21 | Execution Invocation / Serving Request | P5 -> P4 | Initiate or request an AI Execution through a serving/integration boundary. |
| M11-INT-22 | Execution Result / Effect Return | P4 -> P5 | Return execution outcome/effect/status for governed delivery to the consumer. |
F5 - Decision & Action Control
| ID | Interaction | Direction | Principal purpose |
|---|---|---|---|
| M11-INT-23 | Decision / Action Control Request | P4/P5 -> P6 | Present proposed consequential decision, invocation or action for control. |
| M11-INT-24 | Decision Input / Intelligence Effect | P4/P5 -> P6 | Supply recommendation, plan, score or other effect that may inform Decision Formation. |
| M11-INT-25 | Authorization / Decision Outcome | P6 -> P4/P5 | Return permitted, rejected, conditional, deferred or indeterminate control outcome. |
| M11-INT-26 | Authorized Action Commitment | P6 -> P4/P5 | Hand off an authorized/committed action for technical enactment where applicable. |
| M11-INT-27 | Action / Decision Outcome Evidence | P4/P5 -> P6 | Return observed success/failure/indeterminate outcome for reconciliation. |
| M11-INT-28 | Intervention / Revocation / Stop | P6 -> P4/P5 | Propagate stop, override, revocation or intervention controls. |
F6 - Evidence, Assurance & Evolution
| ID | Interaction | Direction | Principal purpose |
|---|---|---|---|
| M11-INT-29 | Runtime Evidence / Outcome | P4 -> P7 | Supply execution, context/primitive binding, Agent/Tool, Environment and effect evidence. |
| M11-INT-30 | Serving / Integration Evidence | P5 -> P7 | Supply consumer, contract, Product/version, provider and delivery evidence. |
| M11-INT-31 | Decision / Authorization / Oversight Evidence | P6 -> P7 | Supply decision, Authorization, human-oversight and action-control evidence. |
| M11-INT-32 | Safety / Currentness / Continuation Feedback | P7 -> P4/P5/P6 | Constrain, suspend, migrate or require additional control for ongoing/future operation. |
| M11-INT-33 | Change / Reassessment Notification | P7 -> affected P1-P6 | Trigger re-evaluation, re-assurance, remediation, migration or governance review. |
M11 does not duplicate plane-internal interactions such as P2 retrieval-to-context assembly or P7 evaluation-to-assurance hand-offs in the inter-plane catalogue; those remain valid within their source plane.
GEAIRA-M11-R11 - Canonical interaction freeze: M11-INT-01 through M11-INT-33 are the canonical GEAIRA 0.1 inter-plane interaction identities for the human-readable baseline.
GEAIRA-M11-R12 - Plane-interaction reconciliation: Plane-specific M04-M10 interaction declarations SHALL map to one or more M11 canonical interactions or remain explicitly plane-internal/external-boundary interactions.
GEAIRA-M11-R13 - No synthetic interaction inflation: Later slices SHALL NOT create new inter-plane interaction identities merely to reflect a different protocol, deployment technology or vendor product.
30.5.1 Reconciliation of M04-M10 source interactions
M11 explicitly reconciles all 70 plane-level interaction declarations frozen in M04-M10. A source interaction may map to a canonical M11 inter-plane interaction, remain plane-internal, or remain a deliberate external/human/environment boundary for M12 or its originating plane to elaborate.
| Source interaction | Source label | Source direction | M11 disposition |
|---|---|---|---|
P1-INT-01 | Intent / Purpose Establishment | External authoritative source -> P1 | Retained external ingress; M12/profile boundary |
P1-INT-02 | Authority / Accountability Binding | External authoritative source -> P1 | Retained external ingress; M12/profile boundary |
P1-INT-03 | Policy / Obligation Projection | External/AIPS/governance -> P1 | Retained external ingress; M12 profile/authority mapping |
P1-INT-04 | Product / Risk Governance Association | P5/AIPS -> P1 | M11-INT-07 plus M12 for external AIPS authority |
P1-INT-05 | Governing Envelope Provision | P1 -> P2/P3/P4/P5/P6/P7 | M11-INT-01..06 |
P1-INT-06 | Authority Basis Provision to P6 | P1 -> P6 | M11-INT-05 |
P1-INT-07 | Assurance / Currentness Feedback | P7 -> P1 | M11-INT-08 |
P1-INT-08 | Revocation / Supersession Notification | P1 -> affected planes | M11-INT-09 |
P2-INT-01 | Governing Context Constraints | P1 -> P2 | M11-INT-01 |
P2-INT-02 | Context Requirement / Request | P4/governed caller -> P2 | M11-INT-10 |
P2-INT-03 | Knowledge / Resource Access | P2 -> external/peer resource authority | Retained external/peer boundary; detailed in M12 |
P2-INT-04 | Retrieval Result Contribution | P2-ABB-02 -> P2-ABB-01 | Plane-internal; retained in M05 |
P2-INT-05 | Grounding Qualification | P2-ABB-03 -> P2-ABB-01 | Plane-internal; retained in M05 |
P2-INT-06 | Context Provision | P2 -> P4 | M11-INT-11 |
P2-INT-07 | Context Trace / Evidence | P2 -> P7 | M11-INT-13 |
P2-INT-08 | Source / Context Currentness Feedback | P7/source authority -> P2 | M11-INT-14 plus M12 for external source authority |
P2-INT-09 | Context Invalidation / Refresh Notification | P2 -> P4/P5 | M11-INT-15 |
P3-INT-01 | Capability Requirement Provision | P1/P5/P4 -> P3 | M11-INT-16 |
P3-INT-02 | Primitive Availability Discovery | P3 < - > authority/P5/runtime | Realization/discovery subinteraction; M11-INT-16/17/20 plus M12 where external |
P3-INT-03 | Capability Contribution Declaration | External primitive authority/evidence -> P3 | Retained external ingress; detailed authority mapping in M12 |
P3-INT-04 | Primitive Evidence / Metadata Access | P3 < - > P7/AIPS/authority | M11-INT-18/19 plus M12 where external |
P3-INT-05 | Compatibility Determination | P3 < - > P4/P5/dependencies | Contained in M11-INT-16/17/20 contract obligations |
P3-INT-06 | Primitive Realization Provision | P3 -> P4 | M11-INT-17 |
P3-INT-07 | Primitive Use / Outcome Evidence | P4 -> P3/P7 | M11-INT-18 |
P3-INT-08 | Assurance / Fitness Feedback | P7 -> P3 | M11-INT-19 |
P3-INT-09 | Primitive Change / Invalidation | P3/authority -> P4/P5/P7 | M11-INT-20 plus M12 where external authority originates change |
P4-INT-01 | Governing Envelope Provision | P1 -> P4 | M11-INT-03 |
P4-INT-02 | AI Context Provision | P2 -> P4 | M11-INT-11 |
P4-INT-03 | Primitive Realization Provision | P3 -> P4 | M11-INT-17 |
P4-INT-04 | Execution Invocation / Serving Request | P5 -> P4 | M11-INT-21 |
P4-INT-05 | Agent / Tool Participation | Agent/Tool < - > P4 | P4/external participant boundary; retained in M07, authority alignment in M12 |
P4-INT-06 | Authorization / Decision Control Request | P4 -> P6 | M11-INT-23 |
P4-INT-07 | Authorization / Action Control Outcome | P6 -> P4 | M11-INT-25/26 |
P4-INT-08 | Environment Observation | Environment -> P4 | Environment boundary; retained in M07 |
P4-INT-09 | Authorized Environment Interaction | P4 -> Environment | Environment boundary; retained in M07 |
P4-INT-10 | Runtime Evidence / Outcome | P4 -> P7 | M11-INT-29 |
P4-INT-11 | Assurance / Currentness / Safety Feedback | P7 -> P4 | M11-INT-32 |
P5-INT-01 | Governing Serving Constraints | P1 -> P5 | M11-INT-04 |
P5-INT-02 | Consumer Consumption Request | Consumer/peer -> P5 | Serving/external ingress; retained in M08, peer interoperability in M12 |
P5-INT-03 | Product / Offering Association | Product authority -> P5 | External Product-authority ingress; M12; association to P1 via M11-INT-07 where required |
P5-INT-04 | Execution Invocation / Serving Request | P5 -> P4 | M11-INT-21 |
P5-INT-05 | Execution Result / Effect Return | P4 -> P5 | M11-INT-22 |
P5-INT-06 | Provider / Integration Invocation | P5 -> external/peer provider | External/peer boundary; detailed in M12 |
P5-INT-07 | Provider / Integration Response | External/peer provider -> P5 | External/peer boundary; detailed in M12 |
P5-INT-08 | Serving / Integration Evidence | P5 -> P7 | M11-INT-30 |
P5-INT-09 | Assurance / Currentness Feedback | P7/authority -> P5 | M11-INT-32 plus M12 where external authority |
P5-INT-10 | Contract Change / Deprecation Notice | P5 -> consumers/peers | Serving/external egress; retained in M08, peer interoperability in M12 |
P6-INT-01 | Governing Authority / Policy Basis | P1 -> P6 | M11-INT-05 |
P6-INT-02 | Decision Context / Evidence Provision | P2/P7 -> P6 | M11-INT-12 plus M11-INT-32 where P7 currentness/assurance constrains control |
P6-INT-03 | Decision / Action Control Request | P4/P5 -> P6 | M11-INT-23 |
P6-INT-04 | Decision Input / Intelligence Effect | P4/P5 -> P6 | M11-INT-24 |
P6-INT-05 | Human Review / Approval Request | P6 -> Human/external authority | Human/external control boundary; retained in M09 |
P6-INT-06 | Human / Authority Outcome | Human/external authority -> P6 | Human/external control boundary; retained in M09 |
P6-INT-07 | Authorization / Decision Outcome | P6 -> P4/P5 | M11-INT-25 |
P6-INT-08 | Authorized Action Commitment | P6 -> P4/P5/external system | M11-INT-26 for P4/P5; external enactment boundary retained/M12 where peer architecture |
P6-INT-09 | Action / Decision Outcome Evidence | P4/P5/external -> P6 | M11-INT-27 for P4/P5; external outcome boundary retained |
P6-INT-10 | Decision / Authorization Evidence | P6 -> P7 | M11-INT-31 |
P6-INT-11 | Assurance / Safety / Currentness Feedback | P7/P1 -> P6 | M11-INT-32 plus M11-INT-05/09 for P1 governing change |
P6-INT-12 | Intervention / Revocation / Stop | P6 -> P4/P5/external system | M11-INT-28 for P4/P5; external enactment boundary retained |
P7-INT-01 | Governing Evaluation / Assurance Requirements | P1/governance -> P7 | M11-INT-06 plus M12 for external governance source |
P7-INT-02 | Context / Source Evidence | P2 -> P7 | M11-INT-13 |
P7-INT-03 | Primitive / Capability Evidence | P3 -> P7 | M11-INT-18 |
P7-INT-04 | Runtime Evidence / Outcome | P4 -> P7 | M11-INT-29 |
P7-INT-05 | Serving / Integration Evidence | P5 -> P7 | M11-INT-30 |
P7-INT-06 | Decision / Authorization / Oversight Evidence | P6 -> P7 | M11-INT-31 |
P7-INT-07 | External Evaluation / Assurance Evidence | External authority/evaluator -> P7 | External evidence boundary; detailed authority/interoperability in M12 |
P7-INT-08 | Governing / Assurance Feedback | P7 -> P1 | M11-INT-08 |
P7-INT-09 | Currentness / Eligibility Feedback | P7 -> P2/P3/P5 | M11-INT-14/19/32 |
P7-INT-10 | Safety / Continuation Feedback | P7 -> P4/P6 | M11-INT-32 |
P7-INT-11 | Change / Reassessment Notification | P7 -> affected P1-P6 | M11-INT-33 |
This reconciliation preserves earlier plane obligations without manufacturing duplicate M11 identities for plane-internal or non-plane boundaries.
GEAIRA-M11-R12A - Complete source reconciliation: All seventy M04-M10 plane-level interaction declarations SHALL remain traceable through the M11 reconciliation table; omission from the 33 canonical inter-plane identities SHALL indicate an explicit plane-internal or external-boundary disposition rather than accidental deletion.
30.6 Interaction direction and reciprocity
Interaction direction identifies the architectural producer/consumer responsibility for a particular hand-off. It does not imply that the overall relationship between two planes is one-way.
Example:
P5 -> P4 M11-INT-21 Execution Invocation
P4 -> P5 M11-INT-22 Execution Result / Effect Return
These are distinct interactions with distinct outcomes and lifecycle semantics.
Likewise:
P4/P5 -> P6 M11-INT-23 Control Request
P6 -> P4/P5 M11-INT-25 Control Outcome
P6 -> P4/P5 M11-INT-26 Action Commitment
P4/P5 -> P6 M11-INT-27 Outcome Evidence
GEAIRA-M11-R14 - Directional semantics: Opposite-direction hand-offs SHALL NOT be collapsed into one ambiguous bidirectional interaction where request, decision, action or evidence semantics differ materially.
GEAIRA-M11-R15 - Response non-equivalence: A transport-level response/acknowledgement SHALL NOT automatically be interpreted as the architectural outcome of the initiating interaction.
30.7 Identity, causality and correlation
GEAIRA does not require one universal enterprise correlation identifier. Different architectural occurrences have different identities.
Relevant identities may include:
governed-use identity
consumer request identity
serving interaction identity
Product / Product Version identity (where valid)
AI Context / context assembly identity
primitive realization / primitive version identity
AI Execution identity
Agent / Tool participant identity
Decision identity
Authorization identity
action / command identity
Evaluation identity
Assurance Assessment identity
Evidence / provenance identity
Currentness Assessment identity
Semantic Change Event identity
The architecture SHOULD preserve causal links between these identities rather than overwrite them with one identifier.
A representative chain is:
ConsumerRequest C17
|
v
ServingInteraction S42
|
v
AIExecution E93
/ \
v v
Context K8 PrimitiveRealization R4
|
+-------> Decision D12
|
v
Authorization A7
|
v
Action X5
|
v
Evidence / Evaluation
Not every scenario requires every identity.
GEAIRA-M11-R16 - Identity separation: Distinct architectural occurrences SHALL retain distinct identities where conflation would impair traceability, authority, evidence or reconciliation.
GEAIRA-M11-R17 - Correlation without identity collapse: Cross-plane correlation SHALL relate identities without asserting semantic identity between them.
GEAIRA-M11-R18 - Causality preservation: Where causality is material, the architecture SHALL preserve sufficient predecessor/trigger/reference information to reconstruct the relevant causal chain.
GEAIRA-M11-R19 - No total-order requirement: GEAIRA SHALL NOT require a single total order of all enterprise-AI events; partial ordering and causal correlation are sufficient where they preserve the required architectural truth.
30.8 Governing envelope and authority propagation
P1 governing information may flow to multiple planes, but authority is not automatically transitive.
P1 governing envelope
|
+--> P2 context constraints
+--> P3 primitive constraints
+--> P4 execution constraints
+--> P5 serving constraints
+--> P6 authority/policy basis
+--> P7 evaluation/assurance requirements
The same governing source may therefore be projected differently for different responsibilities.
GEAIRA-M11-R20 - Purpose-bound propagation: Governing information propagated across planes SHALL remain bound to its applicable purpose, scope, subject/target, conditions and currentness where material.
GEAIRA-M11-R21 - No authority laundering: Receipt of a governing envelope, Delegated Authority reference, policy reference or Product governance context SHALL NOT itself create point-of-use Authorization.
GEAIRA-M11-R22 - Explicit delegation chain: Where an authority decision depends on delegation, sufficient delegation provenance/scope SHALL remain traceable across the relevant hand-offs.
GEAIRA-M11-R23 - Revocation reachability: Material revocation/supersession SHALL be capable of reaching affected dependent responsibilities within the latency required by the governing context.
GEAIRA-M11-R24 - Cached-authority discipline: Cached or embedded governing state MAY be used, but its validity/currentness/revocation strategy SHALL be explicit where stale authority could cause material harm or non-compliance.
30.9 Evidence and provenance propagation
Evidence can originate in any plane and be consumed by P7 or other control responsibilities.
M11 distinguishes:
operational record / telemetry
!= Evidence automatically
Evidence
!= proof
Evidence propagation
!= semantic ownership transfer
An evidence-bearing hand-off SHOULD preserve, where material:
- evidence identity;
- producing source/plane;
- subject/target identity;
- occurrence/event time;
- effective/currentness time where different;
- version/context references;
- provenance/integrity references;
- scope/claim to which the evidence is relevant;
- transformation/aggregation lineage where applicable.
GEAIRA-M11-R25 - Evidence source identity: Evidence propagated across planes SHALL preserve the identity/provenance needed to determine what produced it and what claim/scope it can legitimately support.
GEAIRA-M11-R26 - Telemetry non-promotion: Mere presence of logs, metrics, traces or events SHALL NOT automatically establish that those records constitute sufficient governance/assurance Evidence.
GEAIRA-M11-R27 - Aggregation traceability: Where evidence is aggregated or transformed, the architecture SHALL preserve sufficient lineage to avoid presenting derived evidence as if it were raw source evidence.
30.10 Interaction modes and temporal semantics
A canonical interaction MAY be realized synchronously or asynchronously unless the applicable architecture profile or control obligation requires otherwise.
GEAIRA recognizes at least these generic realization modes:
in-process / embedded
request-response
asynchronous command
asynchronous event
publish-subscribe
stream
batch
shared governed state
artifact/file exchange
human-mediated workflow
federated exchange
GEAIRA-M11-R28 - Mode neutrality: Canonical interactions SHALL remain valid across interaction modes unless a material semantic/control property requires a narrower mode.
GEAIRA-M11-R29 - Synchronous non-completion: A synchronous transport response SHALL NOT imply that the underlying business/AI action has completed unless the contract explicitly defines that outcome.
GEAIRA-M11-R30 - Asynchronous acknowledgement distinction: Acceptance/acknowledgement of an asynchronous hand-off SHALL remain distinct from processing completion, decision outcome, action outcome or evidence finalization.
GEAIRA-M11-R31 - Event-time distinction: Where event time, processing time, effective time and observation time differ materially, the contract SHALL preserve the distinctions needed for currentness, causality or auditability.
GEAIRA-M11-R32 - Late/out-of-order handling: Asynchronous architectures SHALL define how materially late, duplicated or out-of-order interactions affect currentness, reconciliation and downstream control.
30.11 Delivery, idempotency, retry and side-effect semantics
GEAIRA does not promise universal "exactly once" semantics.
The contract SHOULD distinguish, where applicable:
message/request delivered
request accepted
processing started
processing completed
side effect committed
outcome observed
outcome evidence recorded
These states are not interchangeable.
GEAIRA-M11-R33 - Delivery/processing separation: Delivery or receipt of an interaction SHALL NOT be equated with successful processing or side-effect completion.
GEAIRA-M11-R34 - Idempotency explicitness: Where retry/replay is possible, the contract SHALL identify whether the semantic operation is idempotent, conditionally idempotent or non-idempotent.
GEAIRA-M11-R35 - No blind retry for consequential side effects: A timed-out or disconnected consequential operation SHALL NOT be blindly retried when the architecture cannot determine whether the side effect occurred.
GEAIRA-M11-R36 - Indeterminate outcome preservation: Unknown side-effect state SHALL be represented as indeterminate/reconciliation-required rather than coerced into failure.
GEAIRA-M11-R37 - Duplicate tolerance: Where the delivery mechanism may duplicate interactions, consumers SHALL have a strategy appropriate to the operation for duplicate detection, idempotency or reconciliation.
GEAIRA-M11-R38 - Exactly-once restraint: GEAIRA SHALL NOT require a universal exactly-once delivery mechanism; profiles MAY require stronger guarantees where justified and realizable.
30.12 Canonical outcome model
M11 defines the following generic architectural outcome categories. They are not a mandatory wire enum.
| Outcome | Meaning |
|---|---|
| ACCEPTED | The interaction was accepted for processing; completion is not implied. |
| COMPLETED | The contractual operation completed as defined by the applicable contract. |
| REJECTED | The interaction was deliberately not accepted/permitted. |
| DEFERRED | Processing/decision was intentionally postponed pending another condition. |
| CONDITIONAL | Continuation is allowed only under explicit conditions. |
| PARTIAL | A defined subset completed; remaining scope is unresolved or failed. |
| FAILED | The architecture has positive knowledge that the requested operation failed under the contract definition. |
| INDETERMINATE | The true outcome cannot currently be established. |
| CANCELLED / REVOKED | Previously pending/valid continuation was withdrawn or stopped. |
| EXPIRED / STALE | The contract input/outcome is no longer valid/current for the intended use. |
| NOT APPLICABLE | The interaction/control was evaluated as not applicable to the governed scenario. |
GEAIRA-M11-R39 - Outcome preservation: Architecturally distinct outcomes SHALL NOT be flattened into a binary success/failure status where that would lose material control or reconciliation meaning.
GEAIRA-M11-R40 - Accepted/completed distinction: ACCEPTED SHALL NOT be represented as COMPLETED unless the contract semantics make them equivalent for that operation.
GEAIRA-M11-R41 - Failure knowledge requirement: FAILED SHOULD indicate positive knowledge of failure; absence of confirmation alone SHOULD use an indeterminate/timeout-specific state where the true effect is unknown.
30.13 Contract versioning, compatibility and evolution
M11 separates several independent version dimensions:
Interaction Contract version
Protocol/schema version
Product / Product Version
AI Primitive version
AI Context/source version
Policy / governing-envelope version
Evaluation criterion version
Evidence/report version
None is automatically the same as another.
GEAIRA-M11-R42 - Version-dimension separation: Contract, protocol/schema, Product, primitive, context/source and governance versions SHALL remain independently identifiable where material.
GEAIRA-M11-R43 - Semantic compatibility over syntax: Syntactic/schema compatibility SHALL NOT by itself establish semantic, control, capability or assurance compatibility.
GEAIRA-M11-R44 - Contract evolution: Contract evolution SHALL define compatibility, deprecation, supersession and migration behavior appropriate to affected producers/consumers.
GEAIRA-M11-R45 - Historical contract truth: Later contract/version supersession SHALL NOT rewrite which contract/version governed a historical interaction.
GEAIRA-M11-R46 - Consumer acceptance boundary: Consumers SHALL be able to determine whether an interaction contract/version remains acceptable/current for their governed use where that matters.
GEAIRA-M11-R47 - Breaking semantic change: A change that materially alters meaning, authority, capability, effect, failure semantics or evidence obligations SHALL NOT be treated as non-breaking merely because the payload schema remains compatible.
30.14 Cross-plane control points
M11 freezes four logical cross-plane control-point classes.
CP11-01 - Boundary Admission Control Point
Determines whether the incoming hand-off is admissible under contract identity/version, governing envelope, source/target identity and applicable constraints.
CP11-02 - Contract Completeness & Currentness Control Point
Determines whether required contract content, version/currentness and evidence/authority references are sufficiently complete for downstream use.
CP11-03 - Side-Effect & Outcome Reconciliation Control Point
Handles ambiguous outcomes, retries, duplicates and consequential side-effect reconciliation.
CP11-04 - Contract Change / Revocation Control Point
Propagates and enforces material contract, governance, eligibility, assurance or revocation change across dependent boundaries.
These are logical responsibilities and MAY be embedded or distributed.
GEAIRA-M11-R48 - Control-point non-service: CP11-01 through CP11-04 SHALL NOT imply standalone gateway, broker, workflow, policy or reconciliation services.
GEAIRA-M11-R49 - Control placement: A control point MAY be realized at producer, consumer, intermediary, shared platform, embedded runtime or human-governance boundary if its architectural obligation remains traceable.
30.15 Cross-cutting overlays on Interaction Contracts
The M03 X1-X8 overlays apply to M11 contracts as follows:
| Overlay | M11 contract concern |
|---|---|
| X1 Security, Privacy & Trust Boundaries | Authentication, confidentiality, minimization, integrity, trust-boundary crossing. |
| X2 Identity, Authority & Policy Enforcement | Producer/consumer identity, authority source, delegation, Authorization references and revocation. |
| X3 Evidence, Provenance & Traceability | Evidence identity, provenance, causal correlation and historical reconstruction. |
| X4 Observability, Telemetry & Auditability | Contract visibility, latency/outcome telemetry and audit-relevant records. |
| X5 Resilience, Safety & Failure Management | Timeouts, degraded operation, duplicate/retry behavior, fail-safe and indeterminate states. |
| X6 Lifecycle, Version, Currentness & Change | Contract/source/version currentness, deprecation, supersession and migration. |
| X7 Interoperability, Contracts & Semantic Compatibility | Contract meaning, compatibility and boundary transformation. |
| X8 Governance, Human Accountability & Compliance | Human decision/approval/escalation obligations and compliance evidence. |
GEAIRA-M11-R50 - Overlay applicability: Applicable X1-X8 obligations SHALL be reflected in the Interaction Contract rather than assumed to exist invisibly outside the hand-off.
30.16 Representative cross-plane scenarios
Scenario A - Product-serving request to AI Execution
Consumer
|
| P5 serving request
v
P5 -- M11-INT-21 --> P4
<--- M11-INT-22 --
|
v
Consumer
The consumer request, serving interaction and AI Execution retain distinct identities. P5 contract admission does not create P6 Authorization and does not itself prove execution completion.
Scenario B - Agentic tool use requiring point-of-use control
P4 Agent/Tool participation
|
| M11-INT-23
v
P6 Authorization / Action Control
|
| M11-INT-25 / M11-INT-26
v
P4 technical enactment
|
| M11-INT-27
v
P6 outcome reconciliation
A timeout after action dispatch may be INDETERMINATE; blind retry is prohibited where duplicate side effects are unsafe.
Scenario C - Context source becomes non-current
P7 finding
|
| M11-INT-14
v
P2 source/context eligibility changes
|
| M11-INT-15
+------> P4 / P5 affected future use
Historical executions that used the earlier context remain historically true.
Scenario D - Primitive assurance finding affects future execution
P7 -- M11-INT-19 --> P3
P3 updates eligibility
P3 -- M11-INT-20 --> P4/P5/P7
The finding can prevent future use without rewriting historical primitive involvement.
Scenario E - Human-final consequential decision
P4/P5 -- M11-INT-24 --> P6
P2 ------ M11-INT-12 --> P6
P1 ------ M11-INT-05 --> P6
Human review occurs inside/through P6
P6 ------ M11-INT-25/26 --> P4/P5
P6 ------ M11-INT-31 --> P7
Human availability is not proof that review occurred; the oversight occurrence must remain evidentiary where required.
Scenario F - P1 authority revocation
P1 -- M11-INT-09 --> affected P2/P3/P4/P5/P6/P7
The architecture does not assume one central synchronous broadcast service, but the revocation must reach affected responsibilities within the latency required by the governing context.
Scenario G - P7 reassessment after semantic or contract change
P7 -- M11-INT-33 --> affected P1-P6
Reassessment may alter future eligibility, serving, execution or action controls while preserving historical contract and occurrence truth.
30.17 M11 interaction anti-patterns
M11 defines the following anti-patterns:
- Plane pipeline fallacy - treating P1-P7 as seven mandatory sequential runtime stages.
- API-equals-contract - assuming OpenAPI/AsyncAPI/IDL syntax is the whole architectural contract.
- Shared-database contract - relying on undocumented shared tables/state as the integration meaning.
- One global correlation ID - collapsing request, execution, decision, Authorization, action and evidence identities into one identifier.
- Authority laundering - treating propagated purpose/delegation/policy context as point-of-use Authorization.
- Telemetry-equals-evidence - assuming every log/metric/trace is sufficient Evidence.
- Synchronous-equals-complete - treating a synchronous response as proof of downstream completion.
- Acknowledged-equals-done - treating message acceptance as execution/action completion.
- Timeout-equals-failure - declaring a consequential operation failed when its side effect is unknown.
- Blind retry - repeating a non-idempotent/consequential operation without reconciliation.
- Exactly-once mythology - designing the generic architecture around a universal exactly-once assumption.
- Binary-status flattening - removing deferred, conditional, partial or indeterminate outcomes.
- Latest-wins history rewrite - projecting the newest contract/governance/version state onto historical interactions.
- Schema-compatible-equals-semantically-compatible - ignoring material meaning/control changes because syntax still validates.
- Central-bus mandate - assuming all inter-plane interactions must traverse one event bus/service mesh/gateway.
- Cross-plane mega-ABB - collapsing several plane responsibilities into one generic "AI Platform" ABB merely because one platform implements them.
- Protocol-specific canonical interaction - creating new canonical interactions for REST vs event vs stream realization of the same hand-off.
- Implicit revocation - changing governing state without a defined reachability/currentness strategy for affected responsibilities.
30.18 Frozen M11 decisions
M11 freezes the following GEAIRA 0.1 architecture decisions:
plane interaction != semantic relationship one-to-one
Interaction Contract != API/schema/topic/table/service
P1-P7 != mandatory runtime pipeline
co-location != absence of logical contract
shared state != permission for semantic ambiguity
request identity != serving identity != AI Execution identity
Decision identity != Authorization identity != Action identity
correlation != semantic identity
Governing Envelope != point-of-use Authorization
Delegated Authority propagation != transitive permission
telemetry != Evidence automatically
Evidence != proof
receipt != acceptance != completion != side-effect commitment
synchronous response != completion automatically
timeout != failure automatically
unknown side effect -> INDETERMINATE / reconciliation
contract version != Product version != primitive version != context version
schema compatibility != semantic compatibility
supersession != historical mutation
revocation/currentness/change must be reachable to affected responsibilities
external peer-architecture interoperability details remain M12 scope
The following M11 artifacts are canonical human-readable structures:
6 interaction families
33 canonical inter-plane interactions
19-dimensional conceptual Interaction Contract model
11 generic outcome categories
4 cross-plane logical control-point classes
18 interaction anti-patterns
M11 does not freeze:
- a universal payload/schema language;
- a universal error-code vocabulary;
- one protocol or message broker;
- a global correlation-ID format;
- a global event envelope;
- exactly-once delivery;
- one service mesh/API gateway;
- one evidence storage platform;
- detailed external contracts with GEDRA/GEARA/AIPDS/HDIP;
- deployment topology.
GEAIRA-M11-R51 - Frozen catalogue preservation: Later slices SHALL preserve M11-INT-01 through M11-INT-33 unless a reconciled 0.1 architecture change explicitly replaces an interaction without weakening its architectural obligation.
GEAIRA-M11-R52 - Cross-plane semantics preservation: Later realization patterns SHALL preserve the architectural meaning, authority, outcome and currentness semantics of M11 contracts even when physical interfaces are merged or split.
GEAIRA-M11-R53 - External-interoperability hand-off: M12 SHALL define adjacent-architecture interoperability by mapping peer-boundary exchanges onto the M11 Interaction Contract model rather than creating an unrelated integration model.
GEAIRA-M11-R54 - Deployment-neutral continuity: M13 SHALL demonstrate that M11 contracts remain valid across centralized, distributed, federated, edge/physical and mixed realizations.
GEAIRA-M11-R55 - Pattern conformance: M14-M16 patterns, views and conformance rules SHALL reference canonical M11 interactions where material rather than inventing untraceable cross-plane hand-offs.
GEAIRA-M11-R56 - Historical-truth continuity: Contract evolution, revocation, reassessment and migration SHALL constrain current/future use without rewriting the contract/version, identities or outcomes that actually governed historical interactions.
GEAIRA-M11-R57 - Architecture diversity continuity: M11 SHALL remain applicable to Product and non-Product AI, agentic and non-agentic AI, learned/symbolic/hybrid mechanisms, synchronous/asynchronous workloads, and software/physical environments without structural redesign solely because the implementation style differs.
GEAIRA-M11-R58 - M12 readiness: M12 SHALL use the M11 contract model to define explicit interoperability boundaries with adjacent semantic/reference/deployment architectures while preserving their independent authority.
31. M11 Completion State
GEAIRA-M11 - COMPLETE.
M11 establishes:
- one canonical technology-neutral Interaction Contract model;
- six canonical interaction families;
- thirty-three canonical inter-plane interactions;
- explicit identity/correlation/causality rules;
- explicit governing-envelope and authority-propagation rules;
- explicit evidence/provenance propagation rules;
- synchronous/asynchronous/embedded/human-mediated interaction-mode neutrality;
- explicit delivery, acknowledgement, processing and side-effect distinctions;
- eleven generic architectural outcome categories;
- explicit idempotency/retry/indeterminate-side-effect rules;
- explicit version/currentness/compatibility evolution rules;
- four logical cross-plane control-point classes;
- application of all eight X1-X8 cross-cutting overlays;
- seven representative cross-plane scenarios;
- eighteen interaction anti-patterns;
- 59 normative M11 rules.
M11 introduces no change to GEAIRM, BPS, AIPS, Agent Ontology or PROV semantics, does not create a mandatory P1-P7 processing pipeline, and does not prescribe a protocol, global event envelope, integration bus, API gateway, service mesh, evidence store or deployment topology.
Next slice
GEAIRA-M12 - Adjacent Architecture Interoperability
M12 will use the M11 Interaction Contract model to define technology-neutral interoperability responsibilities and authority-preserving boundaries with GEDRA, GEARA, AIPS/AIPDS, BPS, the Agent Ontology, W3C PROV and enterprise-specific realizations such as HDIP.
32. GEAIRA-M12 - Adjacent Architecture Interoperability
32.1 Purpose and architectural boundary
M01 established that GEAIRA is one architecture authority in a broader specification ecosystem. M11 then froze the technology-neutral Interaction Contract model used for material hand-offs between GEAIRA responsibilities. M12 applies that contract discipline at the edges of GEAIRA.
M12 answers:
How does GEAIRA interoperate with adjacent semantic authorities, peer reference architectures, Product/deployment specifications, provenance standards and enterprise-specific realizations without absorbing their jurisdiction or weakening GEAIRA's own architecture responsibilities?
M12 therefore defines:
- adjacent-authority and adjacent-architecture boundary classes;
- authority-preserving mapping rules;
- a canonical interoperability mapping record;
- explicit GEAIRA intersections with GEDRA and GEARA;
- Product and AI Product boundary alignment with BPS/PIN and AIPS;
- deployment alignment with AIPDS;
- Agent identity/taxonomy alignment with the KivaNura Agent Ontology;
- provenance alignment with W3C PROV;
- enterprise-realization mapping rules for HDIP and analogous profiles;
- cross-specification version, currentness, compatibility and conformance rules;
- representative cross-boundary scenarios and anti-patterns.
M12 does not:
- redefine any adjacent specification;
- make GEDRA, GEARA, AIPS, AIPDS, BPS, Agent Ontology, PROV or HDIP subordinate to GEAIRA;
- require one shared namespace or registry across all specifications;
- assert one-to-one equivalence between peer architecture planes or ABBs;
- make conformance to one specification imply conformance to another;
- prescribe a shared implementation platform;
- make HDIP topology intrinsic to generic GEAIRA;
- introduce a new semantic ontology above GEAIRM, GEDRM, GEARM, BPS, AIPS or Agent Ontology.
GEAIRA-M12-R01 - Boundary scope: M12 SHALL define interoperability at specification and architecture boundaries while preserving the independent jurisdiction of every adjacent authority.
GEAIRA-M12-R02 - M11 contract reuse: Material operational exchanges across an adjacent boundary SHALL use the M11 Interaction Contract model rather than a second incompatible integration model.
GEAIRA-M12-R03 - No authority absorption: Referencing, mapping to, consuming from or interoperating with an adjacent specification SHALL NOT transfer semantic or architectural authority to GEAIRA.
GEAIRA-M12-R04 - No architecture flattening: Adjacent architectures SHALL NOT be flattened into one universal plane, ABB, namespace, registry or platform hierarchy merely to simplify integration.
32.2 Interoperability boundary types
M12 distinguishes five boundary types.
| Boundary type | Purpose | Examples |
|---|---|---|
| Semantic Authority Reference | Preserve externally owned meaning used by GEAIRA architecture. | GEAIRM, BPS, AIPS, Agent Ontology, GEDRM, GEARM, PROV |
| Peer Architecture Interoperability | Coordinate responsibilities and contracts between independently governed reference architectures. | GEDRA, GEARA |
| Product / Deployment Specialization | Apply Product- or AI-Product-specific semantics/deployment obligations to a GEAIRA realization. | BPS, AIPS, AIPDS |
| External Standard / Evidence Alignment | Reuse an external standard vocabulary or evidence/provenance model without shadowing it. | W3C PROV |
| Enterprise Realization Mapping | Map generic GEAIRA responsibilities into an enterprise-specific target architecture/profile. | HDIP |
A single adjacent specification MAY participate in more than one type. For example, AIPS is both a semantic authority and the home of AIPDS deployment semantics.
GEAIRA-M12-R05 - Typed boundary: Every normative adjacent-specification mapping SHALL identify the kind of boundary being established rather than using a generic integrates with assertion.
GEAIRA-M12-R06 - Many-to-many mapping: GEAIRA-to-adjacent mappings MAY be many-to-many; similarity of labels SHALL NOT establish architectural equivalence.
GEAIRA-M12-R07 - No transitive identity: If a GEAIRA responsibility maps to an adjacent responsibility and that adjacent responsibility maps to another construct, GEAIRA SHALL NOT infer semantic or architectural identity transitively unless the authoritative specifications explicitly establish it.
GEAIRA-M12-R08 - Co-realization non-equivalence: One deployed platform MAY realize responsibilities from several architectures without causing those responsibilities or architectures to become equivalent.
GEAIRA-M12-R09 - Shared-control preservation: A shared enterprise control MAY satisfy obligations from several specifications, but each specification's applicable obligation and evidence SHALL remain traceable independently.
GEAIRA-M12-R10 - Contract-before-technology: Adjacent runtime interoperability SHALL be defined first in terms of responsibility, semantics, authority, identity, outcome, evidence and change behavior; technology/protocol selection is a realization concern.
GEAIRA-M12-R11 - Explicit externality: A boundary that leaves GEAIRA jurisdiction SHALL remain explicit even when the adjacent responsibility is implemented in the same process, product or platform.
GEAIRA-M12-R12 - Profile strengthening: Enterprise or domain profiles MAY strengthen adjacent-boundary controls and mappings but SHALL NOT weaken the normative authority separation established by M12.
32.3 Canonical interoperability mapping record
An Adjacent Architecture Interoperability Record is the human-readable architecture record used to document a material GEAIRA boundary.
A record SHOULD identify, where applicable:
| Dimension | Required architectural question |
|---|---|
| Boundary ID | Which GEAIRA adjacent boundary is being described? |
| Adjacent authority | Which specification, ontology, architecture or profile is authoritative on the other side? |
| Boundary type | Semantic reference, peer architecture, Product/deployment specialization, external standard or enterprise realization? |
| Purpose | Why does this interoperability boundary exist? |
| GEAIRA responsibilities | Which P1-P7 responsibilities/ABBs participate? |
| Adjacent responsibilities | Which adjacent architecture responsibilities participate, where such constructs exist? |
| Semantic authorities | Which specification owns the meaning of exchanged concepts? |
| M11 contract references | Which M11 interaction(s)/contract family governs operational exchange? |
| Identity bindings | Which Product, version, Agent, Data/Information resource, execution, decision, evidence or other identities remain correlated? |
| Authority/control basis | Which authority, policy, entitlement, governance or authorization context applies? |
| Currentness/version basis | Which versions/currentness states govern the mapping? |
| Evidence/provenance | What trace must cross or remain linkable across the boundary? |
| Compatibility assumptions | Which semantic/contract/deployment compatibility assumptions are material? |
| Failure/degradation behavior | What happens when the adjacent authority or exchange is unavailable, stale, incompatible or indeterminate? |
| Conformance independence | Which conformance claims remain separate? |
| Realization/profile notes | Which enterprise-specific mappings specialize the generic boundary? |
The record is conceptual; M12 does not prescribe one serialization.
GEAIRA-M12-R13 - Mapping-record traceability: A material adjacent mapping SHALL retain enough information to identify GEAIRA responsibility, adjacent authority, semantic ownership and applicable contract/control obligations.
GEAIRA-M12-R14 - External identifier preservation: Authoritative external identifiers SHOULD be retained where available rather than replaced with locally invented aliases that obscure provenance or ownership.
GEAIRA-M12-R15 - Version-visible mapping: A mapping that depends on a specific external specification, Product Version, contract version or profile version SHALL make that dependency visible.
GEAIRA-M12-R16 - Failure visibility: Loss, staleness or incompatibility of an adjacent dependency SHALL be represented explicitly where it materially affects safe/correct operation; it SHALL NOT silently degrade to an ungoverned local assumption.
GEAIRA-M12-R17 - Historical mapping truth: A later mapping/version change MAY constrain current or future interoperability but SHALL NOT rewrite which adjacent version or authority basis governed a historical interaction.
32.4 Canonical adjacent-boundary catalogue
M12 establishes the following canonical boundary catalogue.
| ID | Adjacent authority / architecture | Boundary role | Principal GEAIRA intersection |
|---|---|---|---|
| M12-BND-00 | GEAIRM | Semantic foundation | All P1-P7; semantic meaning remains GEAIRM-owned |
| M12-BND-01 | GEDRA / GEDRM | Peer Data architecture + Data semantics | P1, P2, P5, P7; X1-X8 as applicable |
| M12-BND-02 | GEARA / GEARM | Peer Analytics/Information architecture + semantics | P1-P7, especially P2, P4, P5, P6, P7 |
| M12-BND-03 | BPS / PIN | Generic Product semantic authority | P1 and P5 principally; P7 for Product lifecycle evidence where applicable |
| M12-BND-04 | AIPS | AI Product / primitive / capability / risk-governance authority | P1, P3, P5, P7 and conditional P4/P6 obligations |
| M12-BND-05 | AIPDS | Qualified AI Product deployment specification | P3-P5, P7 and applicable P1/P6 controls |
| M12-BND-06 | KivaNura Agent Ontology | Agent identity/taxonomy authority | P4 principally; P1/P6 for authority/control intersections |
| M12-BND-07 | W3C PROV | Generic provenance authority | X3 and P7, with provenance links from all planes |
| M12-BND-08 | HDIP | Enterprise-specific realization/profile | May realize/specialize P1-P7 and adjacent architecture responsibilities |
GEAIRA-M12-R18 - Canonical boundary catalogue: M12-BND-00 through M12-BND-08 SHALL be the canonical adjacent-boundary identities for the GEAIRA 0.1 human-readable baseline.
GEAIRA-M12-R19 - Semantic-foundation priority: Where an architecture mapping appears to conflict with the authoritative meaning of a GEAIRM/GEDRM/GEARM/BPS/AIPS/Agent/PROV concept, the mapping SHALL be treated as defective rather than silently changing the authoritative concept.
GEAIRA-M12-R20 - Architecture-jurisdiction conflict: Where peer architectures assign materially overlapping responsibilities, the enterprise/profile architecture SHALL make the allocation and contract boundary explicit; it SHALL NOT resolve the overlap by declaring one peer specification semantically subordinate without authority to do so.
32.5 GEDRA / GEDRM interoperability
GEDRA is the peer generic enterprise Data reference architecture and GEDRM remains the Data semantic authority. GEAIRA consumes Data and Data Products where AI responsibilities require them, but it does not absorb Data authority, data productization, data movement or data-governance architecture into P2/P5 merely because AI uses Data.
The following are primary intersections, not equivalences:
| GEDRA architecture responsibility | Principal GEAIRA intersection | Interoperability interpretation |
|---|---|---|
| Semantic & Enterprise Ontology Plane | P2, X7 | GEAIRA preserves Data meaning/classification references; P2 does not redefine them. |
| Authority Plane | P1, P2, P7 | Data-source/scope authority constrains eligibility, context use and evidence/currentness. |
| Processing, Provision & Movement Plane | P2, P5 | Governed Data provision/movement can feed context/resource access and serving/integration contracts. |
| Data Platform & Product Plane | P2, P5, P7 | Data Product/resource identity, version, quality, lineage and lifecycle can participate in AI contracts without becoming AI Product semantics. |
| Consumption & Access Plane | P2, P5 | Access/entitlement and consumption interfaces may realize governed Data access for AI contexts/integrations. |
| Value, Observability & Feedback Plane | P7 | Data quality/value/usage/control evidence may contribute to AI evaluation/assurance where material. |
| Cross-Cutting Governance & Control | X1-X8 | Data policy/privacy/security/lineage/quality/retention/residency obligations remain applicable across AI use. |
| Technology / Utility Foundation | Realization layer | Shared technology may realize both architectures; it is not a GEAIRA plane. |
Typical M11 contract use includes:
GEDRA-governed Data / Data Product
|
| adjacent contract
v
P2 Knowledge & Retrieval Mediation
|
| M11-INT-11
v
P4 AI Execution
and:
GEDRA quality / lineage / currentness evidence
|
v
P7 Evidence / Evaluation / Currentness
|
+--> M11-INT-14 / 32 / 33 as applicable
GEAIRA-M12-R21 - Data authority preservation: GEAIRA SHALL preserve GEDRM/GEDRA authority for Data meaning, authoritative sourcing, Data Product obligations and Data-governance responsibilities.
GEAIRA-M12-R22 - Data Product non-equivalence: A Data Product consumed by or supporting AI SHALL NOT thereby become an AI Product, AI Primitive, AI Context or GEAIRA ABB.
GEAIRA-M12-R23 - P2 non-annexation: P2 Knowledge, Context & Grounding SHALL NOT be treated as replacing GEDRA Data processing, platform, product, authority or governance responsibilities.
GEAIRA-M12-R24 - Data-control continuity: Applicable Data privacy, access, lineage, quality, retention, residency, control and evidence obligations SHALL remain applicable when Data crosses into a GEAIRA context or execution path.
GEAIRA-M12-R25 - Data currentness propagation: Material GEDRA/GEDRM source/version/quality/currentness changes SHALL be capable of constraining future GEAIRA context, primitive, serving, decision or assurance use without rewriting historical AI occurrences.
GEAIRA-M12-R26 - Peer contract explicitness: Material GEAIRA-GEDRA operational exchanges SHALL identify the producer/consumer authority and use the M11 contract dimensions for identity, semantic reference, currentness, evidence and failure behavior.
32.6 GEARA / GEARM interoperability
GEARA is the peer generic enterprise Analytics reference architecture and GEARM remains authoritative for generic Analytics/Information semantics. Analytics may feed AI, AI may participate in analytical realization, and AI outputs may themselves become analytical inputs or Information. None of those interactions makes Analytics and AI the same domain.
The following are primary intersections, not plane equivalences:
| GEARA plane | Principal GEAIRA intersection | Interoperability interpretation |
|---|---|---|
| P1 Intent, Semantics & Design | GEAIRA P1/P2/P3 | Analytical intent/information need and AI purpose/capability requirements may constrain each other through explicit contracts. |
| P2 Discovery, Composition & Realization | GEAIRA P2/P3/P5 | Analytics may discover/compose Data, Information, AI Products/capabilities and dependencies without transferring their authority. |
| P3 Analytical Runtime & Information Formation | GEAIRA P4/P5 | Analytical runtime may invoke AI capability/Product serving; AI Execution may consume or produce analytical/Information artifacts. |
| P4 Provision, Consumption & Decision Integration | GEAIRA P5/P6 | Information/recommendations can be exposed and may inform GEAIRA Decision Formation; Information does not itself create decision authority. |
| P5 Outcome, Value & Evolution | GEAIRA P7 | Analytical outcomes/feedback may contribute evidence; AI assurance/currentness feedback may affect analytical use. |
Representative compositions include:
GEARA analytical output / Information
|
+--> P2 AI Context resource
|
+--> P6 Decision input
and:
GEARA analytical realization
|
| invokes governed AI capability / AI Product
v
GEAIRA P5 -> P4
|
v
result/effect returned to GEARA analytical flow
GEAIRA-M12-R27 - Analytics authority preservation: GEAIRA SHALL preserve GEARM/GEARA authority for analytical/Information semantics and analytical architectural responsibilities.
GEAIRA-M12-R28 - Analytics/AI non-equivalence: An analytical capability, Analytical Output, Information or Information Product SHALL NOT become AI, an AI Primitive or an AI Product solely because AI participates in its realization or consumption.
GEAIRA-M12-R29 - AI-assisted analytics composition: Where GEARA invokes a GEAIRA-governed AI capability, the AI interaction SHALL preserve GEAIRA purpose, primitive/execution, serving, evidence and applicable decision/action obligations rather than being hidden as an opaque analytical function.
GEAIRA-M12-R30 - Decision-support boundary: Analytical Information, recommendation, score or other output MAY inform P6 Decision Formation but SHALL NOT itself establish organizational decision authority or point-of-use Authorization.
GEAIRA-M12-R31 - Outcome/evidence linkage: GEARA outcome/value/feedback evidence and GEAIRA P7 evidence MAY be correlated, but the evidentiary claim, source, scope and currentness SHALL remain explicit.
GEAIRA-M12-R32 - Peer-plane non-equivalence: Similarity between GEARA and GEAIRA plane names or responsibilities SHALL NOT be treated as one-to-one plane identity.
32.7 BPS / PIN interoperability
BPS/PIN remains authoritative for generic Product semantics. GEAIRA consumes Product identity and Product-version context where a GEAIRA realization is intentionally productized, principally through P1 and P5.
BPS / PIN
Product / Product Version / Product Intrinsic Nature
|
v
GEAIRA P5 Product Realization & Contract Boundary
|
+--> GEAIRA serving / integration responsibilities
The absence of BPS Product status does not prevent an internal non-Product AI capability from using GEAIRA architecture responsibilities.
GEAIRA-M12-R33 - Product authority preservation: GEAIRA SHALL reference BPS/PIN for Productness, Product identity, Product Version and Product Intrinsic Nature rather than deriving those semantics from architecture exposure.
GEAIRA-M12-R34 - Exposure non-productness: API exposure, repeated use, serving, persistence, marketplace registration or integration SHALL NOT establish Product status.
GEAIRA-M12-R35 - Product mapping conditionality: P5 Product Realization responsibilities SHALL be conditional on authoritative Product status where Product-specific behavior is claimed.
GEAIRA-M12-R36 - Non-Product support: GEAIRA SHALL remain architecturally applicable to non-Product enterprise AI without requiring creation of artificial BPS Product identities.
32.8 AIPS interoperability
AIPS remains authoritative for AI Product qualification, AI Product Version classifications, AI Primitive/Intelligence Capability taxonomies and applicable AI risk/governance semantics.
Principal intersections are:
| AIPS concern | GEAIRA use |
|---|---|
| AI Primitive / Primitive Type | P3 mechanism realization and P4 occurrence trace |
| Intelligence Capability | P3 capability requirement/realization and P4 exercised capability |
| Execution Mode / Intelligence Effect | P4 occurrence semantics and P5/P6/P7 outcome contracts |
| AI Product / AI Product Version | P1/P5 Product governance/serving context where qualified |
| Agency/Autonomy / Embodiment | P4/P6 architectural control implications, without redefining classifications |
| Risk Tier / Governance Obligation Profile | P1 governing envelope and applicable P6/P7 controls |
| Qualification/currentness/evidence | P5/P7 eligibility, serving and assurance constraints |
GEAIRA-M12-R37 - AIPS authority preservation: GEAIRA SHALL consume AIPS-owned AI Product, primitive, capability, classification, risk and governance semantics without creating competing classifications.
GEAIRA-M12-R38 - No architecture-based qualification: Presence of GEAIRA ABBs, serving interfaces, runtime controls, observability or deployment components SHALL NOT establish AI Product qualification.
GEAIRA-M12-R39 - Classification projection: AIPS classifications MAY constrain GEAIRA responsibilities and control patterns, but the architecture SHALL NOT recompute or reinterpret them outside AIPS authority.
GEAIRA-M12-R40 - Risk/control traceability: Where an AIPS Governance Obligation Profile applies, GEAIRA controls MAY realize its obligations, but obligation identity and evidence SHALL remain traceable to the AIPS-governed Product Version.
GEAIRA-M12-R41 - Occurrence/product separation: P4 occurrence-level evidence about actual primitive use, Execution Mode or Intelligence Effect SHALL NOT automatically rewrite AIPS Product-Version declarations.
32.9 AIPDS interoperability
AIPDS is the deployment-facing specification for a specific qualified AI Product Version. GEAIRA is the generic enterprise-AI reference architecture. Their relationship is specialization, not duplication.
GEAIRA
technology-neutral architecture responsibilities
|
| qualified AI Product realization
v
AIPDS
Product-version deployment / execution / serving /
monitoring / governance / evolution specification
|
v
concrete deployment realization
AIPDS MAY bind concrete deployment targets, packaging, interfaces, resources, runtime controls, observability, evolution and Product-specific operational constraints that GEAIRA deliberately leaves generic.
GEAIRA-M12-R42 - Deployment specialization: AIPDS MAY specialize GEAIRA P3-P5/P7 and applicable P1/P6 responsibilities for a qualified AI Product Version without redefining the generic GEAIRA responsibilities.
GEAIRA-M12-R43 - Qualification prerequisite: GEAIRA SHALL NOT create an AIPDS requirement for a non-Product or non-qualified capability merely to make its architecture look uniform.
GEAIRA-M12-R44 - Control-binding traceability: AIPDS deployment controls that realize GEAIRA/AIPS obligations SHALL preserve traceability to the governing Product Version, obligation and applicable architecture responsibility.
GEAIRA-M12-R45 - Deployment non-equivalence: AIPDS deployment topology, packaging or runtime technology SHALL NOT be treated as the canonical generic GEAIRA topology.
32.10 KivaNura Agent Ontology interoperability
The Agent Ontology owns Agent identity, taxonomy and Agent-specific classifications. GEAIRA owns architectural responsibility for how Agents participate in enterprise-AI execution and controlled decision/action paths.
Agent Ontology
Agent identity / type / Agent classifications
|
v
GEAIRA P4 Agent & Tool Participation
|
+--> P6 Decision / Authorization control where applicable
+--> P1 Delegated Authority envelope where applicable
GEAIRA-M12-R46 - Agent authority preservation: GEAIRA SHALL reference Agent Ontology identity/taxonomy rather than minting shadow Agent classes for architectural convenience.
GEAIRA-M12-R47 - Participation non-identity: An Agent participating in P4 SHALL NOT become an AI Primitive, AI Execution, Product or ABB solely through that participation.
GEAIRA-M12-R48 - Autonomy/authority separation: Agent autonomy or control-mode classification SHALL NOT establish Delegated Authority or point-of-use Authorization.
GEAIRA-M12-R49 - Agent evidence traceability: Where Agent identity materially affects execution, decision/action or assurance, the relevant Agent identity/classification SHALL remain traceable in the applicable M11/P7 evidence chain.
32.11 W3C PROV interoperability
W3C PROV remains the generic provenance authority. GEAIRA X3 and P7 require provenance/evidence traceability but do not create a replacement provenance ontology.
GEAIRA may use PROV-compatible relationships to connect, for example:
- context/resource acquisition;
- primitive/version participation;
- AI Execution occurrences;
- decisions/Authorization/action evidence;
- Evaluation and Assurance artifacts;
- Product/deployment evidence;
- semantic/change events.
GEAIRA-M12-R50 - Provenance authority preservation: GEAIRA SHALL reuse or align with W3C PROV for generic provenance rather than defining semantically overlapping shadow provenance concepts.
GEAIRA-M12-R51 - Provenance/evidence distinction: A provenance assertion MAY support Evidence but SHALL NOT by itself establish that the evidence is sufficient, correct, current or fit for an assurance claim.
GEAIRA-M12-R52 - Cross-boundary provenance continuity: Where an interaction crosses GEAIRA and an adjacent architecture, provenance identifiers/links SHOULD remain correlatable across the boundary where material to traceability or assurance.
32.12 HDIP enterprise-realization mapping
HDIP is an enterprise-specific target conceptual/logical platform architecture for Data, Information and Intelligence. It may realize and specialize GEAIRA, GEDRA and GEARA responsibilities simultaneously. That breadth does not make HDIP a generic reference architecture or cause its plane/layer names to become GEAIRA core constructs.
A profile may map established HDIP constructs such as:
| HDIP construct | Illustrative GEAIRA alignment | Interpretation |
|---|---|---|
| Product Development Experience Plane (PDEP) | P1, P3, P5, P7 | Product development/qualification/compilation lifecycle may realize intent, mechanism, Product/deployment and assurance responsibilities. |
| Marketplace & Consumption Layer | P5 | Product/capability discovery, onboarding, serving and governed consumption. |
| Consumer Experience Plane (CEP) | P5 and conditional P6 | Consumer-facing interaction/consumption; decision/action controls remain P6 where applicable. |
| Consumer Observability & Value Plane (COVP) | P7 | Usage, outcome, value, feedback and continuous-assurance evidence. |
| Governance Plane | P1, P7 and X1-X8 | Policy/governance/metadata/lineage/control/assurance responsibilities across the architecture. |
| Infrastructure / Utility Plane | Realization of multiple planes | Compute, storage, networking, IAM, messaging, hosting, resilience and other enabling technology; not a GEAIRA plane. |
This mapping is intentionally illustrative and profile-level. A future HDIP GEAIRA profile MAY be more precise while preserving GEAIRA and adjacent-specification authority.
GEAIRA-M12-R53 - HDIP profile boundary: HDIP MAY realize or specialize GEAIRA responsibilities, but HDIP plane/layer names SHALL NOT become GEAIRA core planes solely because one enterprise uses them.
GEAIRA-M12-R54 - Multi-architecture realization: One HDIP capability MAY simultaneously realize GEAIRA, GEDRA and GEARA responsibilities, provided each responsibility, authority and conformance claim remains independently traceable.
GEAIRA-M12-R55 - No platform-conformance shortcut: Deployment on HDIP SHALL NOT by itself establish GEAIRA, GEDRA, GEARA, BPS, AIPS or AIPDS conformance.
GEAIRA-M12-R56 - Enterprise-profile explicitness: An HDIP mapping that adds registries, services, lifecycle gates, operating-model roles or platform-specific constructs SHALL identify them as profile/realization constructs rather than generic GEAIRA obligations.
32.13 Cross-specification version, currentness, compatibility and conformance
Adjacent specifications and their governed artifacts evolve independently. M12 therefore treats these dimensions separately:
GEAIRA version
GEDRA version
GEARA version
GEAIRM / GEDRM / GEARM semantic versions
BPS / AIPS / AIPDS versions
Agent Ontology version
PROV vocabulary version
enterprise profile / HDIP version
Product / Product Version
Interaction Contract version
Compatibility must be assessed where a mapping materially depends on one of these versions.
GEAIRA-M12-R57 - Independent version axes: Versions of adjacent specifications, Products, profiles and Interaction Contracts SHALL NOT be collapsed into one global version identifier.
GEAIRA-M12-R58 - Semantic compatibility over syntax: Structural or protocol compatibility SHALL NOT be treated as sufficient where an adjacent semantic or architectural change materially alters meaning, authority, control, evidence or lifecycle obligations.
GEAIRA-M12-R59 - Conformance independence: A claim of GEAIRA conformance SHALL NOT imply GEDRA, GEARA, BPS, AIPS, AIPDS, Agent Ontology, PROV-profile or HDIP conformance, and vice versa, unless a governed compound profile explicitly defines and validates such a combined claim.
GEAIRA-M12-R60 - Currentness propagation: Material adjacent-version/currentness changes SHALL be capable of reaching affected GEAIRA responsibilities within a latency appropriate to their risk/control context.
GEAIRA-M12-R61 - Mapping migration: Where an adjacent change breaks an existing mapping, the architecture SHALL define migration, degradation, suspension, reassessment or other governed disposition rather than silently continuing under a stale assumption.
32.14 Representative adjacent-interoperability scenarios
Scenario A - GEDRA Data Product supplies AI Context
GEDRA Data Product
| authoritative Data / contract / lineage / quality
v
P2 retrieval + grounding/source qualification
|
| M11-INT-11 AI Context Provision
v
P4 AI Execution
|
| M11-INT-29 Runtime Evidence
v
P7
The Data Product remains a GEDRA/BPS-governed Product. Its use does not convert it into an AI Product or AI Primitive.
Scenario B - GEARA analytical output informs a P6 Decision
GEARA Analytical Output / Information
|
| adjacent information contract
v
P6 Decision Formation
|
| M11-INT-25/26 as applicable
v
controlled Decision / Action path
Analytical Information informs the decision but does not carry organizational decision authority merely because it is authoritative Information.
Scenario C - AI Product served through GEAIRA and deployed through AIPDS
BPS Product identity
+
AIPS AI Product qualification / governance
|
v
GEAIRA P5 serving/product contract
|
v
AIPDS Product-version deployment binding
|
v
P4 execution / P7 evidence
Each layer retains its own authority and conformance claim.
Scenario D - Agentic execution with Agent Ontology identity
Agent Ontology identity/type
|
v
P4 Agent participation
|
+--> P1 delegated-authority context
+--> P6 Authorization where consequential action applies
+--> P7 evidence
Agent type/autonomy remains distinct from authority to act.
Scenario E - Cross-architecture provenance chain
GEDRA Data source
-> P2 context resource
-> P4 AI Execution
-> P6 Decision / Action
-> P7 Evaluation / Assurance
PROV-compatible provenance can link the chain, while evidence sufficiency remains a P7/assurance question rather than a PROV assertion.
Scenario F - HDIP realizes several generic architectures
HDIP enterprise platform/profile
/ | \
/ | \
GEDRA roles GEARA roles GEAIRA roles
Shared implementation does not collapse the three reference architectures or their conformance claims.
Scenario G - Adjacent specification version changes
Adjacent spec/profile v1
|
+--> historical interaction H1
later:
Adjacent spec/profile v2
|
+--> mapping reassessment
+--> future contract migration / restriction
H1 remains governed by the version/mapping actually applicable at the time.
32.15 M12 interoperability anti-patterns
M12 defines the following anti-patterns:
- Architecture annexation - treating a peer architecture's responsibilities as a GEAIRA sub-plane merely because AI consumes them.
- Semantic shadowing - minting local GEAIRA copies of BPS/AIPS/Agent/GEDRM/GEARM/PROV concepts.
- Plane-name equivalence - mapping similarly named planes one-to-one without responsibility analysis.
- Conformance laundering - treating conformance to one specification/platform as proof of conformance to all adjacent specifications.
- Shared-platform collapse - assuming co-hosted responsibilities belong to one architecture.
- Universal namespace - forcing independent authorities into one namespace for implementation convenience.
- Data-Product-to-AI-Product promotion - inferring AI Product status because a Data Product supports AI.
- Information-Product-to-AI-Product promotion - inferring AI Product status because Information is AI-assisted.
- Analytics-equals-AI - treating every analytical runtime or predictive output as GEAIRA AI by default.
- Agent-equals-primitive - collapsing Agent identity into the AI Primitive mechanism model.
- Autonomy-equals-authority - transferring Agent/Product autonomy classification into permission to act.
- PROV-equals-assurance - treating provenance presence as sufficient assurance evidence.
- AIPDS-equals-GEAIRA - treating one Product deployment specification as the generic reference architecture.
- HDIP-equals-GEAIRA - treating an enterprise target platform's topology as the generic GEAIRA topology.
- Gateway-equals-interoperability - declaring an API/event/model gateway to be the architecture boundary without an M11 contract.
- Version-latest shortcut - assuming the newest adjacent version/profile is automatically the currently governed compatible version.
- Hidden external dependency - relying on an adjacent authority/resource without traceable identity/currentness/failure semantics.
- Externality bypass - allowing external/provider/peer integration to bypass P1/P6/P7 controls solely because the implementation sits outside the enterprise boundary.
32.16 Frozen M12 decisions
M12 freezes the following 0.1 architecture decisions:
GEAIRA interoperability != authority transfer
adjacent mapping != semantic identity
peer plane != GEAIRA plane one-to-one
co-realization != architecture equivalence
shared platform != shared jurisdiction
GEAIRA != GEDRA != GEARA
Data Product != AI Product
Information Product != AI Product
Analytics != AI
Agent != AI Primitive
Provenance != Assurance
BPS owns generic Product semantics
AIPS owns AI Product / AI classification / risk-governance semantics
AIPDS specializes qualified AI Product deployment
Agent Ontology owns Agent identity/taxonomy
PROV owns generic provenance semantics
HDIP may realize multiple generic architectures
HDIP topology != GEAIRA core topology
cross-boundary runtime exchange -> M11 Interaction Contract discipline
schema/protocol compatibility != semantic/architectural compatibility
conformance claims remain independent unless an explicit compound profile says otherwise
historical boundary/version truth survives later migration or supersession
The following M12 structures are canonical human-readable artifacts:
5 interoperability boundary types
16-dimensional Adjacent Architecture Interoperability Record
9 canonical adjacent boundaries (M12-BND-00..08)
GEDRA/GEARA responsibility-intersection matrices
BPS/AIPS/AIPDS/Agent/PROV/HDIP jurisdiction mappings
7 representative scenarios
18 interoperability anti-patterns
M12 does not freeze:
- a common machine-readable mapping language;
- a universal cross-spec registry;
- a universal enterprise architecture metamodel;
- a shared implementation platform;
- one-to-one peer-plane equivalence;
- an HDIP-specific GEAIRA profile;
- M13 deployment-neutral realization patterns.
GEAIRA-M12-R62 - Frozen-boundary preservation: Later slices SHALL preserve M12-BND-00 through M12-BND-08 and their authority separation unless a reconciled 0.1 architecture change explicitly replaces a mapping without weakening jurisdiction controls.
GEAIRA-M12-R63 - Pattern reuse: M13-M15 patterns/views SHALL reuse the M12 boundary catalogue and M11 Interaction Contract model where adjacent architectures participate rather than inventing hidden cross-specification dependencies.
GEAIRA-M12-R64 - Conformance hand-off: M16 architecture invariants/conformance SHALL distinguish GEAIRA conformance from adjacent-specification conformance and SHALL make required adjacent mappings/evidence explicit where applicable.
GEAIRA-M12-R65 - M13 readiness: M13 SHALL demonstrate technology- and topology-neutral realization patterns for the P1-P7, M11 and M12 obligations across centralized, distributed, federated, edge/physical and mixed environments without converting enterprise-specific implementations into generic requirements.
33. M12 Completion State
GEAIRA-M12 - COMPLETE.
M12 establishes:
- five canonical interoperability boundary types;
- one sixteen-dimensional Adjacent Architecture Interoperability Record;
- nine canonical adjacent boundaries, M12-BND-00 through M12-BND-08;
- explicit GEDRA/GEDRM peer-architecture and Data-authority mappings;
- explicit GEARA/GEARM peer-architecture and Analytics/Information-authority mappings;
- explicit BPS/PIN Product-authority mappings;
- explicit AIPS AI Product/primitive/capability/risk-governance mappings;
- explicit AIPDS qualified-AI-Product deployment specialization rules;
- explicit Agent Ontology identity/taxonomy mappings;
- explicit W3C PROV provenance alignment;
- explicit HDIP enterprise-realization/profile rules;
- cross-specification version/currentness/compatibility/conformance rules;
- seven representative adjacent-interoperability scenarios;
- eighteen interoperability anti-patterns;
- 65 normative M12 rules.
M12 introduces no change to GEAIRM, GEDRM, GEDRA, GEARM, GEARA, BPS, AIPS, AIPDS, Agent Ontology or W3C PROV semantics, does not create a shared namespace or universal platform, does not make HDIP topology generic, and does not imply conformance to one specification from conformance to another.
Next slice
GEAIRA-M13 - Deployment-Neutral Realization Patterns
M13 will demonstrate how the frozen P1-P7 responsibilities, canonical ABBs, M11 interaction contracts and M12 adjacent-boundary mappings can be realized through multiple valid topologies - centralized, distributed, federated, edge/physical and mixed - without turning any one implementation pattern into the definition of GEAIRA.
34. GEAIRA-M13 - Deployment-Neutral Realization Patterns
34.1 Purpose
M13 defines a canonical set of deployment-neutral realization patterns for the architecture frozen through M12.
The purpose is not to prescribe a preferred platform shape. It is to demonstrate that the same GEAIRA architectural responsibilities can remain valid when enterprise AI is realized through materially different deployment topologies.
M13 therefore answers:
How can one GEAIRA responsibility model be realized centrally, distributively, federatively, at the edge, inside embedded environments, through external providers, or through mixed topologies without changing what GEAIRA means?
The M13 pattern catalogue is downstream of:
GEAIRM semantic baseline
|
v
P1-P7 responsibilities / 21 canonical ABBs
|
v
M11 Interaction Contract model / 33 canonical interactions
|
v
M12 adjacent-authority boundaries
|
v
M13 deployment-neutral realization patterns
GEAIRA-M13-R01 - Architecture/topology separation: A conformant GEAIRA architecture SHALL preserve the applicable P1-P7 responsibilities, ABB boundaries, Interaction Contract obligations and adjacent-authority boundaries independently of the selected deployment topology.
GEAIRA-M13-R02 - No preferred topology: GEAIRA 0.1 SHALL NOT treat centralized, distributed, federated, edge, embedded, managed-provider or hybrid realization as intrinsically more conformant or mature than another.
GEAIRA-M13-R03 - Pattern-not-product rule: Selection of an M13 realization pattern SHALL NOT by itself imply Product, AI Product, service, platform or deployment-environment classification.
GEAIRA-M13-R04 - Pattern composition: A real enterprise architecture MAY compose more than one M13 realization pattern, provided the resulting responsibility, authority, contract, evidence and failure boundaries remain explicit.
34.2 Realization metamodel
M13 introduces five architecture-level constructs for reasoning about realization without making them ontology concepts:
| Construct | Meaning |
|---|---|
| Realization Pattern | A reusable topology family showing how GEAIRA responsibilities may be arranged operationally. |
| Realization Unit | A deployable, hosted, embedded, human-operated or externally managed implementation unit that realizes one or more GEAIRA responsibilities/ABBs. |
| Realization Boundary | A material deployment, ownership, trust, locality, failure or administration boundary across which M11 contract obligations may become externally visible. |
| Realization Mapping | Traceable mapping between GEAIRA responsibilities/ABBs and the Realization Units that implement them. |
| Topology Profile | A governed specialization selecting/configuring one or more M13 patterns for a defined enterprise/domain/context. |
A Realization Unit may be:
- a process;
- application;
- container/workload;
- managed service;
- appliance;
- embedded runtime;
- device/controller;
- human/governance process;
- shared platform capability;
- external provider capability;
- or another implementation form.
These implementation forms do not become generic ABB types merely because they realize an ABB.
GEAIRA-M13-R05 - Realization-unit non-equivalence: A Realization Unit SHALL NOT be treated as semantically or architecturally identical to the ABBs it realizes.
GEAIRA-M13-R06 - Many-to-many realization: GEAIRA SHALL permit one Realization Unit to realize multiple ABBs and one ABB to be realized by multiple Realization Units.
GEAIRA-M13-R07 - Responsibility traceability: Every realization claiming GEAIRA alignment SHALL be able to trace applicable canonical responsibilities to the Realization Units, human processes or adjacent architectures that satisfy them.
GEAIRA-M13-R08 - Boundary explicitness: Material changes of trust, administrative authority, deployment locality, ownership, failure domain or external-provider control SHOULD be represented as explicit Realization Boundaries.
GEAIRA-M13-R09 - Contract preservation: Where an M11 canonical interaction crosses a material Realization Boundary, the applicable M11 Interaction Contract obligations SHALL remain explicit even when the implementation uses an internal call, shared memory, local function, event, file, queue, protocol or human workflow.
GEAIRA-M13-R10 - No hidden co-location: Co-location of producer and consumer responsibilities SHALL NOT be used to erase authority, evidence, identity/currentness or failure obligations that would remain material if the responsibilities were separated.
GEAIRA-M13-R11 - Profile specialization: A Topology Profile MAY tighten locality, resilience, security, latency, evidence or deployment constraints but SHALL NOT weaken baseline GEAIRA responsibilities or redefine GEAIRM semantics.
GEAIRA-M13-R12 - Deployment description scope: M13 topology constructs describe realization arrangement only; they SHALL NOT be used as substitute concepts for Product identity, Agent identity, AI Primitive type, Intelligence Capability, Execution Mode or Environment Kind.
34.3 Deployment-realization dimensions
M13 evaluates patterns across eleven independent dimensions:
| Dimension | Typical questions |
|---|---|
| Locality | Central site, regional, domain-local, edge, device, external provider? |
| Administrative control | One operator, multiple domains, partner/federated, managed provider? |
| Trust boundary | Where do identity, authority, privacy or security assumptions change? |
| Execution placement | Where does P4 AI Execution occur? |
| Context/data placement | Where are P2 resources and adjacent GEDRA/GEARA responsibilities realized? |
| Mechanism placement | Where are P3 primitives provisioned/hosted? |
| Decision/action placement | Where are P6 decision, Authorization and oversight controls applied? |
| Evidence placement | Where are P7 evidence, evaluation, assurance and currentness responsibilities realized? |
| Connectivity posture | Continuous, intermittent, asynchronous, disconnected, safety-local? |
| Failure domain | Which responsibilities fail together and where must graceful degradation exist? |
| Lifecycle/control propagation | How do policy, revocation, currentness and change reach distributed realizations? |
These dimensions are deliberately orthogonal. For example, an edge execution can use centrally governed authority, locally cached context, externally hosted primitives and federated assurance evidence.
GEAIRA-M13-R13 - Dimension independence: Locality, ownership, trust, connectivity, lifecycle and failure-domain choices SHALL NOT be collapsed into one topology label.
GEAIRA-M13-R14 - Locality/authority separation: Physical or network proximity SHALL NOT imply semantic authority, Product ownership, decision authority or Authorization.
GEAIRA-M13-R15 - Connectivity-aware control: A realization with intermittent/disconnected connectivity SHALL define how current governing constraints, revocation, currentness and evidence obligations remain safely satisfied during disconnection.
GEAIRA-M13-R16 - Failure-domain awareness: Realization mapping SHALL identify material correlated-failure domains where co-location can cause several GEAIRA responsibilities or controls to fail together.
GEAIRA-M13-R17 - Evidence continuity: Distribution SHALL NOT remove the obligation to correlate execution, decision/action, serving, currentness and assurance evidence where those correlations are material.
GEAIRA-M13-R18 - No centralization inference: A cross-cutting overlay or shared responsibility SHALL NOT imply a centralized implementation.
GEAIRA-M13-R19 - No distribution inference: Domain ownership, federation or multiple Products SHALL NOT imply that every architectural responsibility must be physically distributed.
GEAIRA-M13-R20 - Currentness propagation: Distributed or cached realization state SHALL have a governed mechanism for receiving material policy, authority, version/currentness, assurance or revocation changes within latency appropriate to the context.
34.4 Canonical realization pattern catalogue
M13 freezes eight canonical topology families. They are intentionally broad and MAY be specialized by profiles.
M13-RP-01 Consolidated / Co-Located Realization
M13-RP-02 Distributed Service / Capability Realization
M13-RP-03 Federated Domain / Organizational Realization
M13-RP-04 Edge / Physical Closed-Loop Realization
M13-RP-05 Embedded / Intermittently Connected Realization
M13-RP-06 External / Managed-Provider Realization
M13-RP-07 Hybrid Multi-Locality Realization
M13-RP-08 Shared-Platform Multi-Architecture Realization
The patterns do not represent maturity levels. RP-01 is not "less mature" than RP-03, and RP-07 is not automatically superior because it is more complex.
34.5 M13-RP-01 - Consolidated / Co-Located Realization
Intent
Use one primary platform/runtime or a tightly co-located set of implementation units to realize many GEAIRA responsibilities.
Representative shape:
+-----------------------------------------------------------+
| Consolidated enterprise AI realization |
| |
| P1/P2/P3/P4/P5/P6/P7 responsibilities |
| may be implemented by one platform or tightly coupled |
| realization set |
+-----------------------------------------------------------+
| | |
consumers humans adjacent authorities
This pattern can be valid for smaller enterprises, bounded platforms, controlled environments or highly integrated operating contexts.
Its main architecture risk is co-location collapse: assuming that because several responsibilities run in one platform, their authority and logical boundaries no longer matter.
GEAIRA-M13-R21 - Co-location preservation: RP-01 SHALL preserve logical P1-P7 responsibility and ABB distinctions even where several are implemented in one process/platform.
GEAIRA-M13-R22 - Internal-contract visibility: Material M11 contract obligations SHALL remain testable/auditable in RP-01 even when realized as in-process or shared-state interactions rather than network calls.
GEAIRA-M13-R23 - Correlated-failure treatment: RP-01 SHOULD explicitly assess correlated failure created by consolidating authority, execution, decision/action and evidence responsibilities into a common failure domain.
GEAIRA-M13-R24 - No platform-equals-conformance: Operation on a consolidated "AI Platform" SHALL NOT by itself establish GEAIRA conformance.
34.6 M13-RP-02 - Distributed Service / Capability Realization
Intent
Distribute GEAIRA responsibilities across separately deployed Realization Units while retaining a single coherent architectural authority model.
Representative shape:
P1 unit P2 units P3 units
\ | /
\ | /
M11 contracts
|
P4 units ---- P5 units
| |
+---- P6 -----+
|
P7
This pattern may use services, events, workloads or platform capabilities, but GEAIRA does not prescribe microservices.
GEAIRA-M13-R25 - Distributed-contract explicitness: RP-02 SHALL make material cross-unit M11 contracts explicit, including identity/correlation, version/currentness, failure and evidence obligations.
GEAIRA-M13-R26 - Distribution-not-microservices: RP-02 SHALL NOT require a microservice-per-ABB or service-per-responsibility implementation.
GEAIRA-M13-R27 - Partial-failure semantics: RP-02 SHALL define degraded/partial/indeterminate behavior for material cross-unit failures rather than assuming atomic success across distributed interactions.
GEAIRA-M13-R28 - Distributed lifecycle: Contract, policy, primitive, context and assurance changes SHALL be propagated without requiring historical exchanges to be rewritten.
34.7 M13-RP-03 - Federated Domain / Organizational Realization
Intent
Allow multiple domains, business units, legal entities or partner organizations to realize GEAIRA responsibilities under distributed administrative authority while preserving federated contracts and explicit jurisdiction.
Representative shape:
Domain A GEAIRA realization
|\
| \ governed federation contracts
| \
| Domain B GEAIRA realization
| |
+------------+---- Shared / peer responsibilities
|
P7 / governance federation
Federation may involve separate Products, Agents, authority domains, data jurisdictions, providers or assurance regimes.
GEAIRA-M13-R29 - Federation authority preservation: RP-03 SHALL preserve the distinction between federated interoperability and transfer of semantic, Product, decision or governance authority.
GEAIRA-M13-R30 - Federated contract completeness: Cross-domain interaction SHALL state the applicable authority, semantic mapping, contract version/currentness, evidence and failure obligations rather than rely on organizational trust alone.
GEAIRA-M13-R31 - Local control with global obligations: A federated participant MAY realize controls locally, provided required enterprise/federation obligations and evidence remain demonstrably satisfied.
GEAIRA-M13-R32 - No central-federation paradox: RP-03 SHALL NOT require a single central implementation for every cross-cutting concern merely to call the architecture federated.
34.8 M13-RP-04 - Edge / Physical Closed-Loop Realization
Intent
Realize material P4/P6 responsibilities close to a physical or low-latency Environment while retaining governed linkage to P1/P2/P3/P5/P7 responsibilities.
Representative shape:
Central / regional governance, context, primitive and assurance
| | |
| governed synchronization |
v v v
+------------------------------------------------+
| Edge / physical realization |
| P4 Execution + bounded P6 control |
| local context/primitives as required |
| sensing -> decision/control -> actuation |
+------------------------------------------------+
|
v
Physical / mixed Environment
GEAIRA-M13-R33 - Safety-locality allowance: RP-04 MAY realize latency- or safety-critical Authorization/control locally when the governing envelope, constraints, currentness and evidence obligations are preserved.
GEAIRA-M13-R34 - No remote-call mandate: GEAIRA SHALL NOT require hard-real-time or safety-critical loops to make a synchronous remote call to a centralized P1, P6 or P7 component.
GEAIRA-M13-R35 - Edge stale-state control: RP-04 SHALL define safe behavior for stale, unavailable or revoked governing/context/primitive state, including local safe-state or bounded degraded operation where applicable.
GEAIRA-M13-R36 - Physical-effect evidence: Material actuation/control outcomes SHALL preserve enough occurrence evidence to distinguish commanded, attempted, committed, observed and indeterminate physical effects where relevant.
34.9 M13-RP-05 - Embedded / Intermittently Connected Realization
Intent
Realize AI responsibilities inside an application, device, appliance or constrained local environment that may be intermittently connected or intentionally disconnected.
Representative shape:
Governed package/profile at synchronization time
|
v
+-----------------------------------------------+
| Embedded realization |
| local P2/P3/P4 and conditional P6/P7 duties |
| cached governing/currentness state |
+-----------------------------------------------+
^ |
| periodic evidence/change|
+-------------------------+
GEAIRA-M13-R37 - Disconnected governance: RP-05 SHALL define the maximum acceptable age/scope of cached governing, primitive, context and assurance state for continued operation.
GEAIRA-M13-R38 - Revocation-latency explicitness: Where immediate revocation cannot reach an embedded realization, the architecture SHALL make the revocation latency and compensating safety/control behavior explicit.
GEAIRA-M13-R39 - Evidence buffering: Required evidence MAY be buffered locally during disconnection but SHALL preserve identity, ordering/time semantics, integrity and eventual reconciliation expectations appropriate to the context.
GEAIRA-M13-R40 - Embedded-does-not-mean-uncontrolled: Resource constraint or offline operation SHALL NOT be used to silently omit applicable P1/P6/P7 obligations.
34.10 M13-RP-06 - External / Managed-Provider Realization
Intent
Use an external provider or managed platform to realize one or more GEAIRA responsibilities while the enterprise retains responsibility for its own governing, contract and assurance obligations.
Representative shape:
Enterprise P1/P5/P6/P7 responsibilities
|
| M11 + M12 boundary contract
v
+---------------------------------------+
| External / Managed Provider |
| may realize P2/P3/P4/P5/P7 portions |
+---------------------------------------+
GEAIRA-M13-R41 - Outsourcing-not-authority-transfer: RP-06 SHALL NOT treat provider operation as automatic transfer of enterprise purpose, decision authority, Product accountability, governance or assurance responsibility.
GEAIRA-M13-R42 - External boundary contract: Material provider interaction SHALL use M11 contract discipline plus the applicable M12 authority/interoperability mapping.
GEAIRA-M13-R43 - Provider substitution safety: Provider substitution SHALL assess semantic/capability/effect/control/evidence equivalence and SHALL NOT rely only on protocol compatibility.
GEAIRA-M13-R44 - Provider evidence sufficiency: Where assurance depends on provider-internal behavior or controls, the architecture SHALL obtain sufficient attributable evidence or explicitly record the resulting assurance limitation.
34.11 M13-RP-07 - Hybrid Multi-Locality Realization
Intent
Combine centralized, distributed, edge, embedded and external-provider realizations in one end-to-end architecture.
Representative shape:
Central enterprise controls / context / assurance
| \
| \ provider boundary
domain runtime \---- external capability
|
edge runtime
|
physical Environment
RP-07 is expected to be common in large enterprises, but complexity alone does not make it a preferred architecture.
GEAIRA-M13-R45 - Hybrid explicitness: RP-07 SHALL identify which responsibilities/ABBs are realized at each locality and which M11/M12 contracts cross locality, trust or provider boundaries.
GEAIRA-M13-R46 - No weakest-link assumption hiding: Different localities MAY use different technologies or control implementations, but equivalent architecture obligations SHALL remain explicit where the same responsibility applies.
GEAIRA-M13-R47 - End-to-end correlation: RP-07 SHALL preserve material identity/evidence correlation across locality boundaries without requiring one universal runtime identifier.
GEAIRA-M13-R48 - Failure-containment design: RP-07 SHOULD define which operations remain available, degrade, fail closed, enter safe state or require reconciliation when one locality/provider becomes unavailable.
34.12 M13-RP-08 - Shared-Platform Multi-Architecture Realization
Intent
Allow one enterprise platform capability to realize responsibilities from GEAIRA and adjacent architectures such as GEDRA or GEARA without collapsing their architectural jurisdictions.
Representative shape:
GEAIRA responsibilities
\ |
\ |
Shared Platform Capability
/ |
/ |
GEDRA GEARA
responsibilities responsibilities
HDIP is one possible enterprise-specific example of this pattern, but RP-08 is generic and is not defined by HDIP.
GEAIRA-M13-R49 - Multi-architecture traceability: RP-08 SHALL keep GEAIRA, GEDRA, GEARA and other applicable architecture responsibilities independently traceable even where one platform capability realizes several of them.
GEAIRA-M13-R50 - Shared-platform non-annexation: Co-realization on a shared platform SHALL NOT transfer semantic or architecture authority between GEAIRA and adjacent specifications.
GEAIRA-M13-R51 - Independent conformance: Platform conformance or certification under one architecture/profile SHALL NOT automatically establish conformance to the others.
GEAIRA-M13-R52 - Enterprise profile disclosure: Enterprise-specific registries, planes, services, lifecycle gates or operating-model constructs used by RP-08 SHALL be identified as realization/profile constructs rather than generic GEAIRA requirements.
34.13 Pattern comparison matrix
| Pattern | Primary distribution characteristic | Typical control posture | Principal architecture risk |
|---|---|---|---|
| RP-01 Consolidated | Co-located | Central/co-located logical controls | Logical-boundary collapse; correlated failure |
| RP-02 Distributed | Separate deployable units | Distributed contracts/controls | Partial failure; hidden contract assumptions |
| RP-03 Federated | Separate administrative domains | Federated/local controls | Authority leakage; incompatible governance/currentness |
| RP-04 Edge/Physical | Low-latency/physical locality | Local bounded safety/control | Stale central state; unsafe remote dependency |
| RP-05 Embedded/Disconnected | Device/application local | Cached/bounded controls | Revocation latency; evidence reconciliation |
| RP-06 External Provider | Provider-managed | Enterprise boundary + provider controls | Outsourced opacity; false equivalence |
| RP-07 Hybrid | Multiple localities/topologies | Mixed/federated | Complexity; cross-boundary currentness/failure |
| RP-08 Shared Platform | Multi-architecture co-realization | Shared implementation, separate obligations | Jurisdiction/conformance collapse |
No row in this matrix is a maturity tier.
GEAIRA-M13-R53 - No maturity ordering: M13 pattern identifiers SHALL NOT be interpreted as progression levels, maturity scores or preferred migration sequence.
GEAIRA-M13-R54 - Same obligations, different realization: Selection of a realization pattern MAY change how an obligation is implemented but SHALL NOT silently remove an applicable architectural obligation.
GEAIRA-M13-R55 - Pattern-specific strengthening: A pattern or Topology Profile MAY introduce stricter resilience, locality, evidence, safety, latency or lifecycle requirements where required by its context.
GEAIRA-M13-R56 - Architecture review basis: Architecture review SHOULD evaluate the appropriateness of a selected realization pattern against context, risk, latency, jurisdiction, resilience, operational capability and evidence needs rather than against fashion or platform preference.
34.14 Pattern composition and migration
Real systems frequently compose patterns.
Example:
RP-03 federated domains
|
+-- Domain A: RP-01 consolidated
|
+-- Domain B: RP-02 distributed
|
+-- Factory: RP-04 edge/physical
|
+-- Device fleet: RP-05 embedded
|
+-- External foundation capability: RP-06 provider
The architecture remains one GEAIRA-aligned responsibility system if contracts, authority, evidence, identity and lifecycle remain traceable.
Migration between patterns is also allowed. A capability may move from consolidated to distributed, from provider-hosted to enterprise-hosted, or from cloud-only to hybrid/edge without changing the semantic identity of the underlying enterprise-AI concepts solely because topology changed.
GEAIRA-M13-R57 - Composition traceability: When patterns are composed, each material Realization Boundary SHALL remain traceable to the responsibilities and contracts crossing it.
GEAIRA-M13-R58 - Migration identity preservation: A topology migration SHALL NOT by itself create a new Product, Agent, AI Primitive, Intelligence Capability or semantic identity unless the authoritative semantic/Product rules establish such a change.
GEAIRA-M13-R59 - Migration-currentness control: Pattern migration SHALL preserve or explicitly re-establish governing envelope, contract version, primitive/context eligibility, Authorization behavior, evidence and assurance currentness before the new realization is treated as governed-current.
GEAIRA-M13-R60 - Historical realization truth: After migration, historical records SHALL continue to identify the realization pattern/version and material implementation context that actually governed earlier occurrences rather than projecting the new topology backwards.
34.15 Representative M13 scenarios
Scenario A - Consolidated internal AI capability
A small enterprise hosts context retrieval, primitive realization, AI Execution, serving, decision control and evidence collection inside one managed platform.
GEAIRA allows this as RP-01 provided the architecture still demonstrates:
- governing purpose/authority;
- bounded context and primitive eligibility;
- occurrence-level execution identity;
- point-of-use decision/action controls where applicable;
- contract/version/currentness;
- evidence/assurance responsibilities.
The platform itself is not the proof of conformance.
Scenario B - Distributed enterprise services
P2 context responsibilities, P3 primitive provisioning, P4 execution and P7 assurance are implemented by different enterprise services.
M11 contract obligations become explicit across those deployment boundaries. A timeout between P4 and P6 cannot be flattened into failure if a consequential Authorization/action outcome may already have occurred.
Scenario C - Federated business domains
Two domains each operate their own P2-P5 realization. Enterprise policy and assurance are federated.
Cross-domain AI use remains valid only where semantic mappings, authority, Product identity where applicable, contract currentness and evidence responsibilities are explicit. Domain autonomy does not mean authority transfer.
Scenario D - Physical AI at the edge
A robotic system must sense and act at millisecond latency. P4 and bounded P6 controls execute locally, while P1 policy, P3 primitive packages and P7 assurance are synchronized from regional/central systems.
GEAIRA does not require a central Authorization Service call inside the hard-real-time loop. It requires the local control to remain inside a governed, current and revocable envelope with auditable evidence.
Scenario E - Intermittently connected device
An embedded AI device operates for hours without connectivity.
Its profile defines allowable cached authority/context/primitive age, local safe-state behavior, evidence buffering and reconciliation. The device does not become ungoverned simply because the central control plane is unreachable.
Scenario F - External managed AI provider
An enterprise uses an external generative AI provider behind P5/P4 integration.
Provider API compatibility alone does not make a substitute provider equivalent. Capability/effect differences, evidence access, privacy, currentness, safety, governance and assurance consequences must be assessed under M11/M12.
Scenario G - HDIP-style shared enterprise platform
One enterprise platform realizes GEAIRA AI responsibilities, GEDRA data responsibilities and GEARA analytical responsibilities.
This is RP-08. It is valid if the implementation keeps the three architecture obligations traceable. The platform's internal names, registries and planes remain enterprise-profile constructs rather than new GEAIRA core constructs.
Scenario H - Hybrid cloud, provider and factory edge
P1/P2/P3/P7 are primarily centralized, some primitives are provider-managed, P4/P6 control runs at factory edge, and P5 serves enterprise consumers from regional locations.
The architecture is RP-07 composed with RP-04 and RP-06. Conformance depends on preserved responsibility/contract/control/evidence semantics, not on any single topology label.
34.16 M13 realization anti-patterns
M13 defines the following anti-patterns:
- Topology-as-architecture - treating a deployment drawing as sufficient proof of GEAIRA responsibility coverage.
- Central-is-governed - assuming a centralized platform is governed merely because it is centrally operated.
- Distributed-is-modern - distributing ABBs solely to appear cloud-native or mature.
- Microservice-per-ABB - assigning each canonical ABB its own service by default.
- Plane-per-runtime - creating one deployable runtime per GEAIRA plane.
- Shared-process collapse - erasing logical authority/control/evidence boundaries because responsibilities execute in one process.
- Remote-control reflex - requiring safety/real-time edge loops to depend synchronously on central services.
- Offline-means-exempt - omitting governance/currentness/evidence obligations for disconnected systems.
- Federation-equals-trust - treating organizational federation as proof of semantic, authority or control compatibility.
- Provider-equals-outsourced-accountability - assuming a managed provider owns the enterprise's purpose, decision or assurance obligations.
- Protocol-equals-substitutability - treating provider/runtime substitution as safe because schemas or APIs match.
- One-global-ID - forcing one identifier across requests, executions, decisions, actions and evidence for convenience.
- Central-registry mandate - requiring one enterprise registry merely because several distributed realizations require discovery/currentness.
- Evidence-afterthought - designing distributed execution without a coherent evidence/correlation model.
- Latest-config-wins - projecting current policy/configuration backwards onto historical occurrences.
- Shared-platform annexation - treating co-hosted GEDRA/GEARA/GEAIRA responsibilities as one architecture.
- Topology-migration-equals-new-semantic-identity - changing semantic/Product identity solely because deployment moved.
- Complexity-equals-maturity - preferring hybrid/federated topology when a simpler realization satisfies the same obligations more safely.
34.17 Frozen M13 decisions
M13 freezes the following 0.1 architecture decisions:
GEAIRA architecture != deployment topology
Realization Pattern != maturity level
Realization Unit != ABB
co-location != semantic/architecture equivalence
distribution != microservices
federation != authority transfer
edge locality != authority
offline operation != governance exemption
provider operation != enterprise-accountability transfer
shared platform != shared jurisdiction
schema/API compatibility != safe substitutability
migration != semantic identity change
The following M13 structures are canonical human-readable architecture artifacts:
5 realization metamodel constructs
11 deployment-realization dimensions
8 canonical realization patterns (M13-RP-01..08)
pattern comparison matrix
8 representative realization scenarios
18 realization anti-patterns
60 normative M13 rules
M13 does not freeze:
- a vendor/cloud deployment architecture;
- a universal control plane;
- a service mesh or event bus;
- a mandatory central registry;
- a mandatory microservice topology;
- an HDIP-specific GEAIRA profile;
- implementation sizing/capacity;
- M14 canonical architecture problem/solution patterns;
- M15 reference views/end-to-end flows;
- M16 conformance/invariants.
35. M13 Completion State
GEAIRA-M13 - COMPLETE.
M13 establishes:
- five realization-metamodel constructs;
- eleven independent deployment-realization dimensions;
- eight canonical realization patterns, M13-RP-01 through M13-RP-08;
- explicit centralized/co-located, distributed, federated, edge/physical, embedded/disconnected, external-provider, hybrid and shared-platform realization families;
- preservation rules for all seven planes, twenty-one canonical ABBs, thirty-three M11 canonical inter-plane interactions and nine M12 adjacent boundaries;
- explicit many-to-many ABB/Realization Unit mapping;
- realization-boundary and topology-profile rules;
- topology-aware failure, currentness, authority and evidence obligations;
- eight representative realization scenarios;
- eighteen realization anti-patterns;
- 60 normative M13 rules.
M13 introduces no new GEAIRM semantic concepts or relationships, no new canonical GEAIRA ABBs, no new M11 interactions and no new adjacent-authority boundaries. It does not make any one topology the definition of GEAIRA.
Next slice
GEAIRA-M14 - Canonical Architecture Patterns & Anti-Patterns
M14 will move from where/how responsibilities may be deployed to recurring architecture problem/solution patterns. It will define canonical positive patterns and boundary counterexamples using the frozen P1-P7 responsibility model, 21 ABBs, M11 contract catalogue, M12 adjacent boundaries and M13 realization neutrality.
36. GEAIRA-M14 - Canonical Architecture Patterns & Anti-Patterns
36.1 Purpose
M01-M13 established the GEAIRA architecture vocabulary, seven responsibility planes, twenty-one canonical ABBs, thirty-three canonical inter-plane interactions, adjacent-architecture interoperability boundaries and deployment-neutral realization patterns. M14 now turns those frozen constructs into a reusable architecture pattern language for recurring enterprise-AI problems.
M14 answers six questions:
- Which recurring enterprise-AI architecture problems deserve a canonical positive pattern?
- How should the frozen P1-P7 responsibilities and ABBs be combined without creating new semantics or a mandatory pipeline?
- Which M11 contracts are material to each pattern?
- Which M12 authority/interoperability boundaries are relevant without transferring jurisdiction?
- How can the same pattern remain valid across M13 realization topologies?
- Which tempting architecture shortcuts or conflations must be rejected as canonical anti-patterns?
M14 introduces no new GEAIRM semantic concept, no new GEAIRA plane, no new canonical ABB and no new M11 inter-plane interaction. A pattern is a reusable arrangement of already-frozen responsibilities, ABBs, interactions, controls and adjacent-authority boundaries under stated applicability conditions.
M14 signature proposition: A canonical pattern constrains how existing responsibilities cooperate; it does not manufacture new semantic identity, authority, Product status or deployment topology.
GEAIRA-M14-R01 - Pattern subordination: Every M14 pattern SHALL be interpreted under the normative semantics and architecture rules frozen in M01-M13 and SHALL NOT override them.
GEAIRA-M14-R02 - No semantic expansion by pattern: A label, participant, hand-off or implementation example appearing only in M14 SHALL NOT become a new GEAIRM concept, GEAIRA ABB, Product type, Agent type, risk tier or canonical interaction solely because it appears in a pattern.
GEAIRA-M14-R03 - Deployment neutrality: Every canonical M14 pattern SHALL remain realizable through one or more applicable M13 realization patterns unless its stated problem necessarily requires a specific physical property such as local actuation.
GEAIRA-M14-R04 - No mandatory pipeline: Pattern diagrams SHALL express responsibility/causal relationships rather than impose P1->P7 numeric plane order as a universal runtime sequence.
36.2 Canonical pattern record
Each M14 canonical pattern uses the following record.
| Field | Meaning |
|---|---|
| Pattern ID / Name | Stable human-readable identifier and name |
| Problem | Recurring architecture problem addressed |
| Applicability | Conditions under which the pattern is relevant |
| Forces | Tensions the architecture must balance |
| Participating planes / ABBs | Primary frozen GEAIRA responsibilities used by the pattern |
| Canonical interactions | Material M11 interactions and plane-internal/external boundaries |
| Adjacent authorities | Relevant M12 authority/interoperability boundaries |
| Solution shape | Technology-neutral responsibility arrangement |
| Required controls/evidence | Minimum control, identity, currentness, evidence and failure obligations |
| Permitted realization | M13 realization families that may implement the pattern |
| Must not be concluded | Invalid inference or architecture collapse |
| Boundary counterexample | Deliberately unsafe/invalid architecture interpretation |
The pattern identifiers are documentation identifiers. Machine-readable identifiers remain part of the later executable-artifact programme after M17.
GEAIRA-M14-R05 - Frozen-component use: Canonical patterns SHALL use the frozen P1-P7 responsibilities, P1-P7 ABBs, M11 interactions and M12 boundaries rather than mint duplicate architecture constructs.
GEAIRA-M14-R06 - Applicability discipline: A pattern SHALL be applied only where its problem and applicability conditions exist; absence of a pattern from a solution SHALL NOT imply non-conformance where its conditions do not apply.
GEAIRA-M14-R07 - Pattern composition: Multiple M14 patterns MAY be composed in one architecture, provided responsibility identity, M11 contract identity, authority, evidence and failure semantics remain traceable.
GEAIRA-M14-R08 - Counterexample status: An M14 counterexample is invalid because it violates already-frozen GEAIRA or externally authoritative semantics; it SHALL NOT be treated as creating a new rule solely by example.
36.3 Pattern coverage matrix
M14 freezes fifteen canonical architecture patterns.
| Pattern | Primary concern |
|---|---|
| M14-P01 | Governed AI Invocation & Evidence Chain |
| M14-P02 | Grounded Context Execution |
| M14-P03 | Multi-Primitive Capability Realization |
| M14-P04 | AI Product Serving & Execution Correlation |
| M14-P05 | Human-Reviewed Recommendation & Approval |
| M14-P06 | Bounded Autonomous Digital Action |
| M14-P07 | Closed-Loop Physical AI |
| M14-P08 | Agent Uses Tool Under Bounded Authority |
| M14-P09 | Product / Capability Composition Without Identity Collapse |
| M14-P10 | External Managed AI Provider Mediation |
| M14-P11 | Federated Domain AI Consumption |
| M14-P12 | AI-Assisted Analytics Interoperability |
| M14-P13 | Data-Grounded AI Interoperability |
| M14-P14 | Continuous Assurance, Currentness & Governed Change |
| M14-P15 | Shared-Platform Multi-Architecture Co-Realization |
The catalogue intentionally spans simple advisory AI, AI Products, Agentic AI, autonomous digital action, Physical AI, peer architectures and enterprise platform realization.
36.4 M14-P01 - Governed AI Invocation & Evidence Chain
Problem. An enterprise needs to invoke AI while preserving governing purpose, eligible context/mechanism, execution occurrence identity and later evidence without requiring every control to be a separate service.
Applicability. Any governed AI Execution for which purpose/authority, primitive/context binding or evidence is material.
Participating architecture. P1-ABB-01/02, P2-ABB-01 where context is required, P3-ABB-01/02/03, P4-ABB-01 and P7-ABB-01/03.
Material interactions. M11-INT-01..06 as applicable, M11-INT-10/11, M11-INT-16/17, M11-INT-18/29 and M11-INT-32/33.
Solution shape. The architecture binds an invocation to a current governing envelope, eligible primitive realization and bounded context where applicable; P4 establishes the AI Execution occurrence; P7 receives sufficient evidence to reconstruct the material bindings and constrain future operation when assurance/currentness changes.
P1 governing envelope
| P2 bounded context
| |
+--------+-----------+
v
P3 eligible realization
|
v
P4 AI Execution
|
v
P7 evidence/assurance
Must not be concluded. A single platform request is not proof that purpose, context, primitive eligibility or evidence obligations were satisfied.
M14-C01 - Boundary counterexample. "The call reached the approved AI endpoint, therefore the execution was governed." Invalid: endpoint reachability does not establish governing-envelope currentness, context eligibility, primitive fitness or evidence sufficiency.
GEAIRA-M14-R09 - Governed invocation bindings: M14-P01 SHALL preserve sufficient identity/currentness correlation between governing envelope, applicable context, primitive realization and the AI Execution occurrence where those factors are material.
GEAIRA-M14-R10 - Endpoint non-equivalence: An approved endpoint, runtime or platform SHALL NOT by itself be treated as evidence that M14-P01 governing, eligibility or assurance obligations were satisfied.
36.5 M14-P02 - Grounded Context Execution
Problem. AI must use retrieved enterprise/knowledge resources as bounded context while distinguishing retrieval, Grounding Source qualification, context assembly, execution use and factual correctness.
Applicability. Retrieval-augmented, knowledge-grounded, memory-assisted or other context-enriched execution.
Participating architecture. P2-ABB-02 Knowledge & Retrieval Mediation, P2-ABB-03 Grounding & Source Qualification, P2-ABB-01 Context Assembly & Provision, P4-ABB-01 and P7-ABB-01/03.
Material interactions. M11-INT-01, 10, 11, 13, 14, 15 and 29; external resource/data/analytics authorities through M12 as applicable.
Solution shape. Candidate resources are retrieved under governed eligibility, selected resources are independently qualified for grounding where required, a bounded AI Context is assembled and supplied to the execution, and source/version/currentness evidence is retained.
resource authority
|
v
P2 retrieval -> grounding qualification -> bounded AI Context
|
v
P4 execution
|
v
P7 evidence
Must not be concluded. Retrieval does not automatically make a resource a Grounding Source, and grounding does not prove correctness or authority.
M14-C02 - Boundary counterexample. "The vector search returned the document, therefore the answer is grounded and correct." Invalid: repository technology, retrieval, Grounding Source role, source authority and answer correctness remain separate.
GEAIRA-M14-R11 - Grounding separation: M14-P02 SHALL preserve the distinction between resource retrieval, Grounding Source qualification, context inclusion and execution occurrence.
GEAIRA-M14-R12 - Technology non-entailment: Use of a vector database, knowledge graph, search engine or RAG framework SHALL NOT establish M14-P02 grounding/currentness/correctness semantics by itself.
36.6 M14-P03 - Multi-Primitive Capability Realization
Problem. One Intelligence Capability may require several AI Primitives, or several alternative primitive sets may realize an acceptable capability envelope.
Applicability. Hybrid AI, ensemble, symbolic+learned, planner+reasoner, fallback/substitution or composite primitive realization.
Participating architecture. P3-ABB-01/02/03, P4-ABB-01 and P7-ABB-02/03.
Material interactions. M11-INT-02, 16, 17, 18, 19, 20, 29 and 33.
Solution shape. P3 records capability requirements, candidate primitive contribution, eligibility/currentness/compatibility and the approved composition; P4 records the actual primitive participation in each execution; P7 assesses resulting fitness/evidence.
Must not be concluded. Same capability label does not make primitives substitutable, and the newest primitive version is not automatically the governed current one.
M14-C03 - Boundary counterexample. "Both providers claim Generation, so either model can replace the other without architecture review." Invalid: effect profile, evaluation evidence, latency, risk, compatibility and governance may differ materially.
GEAIRA-M14-R13 - Composition traceability: M14-P03 SHALL preserve the identity, version/currentness and material contribution of each primitive participating in an approved realization set.
GEAIRA-M14-R14 - Capability non-substitutability: Shared Intelligence Capability classification SHALL NOT by itself establish primitive substitutability or semantic/assurance equivalence.
36.7 M14-P04 - AI Product Serving & Execution Correlation
Problem. Consumers invoke an AI Product or other served AI capability, but consumer request, serving contract, Product Version and AI Execution occurrence must remain independently identifiable.
Applicability. API, event, batch, interactive, agent-facing or other governed consumption of an AI Product/capability.
Participating architecture. P5-ABB-01, conditional P5-ABB-02, P4-ABB-01, P1-ABB-01/02 and P7-ABB-01/03.
Material interactions. M11-INT-04, 07, 21, 22, 29, 30, 32 and 33; BPS/AIPS authority through M12 when Product semantics apply.
Solution shape. P5 admits a request under a current serving contract, binds authoritative Product/Version identity where valid, invokes P4, correlates but does not merge the resulting AI Execution identity, and returns a governed result/status to the consumer.
Must not be concluded. API exposure, marketplace listing or repeated reuse does not create Product or AI Product status.
M14-C04 - Boundary counterexample. "This internal endpoint serves an LLM, therefore it is an AI Product." Invalid: serving technology does not bypass BPS Product qualification or AIPS AI Product qualification.
GEAIRA-M14-R15 - Serving/execution identity separation: M14-P04 SHALL correlate consumer request, serving interaction, Product Version where applicable and AI Execution without asserting they are the same occurrence.
GEAIRA-M14-R16 - Product authority preservation: M14-P04 SHALL consume BPS/AIPS Product semantics where applicable and SHALL NOT infer Product or AI Product status from serving/exposure characteristics.
36.8 M14-P05 - Human-Reviewed Recommendation & Approval
Problem. AI provides a recommendation/score/plan that informs a human-controlled decision or approval, and the architecture must prove meaningful oversight rather than merely human presence.
Applicability. Human-final decisions, approval-before-action, regulated review, material exception handling or accountable judgement.
Participating architecture. P4-ABB-01, P6-ABB-01, P6-ABB-02 where action/Authorization applies, P6-ABB-03 and P7-ABB-01/03.
Material interactions. M11-INT-05, 12, 23, 24, 25, 26, 27, 31 and 32; human/external authority boundaries retained from M09.
Solution shape. AI output is presented as Decision input; the human receives sufficient context/evidence and genuine ability to review/reject/approve/intervene; a distinct operative Decision and Authorization are created where required; the oversight occurrence is evidenced.
Must not be concluded. Human participation does not prove review, approval or Authorization. Post-hoc review is not prior approval.
M14-C05 - Boundary counterexample. "A human was copied on the case, so the system is human-in-the-loop." Invalid: passive visibility without timely authority, relevant evidence and intervention ability is not meaningful approval/oversight.
GEAIRA-M14-R17 - Oversight evidence: M14-P05 SHALL distinguish availability of a human from evidence that the required review/approval/intervention actually occurred under the required timing and authority conditions.
GEAIRA-M14-R18 - Recommendation non-commitment: An AI recommendation, score or plan SHALL NOT be treated as the operative enterprise Decision or Authorization solely because a human participant is associated with the process.
36.9 M14-P06 - Bounded Autonomous Digital Action
Problem. AI is permitted to perform digital actions autonomously within a defined envelope without interpreting autonomy as unrestricted authority.
Applicability. AutonomousAction classifications, pre-authorized digital automation, agentic operations or bounded machine-to-machine action.
Participating architecture. P1-ABB-02, P4-ABB-01/02, P6-ABB-02, conditional P6-ABB-03 and P7-ABB-01/03/04.
Material interactions. M11-INT-03, 05, 09, 23, 25, 26, 27, 28, 31, 32 and 33.
Solution shape. P1 establishes the authority envelope; P6 binds point-of-use or pre-authorized conditions and action-control obligations; P4 performs the technical action; P7 captures action/outcome evidence and propagates safety/currentness feedback.
Must not be concluded. Product autonomy classification or Agent autonomy does not itself grant permission to use any Tool or act in any domain.
M14-C06 - Boundary counterexample. "The Product is AutonomousAction, therefore all tool calls are pre-approved." Invalid: autonomy classification remains bounded by authority, policy, target, conditions, currentness and applicable point-of-use controls.
GEAIRA-M14-R19 - Bounded autonomy: M14-P06 SHALL bind autonomous digital action to explicit authority scope, applicable conditions, revocation/currentness and evidence obligations.
GEAIRA-M14-R20 - Autonomy non-authority: Agent/Product autonomy classification SHALL NOT be used as a substitute for Delegated Authority or applicable Authorization semantics.
36.10 M14-P07 - Closed-Loop Physical AI
Problem. AI senses and acts upon a physical Environment under latency/safety constraints while preserving control-loop, authority, safety and evidence boundaries.
Applicability. Robotics, autonomous/assisted physical systems, industrial control, vehicles, devices or other PhysicalAction patterns.
Participating architecture. P4-ABB-03, P4-ABB-01, P6-ABB-02, conditional P6-ABB-03, P1-ABB-02 and P7-ABB-01/03/04.
Material interactions. M11-INT-03, 05, 23, 25, 26, 27, 28, 29, 31, 32 and 33 plus Environment observation/action boundaries retained from M07.
Solution shape. Local control may operate inside a bounded current authority/safety envelope; P4 handles sensing, execution and actuation; P6 provides applicable decision/action control; P7 supplies currentness/safety feedback and evidence. M13-RP-04/RP-05 may place these responsibilities locally without mandatory remote calls.
Must not be concluded. Participation in a Control Loop does not imply autonomous authority, and actuation does not prove authorization or success.
M14-C07 - Boundary counterexample. "The robot controller can actuate the arm, therefore the AI is authorized to perform the movement." Invalid: technical capability, Product Embodiment, authority and point-of-use safety control are separate.
GEAIRA-M14-R21 - Local safety viability: M14-P07 SHALL permit safety/real-time control responsibilities to execute locally where required, while preserving governed envelope, revocation strategy and evidence traceability.
GEAIRA-M14-R22 - Actuation non-authority: Technical ability to actuate an Environment SHALL NOT be treated as proof of authority, Authorization or successful PhysicalAction outcome.
36.11 M14-P08 - Agent Uses Tool Under Bounded Authority
Problem. An Agent participates in AI Execution and uses one or more Tools without inheriting Tool/Product authority or collapsing Agent identity into Product/Primitive identity.
Applicability. Agentic workflows, copilots, autonomous/semi-autonomous Agents or multi-step Tool-using AI.
Participating architecture. P4-ABB-02, P4-ABB-01, P1-ABB-02, P6-ABB-02 and P7-ABB-01/03; Agent Ontology and AIPS/BPS boundaries through M12 as applicable.
Material interactions. M11-INT-03, 05, 23, 25, 26, 28, 29, 31 and 32 plus Agent/Tool participation boundary retained from M07.
Solution shape. Agent identity/taxonomy remains externally authoritative; Tool/Product identity remains independent; P4 mediates contextual participation; P6 applies applicable action/Tool controls; evidence records which Agent/Tool actually participated.
Must not be concluded. Using a Tool does not transfer that Tool's entitlements, Product governance, risk tier or business authority to the Agent.
M14-C08 - Boundary counterexample. "The finance Agent can call the payment Product, so it inherits the payment Product's action authority." Invalid: use relationship is not authority delegation.
GEAIRA-M14-R23 - Participant identity separation: M14-P08 SHALL preserve independent Agent, Tool, Product, primitive and execution identities where applicable.
GEAIRA-M14-R24 - No authority inheritance by Tool use: An Agent's ability to invoke/use a Tool SHALL NOT by itself transfer the Tool's Product governance, entitlement, Delegated Authority or action Authorization to the Agent.
36.12 M14-P09 - Product / Capability Composition Without Identity Collapse
Problem. Several AI Products/capabilities/primitives are composed or orchestrated, but the architecture must distinguish runtime composition from creation of a new Product or Derived AI Product.
Applicability. Chains, ensembles, orchestration, compound experiences, composite Products or upstream AI Product dependencies.
Participating architecture. P3-ABB-03, P4-ABB-01/02 as applicable, P5-ABB-01/02 and P7-ABB-01/03; BPS/AIPS semantics through M12.
Material interactions. M11-INT-07, 16, 17, 18, 20, 21, 22, 29 and 30.
Solution shape. Runtime orchestration/composition preserves identities of participating Products/primitives and execution occurrences. A new Product identity is asserted only when BPS Product qualification and applicable AIPS AI Product/Derived semantics independently establish it.
Must not be concluded. Graph composition or orchestration does not automatically create a Product, collective Agent or Derived AI Product.
M14-C09 - Boundary counterexample. "Product A calls Product B and C, so A+B+C is automatically a Derived AI Product." Invalid: runtime orchestration and Product constitution are different concerns.
GEAIRA-M14-R25 - Composition identity preservation: M14-P09 SHALL preserve participating Product/primitive/Agent identities and SHALL distinguish runtime composition from Product constitution.
GEAIRA-M14-R26 - No graph-shape Product inference: Composition topology, call graph or repeated orchestration SHALL NOT by itself create Product identity, AI Product qualification or Derived AI Product classification.
36.13 M14-P10 - External Managed AI Provider Mediation
Problem. Enterprise AI depends on an external provider while retaining enterprise accountability, governing controls, provider-currentness and evidence sufficiency.
Applicability. External foundation models, AI APIs, managed Agent services, hosted reasoning/search or other provider-operated intelligence.
Participating architecture. P5-ABB-03, P5-ABB-01, P3-ABB-02/03 where provider realizes primitives, P4-ABB-01, P1-ABB-02 and P7-ABB-01/03.
Material interactions. M11-INT-02, 04, 16, 17, 18, 20, 21, 22, 30, 32 and 33 plus M12 external-provider interoperability.
Solution shape. The enterprise mediates provider invocation under explicit contract/version/currentness; maps provider capability to enterprise capability needs; retains necessary evidence; defines substitution/failure behavior; and does not outsource purpose, Product qualification, decision authority or assurance merely by outsourcing execution.
Must not be concluded. Provider certification, API stability or contractual SLA is not sufficient proof of enterprise GEAIRA conformance or safe semantic substitution.
M14-C10 - Boundary counterexample. "Vendor B implements the same OpenAI-compatible API as Vendor A, so we can switch without reassessment." Invalid: capability/effect, governance, data handling, evidence, safety and currentness may differ.
GEAIRA-M14-R27 - Provider accountability preservation: M14-P10 SHALL preserve enterprise responsibility for applicable purpose, authority, decision/action control and assurance even when execution is provider-operated.
GEAIRA-M14-R28 - Provider substitution review: External provider substitution SHALL assess material semantic/capability/effect/control/evidence/currentness differences rather than rely solely on API/schema compatibility.
36.14 M14-P11 - Federated Domain AI Consumption
Problem. One domain consumes AI capability owned/operated by another domain or organization under distributed administrative authority.
Applicability. Data Mesh-like domains, multi-business-unit AI, legal-entity federation, partner AI or cross-domain shared Products.
Participating architecture. P1-ABB-02, P5-ABB-01/02/03, P4-ABB-01, P6-ABB-02 where consequential use applies, P7-ABB-01/03; M12 peer-boundary and M13-RP-03.
Material interactions. M11-INT-04, 05, 07, 21, 22, 23, 25, 30, 31, 32 and 33.
Solution shape. Producer and consumer domains retain their own authority scopes while exchanging governed serving/AI contracts. Semantic mappings, Product/version identity, governing conditions, evidence and failure obligations are explicit at the federation boundary.
Must not be concluded. Organizational trust or shared enterprise identity does not make policy, authority, Product semantics or evidence automatically equivalent across domains.
M14-C11 - Boundary counterexample. "Both domains belong to the same bank, so the consuming Agent can use the producer's AI Product under the producer's permissions." Invalid: consumer-side purpose/authority and point-of-use controls remain independently applicable.
GEAIRA-M14-R29 - Federated jurisdiction: M14-P11 SHALL preserve producer and consumer authority/jurisdiction boundaries and explicitly bind the contract governing cross-domain use.
GEAIRA-M14-R30 - No federation authority transfer: Federation, common ownership or trusted-network membership SHALL NOT itself transfer purpose, Delegated Authority, Product governance or action Authorization across domains.
36.15 M14-P12 - AI-Assisted Analytics Interoperability
Problem. GEARA-governed analytics uses AI capability or AI Products without collapsing analytical Information semantics into AI execution semantics.
Applicability. Predictive/prescriptive analytics with AI, AI-assisted analytics, analytical pipelines invoking AI Products, or AI outputs feeding GEARA Decision integration.
Participating architecture. Relevant GEAIRA P2/P3/P4/P5/P7 responsibilities plus M12 GEARA/GEARM boundary. P6 applies where downstream Decision/action control belongs to the AI-enabled architecture path.
Material interactions. M11-INT-10/11, 16/17, 21/22, 24, 29/30 and 33 plus GEARA peer contracts.
Solution shape. GEARA remains authoritative for analytical Information Need, Analytical Intent, Information formation/provision and analytical Outcome semantics; GEAIRA remains authoritative for enterprise-AI architectural responsibility. Cross-architecture records identify which analytical realization invokes which AI capability/Product and how outputs/evidence are correlated.
Must not be concluded. An analytical capability using AI does not automatically become an AI Product, and AI Execution does not automatically become an Analytics Product or Information Product.
M14-C12 - Boundary counterexample. "The dashboard calls an AI model, therefore the dashboard is an AI Product and GEARA no longer applies." Invalid: Product family and architecture authority do not transfer by dependency.
GEAIRA-M14-R31 - Analytics authority preservation: M14-P12 SHALL preserve GEARA/GEARM authority for analytics/Information semantics while GEAIRA governs the AI architecture responsibilities actually present.
GEAIRA-M14-R32 - No AI annexation of analytics: Use of AI within an analytical realization SHALL NOT cause GEAIRA to absorb or redefine GEARA responsibilities, nor automatically classify the analytical Product as an AI Product.
36.16 M14-P13 - Data-Grounded AI Interoperability
Problem. AI consumes authoritative enterprise data/Data Products while preserving GEDRA/GEDRM authority, data contracts, provenance, access and lifecycle obligations.
Applicability. AI Context sourced from Data Products, training/evaluation/runtime data, retrieval over governed enterprise data, or AI Products depending on Data Products.
Participating architecture. P2-ABB-02/03/01, P3/P4 as applicable, P7-ABB-01/03 and M12 GEDRA/GEDRM boundary.
Material interactions. M11-INT-01, 10, 11, 13, 14, 15, 29 and 33 plus peer Data architecture contracts.
Solution shape. Data meaning/authority/product/quality/lineage remain governed by GEDRM/GEDRA. GEAIRA determines how eligible data/resource references contribute to AI Context or primitive/evaluation needs. Cross-architecture mapping preserves Data Product identity/version and relevant evidence without reclassifying data as an AI-native semantic construct.
Must not be concluded. Retrieval from an Authoritative Data Source/Data Product automatically makes the data a Grounding Source or guarantees downstream AI correctness.
M14-C13 - Boundary counterexample. "Because the source is an ADS/Data Product, anything retrieved from it is valid grounding for every AI purpose." Invalid: purpose, scope, currentness, context applicability and grounding qualification remain distinct.
GEAIRA-M14-R33 - Data-authority preservation: M14-P13 SHALL preserve GEDRM/GEDRA authority for data meaning, sourcing, Data Product and data-governance semantics while projecting only applicable data responsibilities into GEAIRA.
GEAIRA-M14-R34 - Authoritative-source non-universality: Authoritative data status SHALL NOT by itself establish grounding applicability, context eligibility or correctness for every AI purpose/execution.
36.17 M14-P14 - Continuous Assurance, Currentness & Governed Change
Problem. AI architecture must react to new evidence, evaluation findings, source/primitive/Product changes, revocation and semantic change without rewriting historical truth or assuming "latest" equals current.
Applicability. Long-lived AI Products/capabilities, continuously operated Agents, changing providers/primitives/context sources, evolving governance and post-deployment assurance.
Participating architecture. P7-ABB-01/02/03/04 plus affected P1-P6 ABBs.
Material interactions. M11-INT-08/09, 13/14/15, 18/19/20, 29/30/31/32/33.
Solution shape. Evidence is correlated to the actual versions/conditions under which it arose; Evaluation and Assurance remain distinct; Currentness Assessment determines future eligibility; revocation/change propagates to affected responsibilities; historical executions keep their original bindings.
Must not be concluded. A newer version, new evaluation criterion or changed policy retroactively changes what earlier executions used or what earlier assurance findings actually said.
M14-C14 - Boundary counterexample. "Primitive v4 replaced v3, so historical records should show v4 as the model behind last month's executions." Invalid: supersession/currentness changes future governance, not historical occurrence identity.
GEAIRA-M14-R35 - Historical truth preservation: M14-P14 SHALL preserve the versions, evidence, contracts and governing conditions that actually applied to historical occurrences while allowing future currentness/eligibility to change.
GEAIRA-M14-R36 - Feedback non-authorization: Evaluation/Assurance/currentness feedback MAY constrain future operation but SHALL NOT itself be treated as Product approval, risk classification or point-of-use Authorization unless the authoritative model separately establishes that outcome.
36.18 M14-P15 - Shared-Platform Multi-Architecture Co-Realization
Problem. One enterprise platform (for example an HDIP-style platform) realizes GEAIRA, GEDRA and GEARA responsibilities without collapsing the three generic architectures into one platform ontology or one set of planes.
Applicability. Enterprise self-service data/information/intelligence platforms, integrated governance platforms or shared developer/consumer experiences.
Participating architecture. Any applicable GEAIRA P1-P7 ABBs; M12 GEDRA, GEARA, BPS/AIPS/AIPDS and HDIP boundaries; M13-RP-08.
Material interactions. Applicable M11-INT-01..33; adjacent mappings remain explicit.
Solution shape. Shared platform capabilities may co-realize multiple generic architecture responsibilities, but each mapping is traceable to its owning architecture and authority. Enterprise-specific planes, registries, workflows and lifecycle gates are profile constructs, not newly minted GEAIRA core constructs.
Shared enterprise platform capability
| | |
v v v
GEDRA GEARA GEAIRA
responsibilities responsibilities responsibilities
Must not be concluded. Common implementation means common semantic jurisdiction or common conformance.
M14-C15 - Boundary counterexample. "HDIP has one Governance Plane, therefore that plane replaces GEAIRA P1/P6/P7, GEDRA governance and GEARA governance." Invalid: a shared implementation construct may realize several responsibilities but does not erase their distinct generic architecture contracts.
GEAIRA-M14-R37 - Multi-architecture traceability: M14-P15 SHALL keep co-realized GEAIRA, GEDRA, GEARA and other applicable responsibilities independently traceable to their owning specification/profile.
GEAIRA-M14-R38 - Shared-platform non-annexation: A shared enterprise platform construct SHALL NOT be promoted into a generic GEAIRA plane/ABB or used to transfer semantic/architecture authority merely because it realizes several adjacent responsibilities.
36.19 Cross-pattern composition rules
Canonical patterns are deliberately composable. A realistic enterprise architecture may combine, for example:
M14-P04 AI Product Serving
+ M14-P02 Grounded Context
+ M14-P08 Agent Tool Use
+ M14-P06 Bounded Autonomous Action
+ M14-P14 Continuous Assurance
or:
M14-P13 Data-Grounded AI
+ M14-P12 AI-Assisted Analytics
+ M14-P15 Shared Platform Co-Realization
Pattern composition does not create an extra super-plane, orchestration bus or "AI Architecture Manager" ABB.
GEAIRA-M14-R39 - Responsibility deduplication: When patterns share the same canonical responsibility/ABB, the composed architecture SHALL preserve one logical responsibility identity rather than duplicate it solely because multiple patterns reference it.
GEAIRA-M14-R40 - Interaction deduplication: Shared M11 interactions MAY satisfy several composed patterns; the architecture SHALL NOT mint duplicate canonical interaction identities for each pattern instance.
GEAIRA-M14-R41 - Contract specialization: Pattern instances MAY specialize M11 contract obligations for context, risk, topology or profile, but SHALL NOT weaken applicable M01-M13 requirements.
GEAIRA-M14-R42 - Control accumulation: If several applicable patterns introduce independent control obligations, composition SHALL preserve the union of material obligations unless authoritative governance explicitly resolves an overlap or conflict.
GEAIRA-M14-R43 - Conflict visibility: Pattern conflicts involving authority, purpose, currentness, safety, privacy, evidence or failure behavior SHALL be made explicit and resolved through authoritative governance rather than hidden by implementation precedence.
GEAIRA-M14-R44 - Causal traceability: Pattern composition SHALL preserve sufficient causal/correlation information to reconstruct material request->context/primitive->execution->decision/action->evidence chains without requiring one universal identifier.
GEAIRA-M14-R45 - Failure composition: The failure behavior of a composed architecture SHALL consider cross-pattern partial/indeterminate states and SHALL NOT assume atomic success merely because each local pattern instance is individually valid.
GEAIRA-M14-R46 - Adjacent-boundary preservation: Composed patterns crossing GEDRA, GEARA, BPS/AIPS, Agent Ontology, W3C PROV or HDIP boundaries SHALL preserve the M12 authority and conformance rules of each boundary.
GEAIRA-M14-R47 - Realization independence: Pattern composition SHALL remain distinct from M13 topology composition; the same logical pattern composition MAY be realized centrally, distributed, federated, edge, provider-managed or hybrid where its physical requirements permit.
GEAIRA-M14-R48 - No pattern-inferred Product identity: Combining multiple patterns SHALL NOT by itself create Product, AI Product, Derived AI Product, Agent or Intelligence Capability identity.
36.20 Canonical M14 anti-pattern catalogue
M14 consolidates the most important unsafe architecture conflations discovered across M02-M13 into twenty canonical anti-patterns. These do not replace the more detailed slice-specific anti-pattern catalogues; they provide a cross-plane review lens.
- Endpoint-is-governed - treating use of an approved endpoint/platform as proof that purpose, context, primitive and evidence obligations are satisfied.
- RAG-is-grounded - treating retrieval technology as proof of Grounding Source qualification or correctness.
- Capability-label substitution - treating two primitives/providers as substitutable because they advertise the same capability label.
- API-is-Product - inferring Product/AI Product identity from serving, reuse, marketplace presence or monetization.
- Human-present-is-approved - treating human visibility or participation as proof of review, approval or meaningful oversight.
- Autonomous-is-authorized - treating Product/Agent autonomy classification as unrestricted permission.
- Actuator-is-authority - treating physical capability to act as proof of permitted/safe action.
- Tool-use authority inheritance - allowing Agent Tool use to transfer Product/Tool entitlements or authority.
- Orchestration-is-derived-product - inferring new Product/Derived AI Product identity from call graph or composition topology.
- Provider-is-accountability-owner - outsourcing enterprise purpose, decision/action control or assurance responsibility with execution.
- Federation-is-shared-authority - assuming common ownership/trust removes cross-domain authority and contract boundaries.
- AI-annexes-analytics - treating AI dependency as reason for GEAIRA to absorb GEARA/GEARM responsibilities.
- Authoritative-data-is-universal-grounding - treating an authoritative data source as suitable grounding for every AI context/purpose.
- Latest-rewrites-history - replacing historical contract/context/primitive/policy identity with current versions.
- Shared-platform-is-one-architecture - collapsing GEDRA/GEARA/GEAIRA because one enterprise platform implements all three.
- Pattern-is-topology - treating an M14 problem/solution pattern as requiring one deployment topology.
- Pattern-per-service - deploying a new service for every canonical pattern or ABB reference.
- Control-by-observability - treating telemetry, monitoring or an assurance dashboard as the control/Authorization itself.
- Happy-path composition - composing patterns without defining partial, indeterminate, revocation and reconciliation semantics.
- Conformance-by-brand - claiming GEAIRA alignment because a vendor/product/platform uses architecture-like terminology rather than demonstrating responsibility/contract/control coverage.
GEAIRA-M14-R49 - Canonical anti-pattern rejection: Architectures claiming GEAIRA 0.1 alignment SHALL NOT intentionally rely on M14 anti-patterns 1-20 as substitutes for the corresponding canonical responsibilities, authority boundaries or contracts.
GEAIRA-M14-R50 - Endpoint-is-governed rejection: Approved platform/endpoint status SHALL NOT substitute for explicit governing, eligibility and evidence obligations.
GEAIRA-M14-R51 - Retrieval-is-grounding rejection: Retrieval success or repository technology SHALL NOT substitute for grounding/source qualification and context applicability.
GEAIRA-M14-R52 - Human-presence rejection: Human presence, notification or downstream audit SHALL NOT be represented as prior review/approval unless that oversight actually occurred under the required conditions.
GEAIRA-M14-R53 - Autonomy/actuation rejection: Autonomy classification or technical actuation capability SHALL NOT substitute for applicable authority, Authorization and safety controls.
GEAIRA-M14-R54 - Composition-inference rejection: Runtime orchestration/composition SHALL NOT create Product, Agent or authority identity by graph shape alone.
GEAIRA-M14-R55 - Provider/federation rejection: External provision or federation SHALL NOT transfer enterprise accountability or authority unless an authoritative delegation/contract explicitly establishes the relevant scope.
GEAIRA-M14-R56 - Adjacent-annexation rejection: GEAIRA SHALL NOT absorb GEDRA/GEARA/BPS/AIPS/Agent-Ontology/W3C-PROV authority merely because one pattern crosses those boundaries.
GEAIRA-M14-R57 - History-rewrite rejection: Current/latest versions SHALL NOT be projected backwards over historical executions, contracts, decisions or assurance evidence.
GEAIRA-M14-R58 - Topology/service inflation rejection: Canonical patterns SHALL NOT be interpreted as mandatory deployment tiers, services, registries, gateways or control planes.
GEAIRA-M14-R59 - Observability-control rejection: Telemetry, dashboards, logs or Evidence Stores SHALL NOT be treated as equivalent to Evaluation, Assurance, Authorization or control effectiveness.
GEAIRA-M14-R60 - Composition failure discipline: Composed patterns SHALL explicitly address material partial, failed, indeterminate, revoked and reconciliation-required states where those states can occur.
GEAIRA-M14-R61 - Conformance evidence: Claims that an implementation realizes an M14 pattern SHOULD identify the participating responsibilities/ABBs, contracts, controls, evidence and external authority mappings rather than rely on product naming.
GEAIRA-M14-R62 - Profile strengthening: Enterprise/domain profiles MAY prohibit additional pattern variants or introduce stricter controls but SHALL NOT weaken the M14 hard distinctions or M01-M13 core rules.
GEAIRA-M14-R63 - Anti-pattern non-exhaustiveness: The M14 anti-pattern catalogue is canonical but not exhaustive; absence from the catalogue SHALL NOT make an architecture valid if it violates another frozen GEAIRA or external-authority rule.
36.21 Pattern selection and architecture-review method
M14 does not require every architecture to instantiate all fifteen patterns. Architecture review SHOULD proceed in the following order:
1. Identify the actual enterprise-AI problem and semantic participants.
2. Determine applicable P1-P7 responsibilities and external authorities.
3. Select one or more M14 patterns whose applicability conditions hold.
4. Map each pattern to frozen ABBs/interactions rather than inventing parallel components.
5. Select an M13 realization topology appropriate to latency, jurisdiction, resilience and operational constraints.
6. Define contract/currentness/failure/evidence obligations.
7. Test the architecture against the M14 boundary counterexample and anti-pattern catalogue.
8. Preserve historical/version/conformance traceability.
A pattern MAY be used informatively even when some optional participants are absent. The invariant is preservation of the responsibilities that actually apply.
GEAIRA-M14-R64 - Problem-first selection: Pattern selection SHALL begin from the recurring architecture problem and applicability conditions rather than from a vendor feature, product catalogue or desired topology.
GEAIRA-M14-R65 - Minimal sufficient pattern set: Architecture SHOULD select the smallest sufficient set of canonical patterns that covers the material problem/controls, avoiding pattern inflation for documentation symmetry.
GEAIRA-M14-R66 - Review against counterexample: A pattern instance SHOULD be explicitly tested against its M14-Cxx boundary counterexample and relevant cross-pattern anti-patterns during architecture review.
36.22 Frozen M14 decisions
M14 freezes the following decisions for GEAIRA 0.1:
Architecture Pattern != deployment topology
Architecture Pattern != Product / Product Type
Architecture Pattern != service / platform
pattern participation != authority transfer
pattern composition != new semantic identity
retrieval != grounding
human participation != approval
autonomy != authority
actuation != Authorization
orchestration != Product constitution
provider operation != accountability transfer
federation != common authority
AI integration != annexation of Data/Analytics architecture
latest/current != historical truth
shared platform != shared architecture jurisdiction
The following are canonical M14 human-readable artifacts:
15 canonical architecture patterns (M14-P01..M14-P15)
15 pattern boundary counterexamples (M14-C01..M14-C15)
20 consolidated cross-plane anti-patterns
pattern coverage matrix
pattern record / review method
66 normative M14 rules
M14 does not freeze:
- M15 canonical reference views or end-to-end flows;
- a universal enterprise-AI workflow;
- a mandatory deployment topology;
- a vendor/platform blueprint;
- a complete enterprise Profile;
- machine-readable pattern registries;
- M16 conformance/invariant/profile rules;
- M17 final human-readable baseline freeze.
37. M14 Completion State
GEAIRA-M14 - COMPLETE.
M14 establishes:
- one canonical architecture-pattern record;
- fifteen canonical positive architecture patterns, M14-P01 through M14-P15;
- fifteen paired boundary counterexamples, M14-C01 through M14-C15;
- explicit coverage of governed invocation, grounding, multi-primitive realization, AI Product serving, human oversight, autonomous action, Physical AI, Agent/Tool use, Product/capability composition, external providers, federation, GEARA interoperability, GEDRA interoperability, continuous assurance/change and shared-platform co-realization;
- a twenty-item consolidated anti-pattern catalogue;
- cross-pattern composition rules;
- a problem-first pattern-selection and architecture-review method;
- preservation of all seven planes, twenty-one canonical ABBs, thirty-three M11 interactions, nine M12 adjacent boundaries and eight M13 realization patterns;
- 66 normative M14 rules.
M14 introduces no new GEAIRM semantic concepts/relationships, no new GEAIRA plane/ABB, no new M11 canonical interaction and no new M12 adjacent boundary.
38. GEAIRA-M15 - Canonical Reference Views & End-to-End Flows
38.1 Purpose
GEAIRA-M15 turns the frozen GEAIRA 0.1 architecture into a stable set of stakeholder-oriented reference views and representative end-to-end flows.
M15 answers two different questions:
- Reference view: What part of the architecture must a stakeholder see to answer a particular architecture question?
- End-to-end flow: How may responsibilities, contracts and evidence participate across time for a representative enterprise-AI scenario?
These constructs are projections of the existing architecture. They do not create a second architecture model.
Reference View
= purposeful projection of frozen architecture
End-to-End Flow
= representative traversal of frozen responsibilities/contracts
Reference View / Flow
!= new Plane
!= new ABB
!= new M11 interaction
!= new semantic concept
!= mandatory implementation topology
M15 preserves the M01 rule that a Reference View does not create duplicate responsibility merely for presentation, and the M13/M14 rule that logical architecture remains independent of physical topology.
GEAIRA-M15-R01 - Projection rule: An M15 Reference View SHALL project existing GEAIRA responsibilities, ABBs, interactions, overlays, adjacent boundaries and patterns; it SHALL NOT create parallel architecture ownership.
GEAIRA-M15-R02 - Flow non-universality: An M15 end-to-end flow SHALL be interpreted as a representative architecture traversal under stated conditions and SHALL NOT be treated as a mandatory universal enterprise-AI lifecycle or runtime sequence.
GEAIRA-M15-R03 - Identity preservation: Participation in a Reference View or flow SHALL NOT create, merge or redefine Product, AI Product, Agent, AI Primitive, Intelligence Capability, AI Execution, Decision, Authorization, Action, Evaluation or Assurance identity.
GEAIRA-M15-R04 - Authority preservation: A view or flow SHALL preserve the authoritative source of semantic, Product, AI Product, Agent, Data, Analytics, provenance and enterprise-profile semantics established by M12.
GEAIRA-M15-R05 - Interaction reuse: M15 SHALL reuse M11-INT-01 through M11-INT-33 for inter-plane hand-offs and SHALL NOT introduce synthetic interaction identities for diagrammatic convenience.
GEAIRA-M15-R06 - Topology neutrality: A Reference View or flow MAY be realized by any applicable M13 realization pattern and SHALL NOT imply centralization, distribution, federation, edge placement or provider usage unless that characteristic is part of the scenario being shown.
GEAIRA-M15-R07 - Pattern traceability: Representative flows SHOULD identify the M14 canonical pattern or pattern composition they illustrate so that views, flows and problem/solution patterns remain traceable.
GEAIRA-M15-R08 - Historical truth: M15 views and flows SHALL preserve the version, contract, context, primitive, authority and evidence state that actually applied to a historical occurrence; later currentness findings SHALL NOT rewrite earlier facts.
38.2 Reference-view metamodel
Each canonical Reference View contains the following conceptual fields:
viewId
name
stakeholderQuestion
primaryStakeholders
includedPlanes
participatingABBs
principalM11Interactions
crossCuttingOverlays
adjacentAuthorityBoundaries
applicableM14Patterns
applicableM13RealizationPatterns
requiredEvidenceOrTraceability
explicitExclusions
The view is intentionally selective. Not every applicable responsibility has to be drawn if omission does not distort the stakeholder question.
GEAIRA-M15-R09 - View selectivity: A Reference View MAY omit architecture elements that are irrelevant to its stakeholder question, provided the omission does not imply that an applicable responsibility or control ceases to exist.
GEAIRA-M15-R10 - View consistency: If the same ABB, interaction or semantic identity appears in multiple Reference Views, it SHALL retain the same canonical identity and meaning.
GEAIRA-M15-R11 - View independence from diagram notation: The canonical meaning of an M15 Reference View SHALL NOT depend on a specific diagramming notation, vendor tool, icon set, layer style or visual layout.
GEAIRA-M15-R12 - Cross-cutting visibility: A view SHOULD show the cross-cutting overlays materially relevant to its concern rather than treating overlays as an optional legend detached from architecture obligations.
GEAIRA-M15-R13 - External-boundary visibility: Where a view crosses a GEDRA, GEARA, BPS/AIPS, AIPDS, Agent Ontology, W3C PROV or HDIP boundary, that boundary SHOULD be visible enough to prevent authority conflation.
GEAIRA-M15-R14 - View non-deployment: Boxes drawn in a Reference View SHALL NOT be interpreted as mandatory deployable services unless a realization Profile separately establishes that choice.
GEAIRA-M15-R15 - View conformance: A solution MAY conform to GEAIRA while using different solution diagrams, provided it can map material responsibilities and contracts to the canonical Reference Views where those views are applicable.
GEAIRA-M15-R16 - View non-proliferation: Enterprise/domain Profiles MAY add concern-specific views, but SHOULD reuse canonical M15 views where possible and SHALL NOT duplicate architecture responsibility solely to obtain a different presentation.
38.3 Canonical Reference View catalogue
M15 freezes ten canonical views, materializing and refining the ten candidate views first identified in M01.
| View | Name | Primary stakeholder question |
|---|---|---|
| M15-V01 | Architecture Authority & Responsibility View | Who owns meaning, authority and architectural responsibility, and where do those boundaries meet? |
| M15-V02 | Enterprise AI End-to-End View | How can governed intent progress into AI Execution, optional decision/action, evidence and feedback without implying one universal pipeline? |
| M15-V03 | Context & Grounding View | How are resources retrieved, qualified, assembled and provided as bounded context while preserving source/currentness? |
| M15-V04 | Agentic & Tool-Use View | How may Agents participate in executions and use Tools without inheriting unbounded authority? |
| M15-V05 | Physical AI & Control View | How do sensing, execution, Authorization, Actuation, safety and control-loop feedback interact? |
| M15-V06 | Product, Serving & Integration View | How are AI capabilities or qualified AI Products exposed, contracted, integrated and correlated to execution? |
| M15-V07 | Decision, Authorization & Human Oversight View | How are recommendations, Decisions, Authorization, Action and human oversight kept distinct and governed? |
| M15-V08 | Evaluation, Assurance & Observability View | How do telemetry/evidence, Evaluation, Assurance and continuation constraints relate? |
| M15-V09 | Lifecycle, Provenance & Evolution View | How are version, currentness, supersession, provenance and change propagated without rewriting history? |
| M15-V10 | Federation, Composition & Shared-Platform View | How can domains, Products, peer architectures and shared platforms compose without identity or authority collapse? |
GEAIRA-M15-R17 - Canonical view freeze: M15-V01 through M15-V10 are the canonical GEAIRA 0.1 human-readable Reference Views.
GEAIRA-M15-R18 - Candidate reconciliation: The ten M01 candidate views are considered reconciled and frozen through M15-V01 through M15-V10; refinements in naming SHALL NOT be interpreted as new architecture domains.
38.4 M15-V01 - Architecture Authority & Responsibility View
Question. Who owns semantic meaning, Product/AI Product qualification, Agent identity, provenance, data/analytics architecture, and enterprise-AI architectural responsibility?
Primary stakeholders. Enterprise Architects, Domain Architects, CDO/CIO architecture functions, governance authorities, Product governance, risk/compliance and specification custodians.
External semantic / architecture authorities
GEAIRM GEDRM/GEDRA GEARM/GEARA BPS/PIN AIPS/AIPDS
| | | | |
+----------+-------------+-----------+-----------+
|
v
GEAIRA responsibility model
P1 P2 P3 P4 P5 P6 P7 + X1..X8
|
+-----------------+------------------+
| |
v v
Agent Ontology W3C PROV
Agent identity provenance
|
v
Enterprise Profile
e.g. HDIP
This view emphasizes jurisdiction rather than runtime sequence.
Principal M11 interactions: M11-INT-01..09 and M11-INT-31..33 where authority/assurance feedback is material.
Principal overlays: X2 Identity, Authority & Policy Enforcement; X3 Evidence, Provenance & Traceability; X6 Lifecycle, Version, Currentness & Change; X8 Governance, Human Accountability & Compliance.
GEAIRA-M15-R19 - Authority-view non-annexation: M15-V01 SHALL show adjacent semantic/architecture authorities as peers or foundations and SHALL NOT depict them as GEAIRA-owned subcomponents.
38.5 M15-V02 - Enterprise AI End-to-End View
Question. What is the smallest complete enterprise-AI architecture picture that shows intent, context, capability, execution, serving, optional decision/action control and evidence/evolution?
P1
Intent / Purpose / Authority
/ | \
v v v
P2 P3 P5
Context Primitive Serving
\ | /
\ v /
P4
AI Execution
/ \
optional v v evidence
P6 ----> P7
Decision/Action Evaluation /
/ Oversight Assurance /
Evolution
^ |
+------------+
feedback
The arrows show representative responsibility hand-offs, not one required order. For example, P5 may invoke P4, P4 may request context from P2, P7 may pre-empt future execution through currentness feedback, and P6 may be entirely absent from a non-consequential advisory use case.
This view provides the primary whole-architecture orientation for M15 flows.
GEAIRA-M15-R20 - End-to-end optionality: M15-V02 SHALL keep P5 and P6 participation conditional where the scenario has no serving/Product or consequential decision/action concern.
38.6 M15-V03 - Context & Grounding View
Question. How does GEAIRA distinguish knowledge/resource access, retrieval, Grounding Source qualification, bounded AI Context, memory contribution and currentness?
Resource / Data / Information authorities
|
v
P2-ABB-02 Retrieval Mediation
|
candidate resources
v
P2-ABB-03 Grounding &
Source Qualification
|
qualified contribution
v
P2-ABB-01 Context Assembly
|
M11-INT-11 AI Context
v
P4 Execution
|
v
P7 evidence
|
currentness feedback
+------> P2
This view MAY expose GEDRA/GEARA boundaries for Data or Information resources. It MUST preserve:
retrieved resource != Grounding Source
Grounding Source != truth guarantee
AI Context != Environment
Memory contribution != learning / retraining
GEAIRA-M15-R21 - Grounding-view distinction: M15-V03 SHALL visually or textually preserve the distinction between retrieval, grounding qualification and bounded context assembly.
38.7 M15-V04 - Agentic & Tool-Use View
Question. How may an Agent participate in AI Execution and use Tools while authority, Authorization and execution occurrence remain explicit?
P1 authority envelope
|
v
Agent identity --> P4-ABB-02 Agent & Tool Participation
|
+---- participates in ----+
| |
v v
P4-ABB-01 Tool / external
AI Execution capability
|
consequential request
v
P6
Authorization / control
|
outcome / commit
v
P4 technical enactment
|
v
P7 evidence
The KivaNura Agent Ontology remains authoritative for Agent identity/taxonomy. GEAIRA owns the architectural responsibility for participation, Tool-use mediation and occurrence-level runtime boundaries.
GEAIRA-M15-R22 - Agent-view authority discipline: M15-V04 SHALL NOT visually imply that Agent identity, autonomy classification, Tool availability or prior participation creates Authorization for a consequential invocation/action.
38.8 M15-V05 - Physical AI & Control View
Question. How do sensing, AI Execution, decision/action control, Actuation, Environment feedback and safety interact in Physical AI?
Environment
|
observation
v
P4-ABB-03 / P4
sensing + execution
|
proposed effect / action
v
P6
Authorization / safety /
human oversight
|
authorized command
v
P4-ABB-03
Actuation
|
v
Environment
|
+----> P7 evidence / safety feedback
Local pre-authorization and embedded safety are permitted where latency/safety require them. The view does not require a remote control service in the hard-real-time path.
GEAIRA-M15-R23 - Physical-view locality: M15-V05 SHALL allow safety/Authorization controls to be embedded or local where required and SHALL NOT imply that centralized network access is necessary for every control-loop cycle.
38.9 M15-V06 - Product, Serving & Integration View
Question. How are capabilities and, where valid, Products/AI Products exposed without confusing serving with Product qualification or execution occurrence?
BPS/AIPS Product authority (conditional)
|
v
P5-ABB-02
Product Realization & Contract
|
Consumer --> P5-ABB-01 Serving Boundary
|
M11-INT-21 request
v
P4
AI Execution
|
M11-INT-22 result
v
P5 delivery
|
v
Consumer
External provider / peer integration
^
|
P5-ABB-03
For a non-Product internal AI capability, P5-ABB-02 may be absent.
GEAIRA-M15-R24 - Serving-view Product discipline: M15-V06 SHALL distinguish serving exposure from Product and AI Product qualification and SHALL keep P5-ABB-02 conditional on authoritative Product semantics.
38.10 M15-V07 - Decision, Authorization & Human Oversight View
Question. How does AI-derived information become, or fail to become, an operative Decision or authorized Action?
P4/P5 Intelligence Effect / recommendation
|
v
P6-ABB-01
Decision Formation & Commitment
|
if action/control
v
P6-ABB-02
Authorization & Action Control
|
+---------+---------+
| |
v v
approve/commit reject/defer/
| escalate/condition
v
P4/P5 enactment
P6-ABB-03 Human Oversight & Escalation may
participate before Decision, before Action, during
supervision, on exception, or after-the-fact review.
GEAIRA-M15-R25 - Decision-view temporal truth: M15-V07 SHALL distinguish prior review/approval, point-of-use Authorization, supervisory intervention and post-hoc review; a later human review SHALL NOT be depicted as prior approval.
38.11 M15-V08 - Evaluation, Assurance & Observability View
Question. How do runtime/serving/decision observations become evidence, Evaluation Results, Assurance Findings and continuation/change constraints?
P2 / P3 / P4 / P5 / P6
evidence inputs
|
v
P7-ABB-01
Observability & Evidence
|
v
P7-ABB-02
Evaluation
|
v
P7-ABB-03
Assurance & Currentness
|
v
P7-ABB-04
Evolution / Feedback
|
+----> affected P1-P6
The view preserves:
telemetry != Evidence automatically
Evaluation != Assurance
Assurance != Authorization
PASS != universal fitness
GEAIRA-M15-R26 - Assurance-view separation: M15-V08 SHALL preserve Observation/Evidence, Evaluation, Assurance and continuation/change decisions as distinct responsibilities even when realized by one platform.
38.12 M15-V09 - Lifecycle, Provenance & Evolution View
Question. How do identities and versions change while historical truth remains reconstructible?
Object/version v3 --used by--> Execution E1 @ T1
|
| later superseded by
v
Object/version v4 --current for--> future use @ T2
P7 currentness/evolution findings
|
+--> constrain future P1-P6 behavior
Historical evidence remains:
E1 used v3 under the contract/context/authority valid at T1.
This view spans X3 and X6 heavily and may reference W3C PROV for generic provenance representation.
GEAIRA-M15-R27 - Lifecycle-view history rule: M15-V09 SHALL distinguish supersession/currentness from deletion or retroactive mutation and SHALL preserve occurrence-time provenance.
38.13 M15-V10 - Federation, Composition & Shared-Platform View
Question. How can multiple domains, Products, Agents, peer architectures and shared enterprise platforms compose while keeping semantic/architecture jurisdiction and conformance independent?
Domain A GEAIRA profile ----M11/M12 contracts---- Domain B GEAIRA profile
| |
+------------------+----------------------+
|
v
Shared enterprise platform
e.g. HDIP
/ | \
v v v
GEDRA GEARA GEAIRA
A platform realization may host responsibilities from several reference architectures. This does not merge the architectures or their semantic foundations.
The view also covers Product/capability composition:
orchestrated capabilities
!= automatically new Product
!= automatically new Agent
!= automatically new authority domain
GEAIRA-M15-R28 - Federation/composition discipline: M15-V10 SHALL preserve independent semantic authority, architecture conformance and identity across federated/composed participants even when they are co-hosted by one shared platform.
38.14 Canonical view coverage matrix
The views are intentionally overlapping projections.
| View | Principal planes / ABBs | Principal M11 coverage | Typical M14 pattern coverage |
|---|---|---|---|
| V01 | P1 plus authority-bearing responsibilities across P2-P7 | 01-09, 31-33 | P01, P06, P08, P10, P11, P15 |
| V02 | P1-P7 | 01-33 as applicable | P01-P15 as composition context |
| V03 | P2-ABB-01/02/03, P4-ABB-01, P7-ABB-01/03 | 01, 10-15, 29, 33 | P02, P13, P14 |
| V04 | P1-ABB-02, P4-ABB-01/02, P6-ABB-02/03, P7 | 03,05,11,17,21,23,25-29,31-33 | P08, P06, P14 |
| V05 | P1-ABB-02, P4-ABB-01/03, P6-ABB-02/03, P7 | 03,05,11,17,23,25-29,31-33 | P07, P14 |
| V06 | P5-ABB-01/02/03, P4-ABB-01, P1/P3/P7 as applicable | 04,07,09,16,20-23,25,30,32,33 | P04, P09, P10, P11 |
| V07 | P6-ABB-01/02/03 plus P1/P2/P4/P5/P7 | 05,12,23-28,31-33 | P05, P06, P07, P08 |
| V08 | P7-ABB-01/02/03/04 plus evidence-producing P1-P6 | 06,08,13,14,18-20,29-33 | P14 plus all evidence-bearing patterns |
| V09 | P7 plus X3/X6 across P1-P6 | 08,09,13-15,18-20,29-33 | P14 |
| V10 | Cross-plane / cross-domain / M12 boundaries | Any applicable M11 contract | P09-P13, P15 |
ABB coverage across V01-V10 includes all canonical ABBs:
P1-ABB-01 P1-ABB-02
P2-ABB-01 P2-ABB-02 P2-ABB-03
P3-ABB-01 P3-ABB-02 P3-ABB-03
P4-ABB-01 P4-ABB-02 P4-ABB-03
P5-ABB-01 P5-ABB-02 P5-ABB-03
P6-ABB-01 P6-ABB-02 P6-ABB-03
P7-ABB-01 P7-ABB-02 P7-ABB-03 P7-ABB-04
Canonical interaction coverage is likewise closed against the frozen M11 identifiers:
M11-INT-01 M11-INT-02 M11-INT-03 M11-INT-04 M11-INT-05 M11-INT-06
M11-INT-07 M11-INT-08 M11-INT-09 M11-INT-10 M11-INT-11 M11-INT-12
M11-INT-13 M11-INT-14 M11-INT-15 M11-INT-16 M11-INT-17 M11-INT-18
M11-INT-19 M11-INT-20 M11-INT-21 M11-INT-22 M11-INT-23 M11-INT-24
M11-INT-25 M11-INT-26 M11-INT-27 M11-INT-28 M11-INT-29 M11-INT-30
M11-INT-31 M11-INT-32 M11-INT-33
GEAIRA-M15-R29 - ABB coverage: The canonical view set SHALL collectively cover all twenty-one canonical GEAIRA ABBs without creating view-specific duplicate ABB identities.
GEAIRA-M15-R30 - Interaction coverage: The canonical view set SHALL collectively cover M11-INT-01 through M11-INT-33, while a particular view MAY show only the interactions relevant to its stakeholder question.
GEAIRA-M15-R31 - Overlay coverage: Cross-cutting overlays X1-X8 SHALL remain applicable in every view according to material relevance; absence from a simplified diagram SHALL NOT mean an overlay is waived.
GEAIRA-M15-R32 - Adjacent-boundary coverage: The canonical view set SHALL be capable of representing all nine M12 adjacent boundaries without redefining their authority.
38.15 End-to-end flow metamodel
An M15 end-to-end flow is defined by:
flowId
name
scenarioIntent
entryConditions
participatingViews
participatingPatterns
participatingPlanesAndABBs
interactionSequenceOrPartialOrder
optionalBranches
externalBoundaries
identityAndCorrelationPoints
controlAndAuthorizationPoints
evidenceAndAssurancePoints
failureOrIndeterminateStates
historicalTraceability
realizationNeutrality
The field interactionSequenceOrPartialOrder is deliberate. Many flows are partially ordered, event-driven, feedback-based or concurrent rather than linear.
GEAIRA-M15-R33 - Partial-order semantics: A flow MAY define ordering constraints between material events without asserting a single globally linear sequence.
GEAIRA-M15-R34 - Optional-path semantics: Conditional responsibilities such as Product realization, Agent participation, human approval, Action and Physical AI SHALL appear as conditional branches rather than mandatory stages where they do not always apply.
GEAIRA-M15-R35 - Correlation without collapse: A flow SHALL preserve distinct request, serving, execution, Decision, Authorization, Action, Evaluation and Assurance identities even when they are linked in one causal chain.
GEAIRA-M15-R36 - Control-point visibility: Consequential flows SHALL show where applicable Authorization, human oversight, safety, revocation or continuation controls occur rather than hiding those obligations inside generic arrows.
GEAIRA-M15-R37 - Failure visibility: Representative flows SHALL show material failed, rejected, deferred, conditional, revoked or INDETERMINATE outcomes where omission would create a false happy-path guarantee.
GEAIRA-M15-R38 - Evidence visibility: Representative flows SHALL identify material evidence-producing points sufficient to reconstruct the causal path required by P7 and X3/X4.
38.16 M15-FLOW-01 - Governed AI Execution
Pattern basis: M14-P01 Governed AI Invocation & Evidence Chain; optionally M14-P03 Multi-Primitive Capability Realization.
Scenario. An enterprise capability invokes AI under a valid governing envelope without requiring Product serving, Agent participation or consequential Action.
1. P1 establishes applicable purpose / authority constraints.
2. P3 resolves an eligible primitive realization for the required capability.
3. P4 binds an AI Execution occurrence to the governing context and primitive realization.
4. P2 provides bounded AI Context if the execution requires context.
5. P4 invokes the primitive(s) and records actual participation/effect.
6. P7 receives runtime / context / primitive evidence.
7. P7 Evaluation/Assurance may constrain future execution.
Representative M11 interactions: 02,03,10,11,16-19,29,32,33.
No P5 or P6 step is mandatory.
GEAIRA-M15-R39 - Base execution flow: M15-FLOW-01 SHALL remain valid for non-Product, non-Agentic and non-consequential AI execution and SHALL NOT force optional P5/P6 responsibilities into the path.
38.17 M15-FLOW-02 - Grounded AI Product Consumption
Pattern basis: M14-P02 Grounded Context Execution + M14-P04 AI Product Serving & Execution Correlation.
Consumer
|
v
P5 validates serving contract / Product Version association
|
+--> P1 governing serving constraints
|
v
P4 execution request
|
+--> P2 context requirement
|
v
retrieve -> qualify grounding -> assemble context
|
v
P4 AI Execution <---- P3 primitive realization
|
v
P5 result delivery
|
v
Consumer
P2/P3/P4/P5 evidence ----------------------> P7
The flow preserves the distinction between Product Version, serving-contract version and AI Execution identity.
GEAIRA-M15-R40 - Grounded Product flow: M15-FLOW-02 SHALL preserve retrieval/grounding/context distinctions and SHALL NOT infer Product or AI Product qualification from the existence of a serving interface.
38.18 M15-FLOW-03 - Human-Reviewed Recommendation and Decision
Pattern basis: M14-P05 Human-Reviewed Recommendation & Approval.
P4/P5 produces recommendation / score / plan
|
v
P6-ABB-01
Decision Formation
|
human review where required
v
P6-ABB-03
|
decision / approval
v
P6-ABB-02
Authorization if Action applies
|
v
P4/P5 technical enactment
|
v
outcome + oversight evidence -> P7
Possible outcomes include reject, defer, conditional approval and escalation.
GEAIRA-M15-R41 - Human-review flow: M15-FLOW-03 SHALL record actual human review/approval as an occurrence with role, timing and evidence and SHALL NOT represent availability or notification as completed oversight.
38.19 M15-FLOW-04 - Bounded Autonomous Digital Action
Pattern basis: M14-P06 Bounded Autonomous Digital Action.
P1 authority / purpose envelope
|
v
P4 execution / Agent participation (optional)
|
proposed action
v
P6 evaluates applicable pre-authorization /
point-of-use Authorization boundary
|
authorized?
/ \
yes no/indeterminate
| |
v v
P4/P5 enact reject/escalate/stop
|
v
outcome evidence -> P6/P7
Bounded pre-authorization is permitted. Autonomy classification alone is not permission.
GEAIRA-M15-R42 - Autonomous-action flow: M15-FLOW-04 SHALL make the applicable authority/Authorization basis explicit and SHALL NOT use autonomy, Agent identity or prior success as a substitute for permission.
38.20 M15-FLOW-05 - Agentic Tool Use
Pattern basis: M14-P08 Agent Uses Tool Under Bounded Authority.
Agent identity (Agent Ontology)
|
v
P4-ABB-02 Agent & Tool Participation
|
tool invocation proposed
|
consequential?
/ \
no yes
| |
| v
| P6 control
| |
+-----------+----> approved invocation
|
v
Tool
|
result / side effect
v
P4 correlation + evidence
|
v
P7
If a side effect times out and occurrence is unknown, the state remains INDETERMINATE until reconciled.
GEAIRA-M15-R43 - Tool-use flow: M15-FLOW-05 SHALL distinguish Tool availability, Tool invocation, Authorization and side-effect outcome and SHALL prohibit blind retry where duplicate effects may be harmful.
38.21 M15-FLOW-06 - Physical AI Closed Loop
Pattern basis: M14-P07 Closed-Loop Physical AI.
Environment observation
|
v
P4 sensing / AI Execution <--- P2 context / P3 primitive
|
proposed control effect
v
P6 local/embedded or remote control boundary
|
authorized command
v
P4 Actuation
|
v
Environment changes
|
+--> next observation
|
+--> P7 safety / outcome evidence
A governed pre-authorized envelope and local fail-safe behavior may support hard-real-time operation.
GEAIRA-M15-R44 - Physical-loop flow: M15-FLOW-06 SHALL permit local or embedded safety/control realization and SHALL preserve an explicit bound on pre-authorized behavior and revocation/currentness handling.
38.22 M15-FLOW-07 - External Managed AI Provider
Pattern basis: M14-P10 External Managed AI Provider Mediation.
Consumer / Product path
|
v
P5-ABB-01 serving
|
v
P5-ABB-03 provider mediation
|
+--> P3 eligibility / provider-backed primitive rules
|
v
External AI provider
|
v
provider result / execution evidence
|
v
P5/P4 correlation -> P7 assurance
Externality does not bypass P1, P6 or P7 obligations where they apply, and provider evidence may be insufficient for some assurance claims.
GEAIRA-M15-R45 - Provider flow: M15-FLOW-07 SHALL preserve enterprise accountability and SHALL NOT infer semantic, risk or assurance equivalence merely because two providers expose compatible APIs.
38.23 M15-FLOW-08 - Federated / Shared-Platform Composition
Pattern basis: M14-P09 Product/Capability Composition Without Identity Collapse + M14-P11 Federated Domain AI Consumption + M14-P15 Shared-Platform Multi-Architecture Co-Realization.
Domain A consumer / Product
|
v
Domain A P5 ---- M11/M12 contract ---- Domain B P5/P4
| |
| v
| AI capability
| |
+---------- correlated outcome ----+
Shared platform MAY realize responsibilities for both domains and
for GEDRA / GEARA / GEAIRA without merging authority or identity.
GEAIRA-M15-R46 - Federated flow: M15-FLOW-08 SHALL keep domain authority, Product identity, contract version and conformance claims independently traceable across federation/shared-platform boundaries.
38.24 M15-FLOW-09 - Data-Grounded and AI-Assisted Analytics Interoperability
Pattern basis: M14-P12 AI-Assisted Analytics Interoperability + M14-P13 Data-Grounded AI Interoperability.
GEDRA-governed Data / Data Product
|
v
GEAIRA P2 context path
|
v
P4 AI Execution
|
+----------------------+
| |
v v
GEARA analytics direct AI effect
runtime/information |
| |
+----------+-----------+
v
Decision / consumer
Data authority remains with GEDRM/GEDRA; analytics semantics remain with GEARM/GEARA; enterprise-AI execution responsibility remains with GEAIRA.
GEAIRA-M15-R47 - Peer-architecture flow: M15-FLOW-09 SHALL preserve GEDRA, GEARA and GEAIRA authority boundaries even where one runtime/platform co-realizes responsibilities from all three.
38.25 M15-FLOW-10 - Continuous Assurance, Currentness & Evolution
Pattern basis: M14-P14 Continuous Assurance, Currentness & Governed Change.
P1-P6 operational / governance evidence
|
v
P7-ABB-01
|
v
P7-ABB-02 Evaluation
|
v
P7-ABB-03 Assurance / Currentness
|
v
P7-ABB-04 Change Feedback
|
+-------+-------+-------+
v v v v
P1 P2 P3 P4/P5/P6
governance context primitive continuation /
review refresh change controls
The feedback may suspend, constrain, migrate or require reassessment for future use. It does not alter the historical truth of prior executions/evaluations.
GEAIRA-M15-R48 - Assurance-loop flow: M15-FLOW-10 SHALL propagate currentness/change findings into future architecture behavior while preserving historical occurrence evidence and earlier assessment context.
38.26 Flow-to-view / pattern coverage
| Flow | Primary views | Principal M14 patterns | Representative M13 realization choices |
|---|---|---|---|
| FLOW-01 | V01,V02,V08,V09 | P01, optional P03,P14 | RP-01,02,05,07 |
| FLOW-02 | V02,V03,V06,V08,V09 | P02,P04,P14 | RP-01,02,06,07 |
| FLOW-03 | V02,V07,V08 | P05,P14 | RP-01,02,03,07 |
| FLOW-04 | V01,V04,V07,V08 | P06,P14 | RP-02,03,04,05,07 |
| FLOW-05 | V04,V07,V08 | P08,P14 | RP-02,04,05,06,07 |
| FLOW-06 | V05,V07,V08,V09 | P07,P14 | RP-04,05,07 |
| FLOW-07 | V06,V08,V10 | P10,P14 | RP-06,07 |
| FLOW-08 | V06,V10 | P09,P11,P15 | RP-03,07,08 |
| FLOW-09 | V03,V06,V10 | P12,P13,P15 | RP-02,03,07,08 |
| FLOW-10 | V01,V08,V09,V10 | P14 | RP-01..08 |
Across FLOW-01 through FLOW-10, all fifteen M14 patterns are represented either as a primary basis or an explicit composition option.
Coverage closure uses the frozen identifiers rather than inventing M15 aliases:
M12-BND-00 M12-BND-01 M12-BND-02 M12-BND-03 M12-BND-04
M12-BND-05 M12-BND-06 M12-BND-07 M12-BND-08
M13-RP-01 M13-RP-02 M13-RP-03 M13-RP-04
M13-RP-05 M13-RP-06 M13-RP-07 M13-RP-08
M14-P01 M14-P02 M14-P03 M14-P04 M14-P05
M14-P06 M14-P07 M14-P08 M14-P09 M14-P10
M14-P11 M14-P12 M14-P13 M14-P14 M14-P15
GEAIRA-M15-R49 - Pattern coverage: The M15 flow set SHALL collectively demonstrate all fifteen M14 canonical patterns without redefining any pattern as a mandatory lifecycle stage.
GEAIRA-M15-R50 - Realization-pattern compatibility: Each M15 flow SHALL remain separable from M13 topology choice; the table above is illustrative and not an exhaustive or exclusive mapping.
GEAIRA-M15-R51 - Cross-view consistency: If a flow traverses several canonical views, it SHALL reference one coherent architecture state rather than create view-specific versions of the same occurrence.
GEAIRA-M15-R52 - Flow composability: Enterprise scenarios MAY compose multiple M15 flows, provided shared responsibilities/interactions retain one canonical identity and material ordering/control conflicts are explicitly resolved.
38.27 Identity and causal-chain guidance
M15 freezes a generic causal correlation model without requiring one global identifier:
ConsumerRequest C17
-> ServingInteraction S42
-> AIExecution E93
-> ContextAssembly K8
-> PrimitiveRealization R4
-> AgentParticipation G2 (optional)
-> ToolInvocation T6 (optional)
-> Decision D12 (optional)
-> Authorization A7 (optional)
-> Action X5 (optional)
-> Evaluation EV3
-> AssuranceFinding AF2
The architecture requires enough correlation to reconstruct material causality, not identity collapse.
GEAIRA-M15-R53 - No universal correlation identity: Implementations SHALL NOT be required to represent all occurrences with one universal identifier, but SHALL preserve sufficient correlation for material causal reconstruction.
GEAIRA-M15-R54 - Occurrence/version binding: Where material, a flow SHALL retain the Product/contract/context/primitive/policy/version identities that actually governed each occurrence.
GEAIRA-M15-R55 - Evidence chain: Evidence SHOULD identify both the occurrence it describes and the authoritative semantic/version references needed to interpret that occurrence.
GEAIRA-M15-R56 - Causal uncertainty: Where causal linkage or side-effect outcome cannot be established reliably, the architecture SHALL preserve the uncertainty/INDETERMINATE state rather than invent certainty for diagrammatic closure.
38.28 Flow failure and branch discipline
A canonical flow is not a happy-path promise. Material branch outcomes include:
ACCEPTED
COMPLETED
REJECTED
DEFERRED
CONDITIONAL
PARTIAL
FAILED
INDETERMINATE
CANCELLED / REVOKED
EXPIRED / STALE
NOT APPLICABLE
Examples:
- grounding may fail while retrieval succeeds;
- no eligible primitive may exist;
- a serving request may be admitted while execution later fails;
- Authorization may reject an otherwise successful AI result;
- human review may escalate rather than approve;
- an external provider timeout may leave side-effect state indeterminate;
- P7 may invalidate future use after an earlier successful execution.
GEAIRA-M15-R57 - Non-happy-path requirement: Material flows SHALL represent non-success outcomes where those outcomes alter control, reconciliation, consumer behavior or future eligibility.
GEAIRA-M15-R58 - No success propagation: Success at one flow boundary SHALL NOT automatically imply success at downstream boundaries; delivered, accepted, completed, committed and observed remain distinct where material.
GEAIRA-M15-R59 - Revocation propagation: A flow SHALL account for applicable revocation/supersession/currentness changes that occur before a consequential commit or during long-running operation.
GEAIRA-M15-R60 - Retry safety: A flow involving potentially consequential side effects SHALL specify idempotency/reconciliation expectations before retry is considered safe.
38.29 Canonical M15 anti-patterns
M15 freezes the following view/flow anti-patterns:
- Universal Seven-Plane Pipeline - depicting P1->P2->P3->P4->P5->P6->P7 as mandatory order.
- One Diagram Is the Architecture - treating one reference view as the complete responsibility model.
- View-Specific ABB Duplication - cloning an ABB because several stakeholder views show it.
- Flow Creates Product - inferring Product/AI Product identity from a serving or orchestration flow.
- Flow Creates Authority - inferring permission from earlier successful stages.
- Context Arrow Means Grounded - collapsing retrieval/context/grounding distinctions into one arrow.
- Human Box Means Approval - depicting human presence as completed oversight.
- Agent Box Means Autonomy/Authority - inferring autonomy or Authorization from Agent participation.
- Telemetry Arrow Means Assurance - collapsing observation, Evaluation and Assurance.
- Timeout Means Failed - erasing indeterminate side-effect state.
- Latest-Version Diagram - drawing all historical occurrences as if they used the latest/current artifacts.
- Platform Box Owns All Semantics - making a shared platform the semantic authority for peer architectures.
- Federation Means Shared Authority - treating domain federation as one authority scope.
- Provider Box Transfers Accountability - making external provider operation remove enterprise accountability.
- Topology-by-Diagram Position - interpreting diagram columns/rows as mandatory deployment tiers.
- Happy-Path-Only Flow - omitting material reject/defer/revoke/partial/indeterminate branches.
- Single Global Correlation ID - collapsing occurrence identities into one identifier.
- Diagrammatic Conformance - claiming GEAIRA alignment because a solution picture resembles an M15 diagram.
GEAIRA-M15-R61 - Anti-pattern rejection: Architectures claiming GEAIRA alignment SHALL NOT use M15 anti-patterns 1-18 as substitutes for the frozen responsibilities, contracts, authority boundaries or evidence obligations.
GEAIRA-M15-R62 - Diagram is informative projection: Visual similarity to an M15 view or flow SHALL NOT by itself establish conformance; traceable responsibility/contract/control evidence is required.
38.30 M15 review method
A GEAIRA architecture review using M15 SHOULD proceed as follows:
1. Identify the stakeholder question and select the smallest relevant M15 view set.
2. Identify the scenario and select the closest representative M15 flow(s).
3. Trace participating elements back to canonical P1-P7 responsibilities and ABBs.
4. Trace hand-offs to M11 interaction contracts.
5. Identify M12 adjacent authorities and preserve jurisdiction.
6. Select/confirm the actual M13 realization topology.
7. Confirm applicable M14 patterns and paired counterexamples.
8. Verify identity/correlation, authority, currentness, evidence and failure branches.
9. Confirm historical truth and change/revocation behavior.
10. Record deviations or Profile specializations explicitly.
GEAIRA-M15-R63 - View/flow review use: M15 views and flows SHOULD be used as review lenses and traceability aids rather than copied as mandatory solution blueprints.
GEAIRA-M15-R64 - Profile strengthening: Profiles MAY add specialized views/flows, stricter control points or narrower permitted branches, but SHALL NOT weaken M01-M15 authority, identity, currentness, evidence, failure or semantic-boundary rules.
38.31 Frozen M15 decisions
M15 freezes:
10 canonical Reference Views (M15-V01..M15-V10)
10 representative end-to-end flows (M15-FLOW-01..M15-FLOW-10)
1 Reference View metamodel
1 End-to-End Flow metamodel
18 canonical M15 anti-patterns
64 normative M15 rules
The following hard distinctions remain explicit:
Reference View != architecture ownership
Flow != lifecycle
Flow != topology
flow stage != semantic identity
interaction arrow != service/API mandate
flow success != downstream success
human presence != approval
Agent participation != authority
retrieval != grounding
telemetry != Assurance
current/latest != historical truth
shared platform != shared jurisdiction
M15 introduces no new:
- GEAIRM semantic concept or relationship;
- GEAIRA plane;
- canonical ABB;
- M11 canonical inter-plane interaction;
- M12 adjacent-authority boundary;
- M13 realization pattern;
- M14 canonical architecture pattern.
M15 therefore completes the stakeholder-view and representative-flow layer without changing the frozen core architecture.
39. M15 Completion State
GEAIRA-M15 - COMPLETE.
M15 establishes:
- ten canonical stakeholder/reference views, M15-V01 through M15-V10;
- ten representative end-to-end architecture flows, M15-FLOW-01 through M15-FLOW-10;
- one reusable Reference View metamodel;
- one reusable End-to-End Flow metamodel;
- explicit coverage of all seven planes and all twenty-one canonical ABBs;
- collective coverage of all thirty-three M11 canonical interactions;
- representability of all nine M12 adjacent boundaries;
- topology independence from all eight M13 realization patterns;
- representative coverage of all fifteen M14 canonical patterns;
- identity/correlation, causal-chain, failure/branch and historical-truth guidance;
- an eighteen-item canonical M15 view/flow anti-pattern catalogue;
- 64 normative M15 rules.
M15 introduces no new GEAIRM semantic concept/relationship and no new GEAIRA plane, canonical ABB, M11 interaction, M12 boundary, M13 realization pattern or M14 architecture pattern.
Next slice
GEAIRA-M16 - Architecture Invariants, Conformance & Profiling
M16 will consolidate the frozen M01-M15 architecture into explicit architecture invariants, conformance expectations and governed Profile rules. It will define what a solution, domain or enterprise architecture must preserve when claiming GEAIRA alignment, how Profiles may tighten/specialize the baseline without weakening it, and which obligations remain human/assessment-driven versus safely machine-checkable in a later executable formalization programme.
40. GEAIRA-M16 - Architecture Invariants, Conformance & Profiling
40.1 Purpose
GEAIRA-M16 consolidates the architectural commitments established in M01-M15 into a governed conformance model.
M16 answers three questions:
- Which architectural truths must remain valid in every GEAIRA-aligned architecture?
- What evidence is required before an enterprise, domain, solution, Profile or realization may claim GEAIRA alignment?
- How may a Profile specialize the generic architecture without weakening, redefining or bypassing it?
M16 does not introduce a new architecture plane, runtime service, Product type, semantic concept or deployment topology. It is the conformance and specialization layer over the architecture already frozen through M15.
The central rule is:
GEAIRA conformance is conformance to architectural obligations and invariants, not visual similarity to a diagram, use of a preferred platform, implementation of a named technology, or adoption of GEAIRA terminology alone.
GEAIRA-M16-R01 - Obligation-based conformance: A GEAIRA conformance claim SHALL be based on satisfaction of applicable architectural invariants and traceable responsibilities rather than naming, branding, diagram similarity or technology selection.
GEAIRA-M16-R02 - No new semantic authority: M16 SHALL NOT create semantic authority that belongs to GEAIRM, BPS/PIN, AIPS, GEDRM, GEARM, the Agent Ontology, W3C PROV or another authoritative specification.
GEAIRA-M16-R03 - Frozen architecture preservation: M16 SHALL preserve the seven P1-P7 planes, eight X1-X8 overlays, twenty-one canonical ABBs, thirty-three M11 canonical interactions, nine M12 adjacent boundaries, eight M13 realization patterns, fifteen M14 architecture patterns and ten M15 Reference Views/flows without redefining them.
GEAIRA-M16-R04 - No conformance-by-implementation: Use of a product, vendor, framework, gateway, model platform, Agent framework, observability system, HDIP capability or other implementation technology SHALL NOT by itself establish GEAIRA conformance.
40.2 Architecture Invariant metamodel
An Architecture Invariant is a condition that must remain true for every subject to which the invariant applies if that subject is to claim GEAIRA conformance.
Each invariant has the following conceptual fields:
| Field | Meaning |
|---|---|
invariantId | Stable GEAIRA identifier |
name | Human-readable invariant name |
statement | Normative condition that must remain true |
sourceTrace | M01-M15 rules/decisions from which the invariant is derived |
applicability | Circumstances under which it applies |
assessmentMode | STRUCTURAL, HYBRID, or ASSESSMENT |
requiredEvidence | Evidence sufficient to assess the invariant |
failureMeaning | Meaning of violation or inability to determine compliance |
profileBehavior | How a Profile may tighten or specialize the invariant |
M16 uses three assessment modes:
STRUCTURAL
Safely machine-checkable from explicit architecture/conformance records
without pretending to determine architectural judgement.
HYBRID
Partly machine-checkable, but final conformance requires
architecture/governance judgement or contextual evidence.
ASSESSMENT
Requires accountable architecture/governance assessment;
automation may support evidence gathering but cannot safely decide it alone.
GEAIRA-M16-R05 - Stable invariant identity: Every M16 invariant SHALL have a stable identifier independent of implementation technology and Profile naming.
GEAIRA-M16-R06 - Conservative automation: An invariant SHALL be classified STRUCTURAL only where its required condition can be determined safely from explicit machine-readable facts without inferring semantic equivalence, authority, quality, assurance or human judgement.
GEAIRA-M16-R07 - Hybrid honesty: A HYBRID invariant MAY use automated checks for structural portions but SHALL retain an explicit assessment step for the judgement-dependent portion.
GEAIRA-M16-R08 - Assessment not simulated: An ASSESSMENT invariant SHALL NOT be reported as machine-proven merely because supporting telemetry, metadata, policy or configuration records exist.
40.3 Canonical M16 Architecture Invariant Catalogue
M16 freezes forty-eight architecture invariants.
40.3.1 Semantic authority and jurisdiction invariants
| ID | Name | Canonical invariant | Mode |
|---|---|---|---|
| M16-I01 | GEAIRM semantic authority | GEAIRA architecture SHALL NOT redefine GEAIRM-owned enterprise-AI concepts, relationships or invariants incompatibly. | HYBRID |
| M16-I02 | External authority preservation | BPS/PIN, AIPS, GEDRM/GEDRA, GEARM/GEARA, Agent Ontology and W3C PROV authority SHALL remain external where M12 assigns it externally. | HYBRID |
| M16-I03 | Relationship-not-component | A semantic relationship SHALL NOT by itself be treated as requiring a standalone ABB, service, registry, store or API. | ASSESSMENT |
| M16-I04 | Architecture-versus-semantics | New architecture constructs SHALL NOT silently redefine semantic concepts; semantic gaps SHALL be referred to the appropriate semantic authority. | HYBRID |
| M16-I05 | Core identity distinctions | Product, AI Product, Agent, AI Primitive and AI Execution identities SHALL remain distinguishable where applicable. | HYBRID |
| M16-I06 | Domain-plane distinction | A GEAIRM semantic domain or ontology class SHALL NOT be treated as a one-to-one GEAIRA Plane. | STRUCTURAL |
40.3.2 Responsibility, ABB, plane and overlay invariants
| ID | Name | Canonical invariant | Mode |
|---|---|---|---|
| M16-I07 | Responsibility-before-component | Mandatory architecture responsibility SHALL be established before deciding whether a standalone ABB or implementation component is required. | ASSESSMENT |
| M16-I08 | Responsibility-not-standalone-component | Mandatory responsibility SHALL NOT imply a mandatory standalone service/component. | ASSESSMENT |
| M16-I09 | ABB admission discipline | A claimed canonical/specialized ABB SHALL preserve M02 cohesion, traceability, non-duplication, technology-neutrality and realization-independence criteria. | HYBRID |
| M16-I10 | Plane logicality | P1-P7 SHALL remain logical responsibility planes, not mandatory physical tiers, teams, systems or lifecycle stages. | HYBRID |
| M16-I11 | Overlay orthogonality | X1-X8 SHALL remain cross-cutting obligations and SHALL NOT automatically become additional planes or centralized shared services. | HYBRID |
| M16-I12 | Many-to-many realization | One implementation MAY realize several planes/ABBs and one ABB MAY span several realization units without changing canonical identity. | STRUCTURAL |
| M16-I13 | Canonical identity reuse | Views, flows, patterns, Profiles and solution diagrams SHALL reuse canonical ABB/responsibility identities rather than clone them for presentation convenience. | STRUCTURAL |
40.3.3 Governing intent, context and intelligence-mechanism invariants
| ID | Name | Canonical invariant | Mode |
|---|---|---|---|
| M16-I14 | P1/P6 authority separation | P1 governing intent/authority envelope SHALL remain distinct from P6 point-of-use Authorization and action control. | HYBRID |
| M16-I15 | Purpose/authority distinction | Intended purpose SHALL remain distinct from Prompt/Instruction; Delegated Authority and autonomy SHALL remain distinct from Authorization. | HYBRID |
| M16-I16 | Retrieval-grounding distinction | Retrieval SHALL remain distinct from grounding, source authority and correctness/truth. | HYBRID |
| M16-I17 | Context/environment/memory distinction | AI Context SHALL remain distinct from Environment; Memory SHALL NOT automatically imply learning/retraining. | HYBRID |
| M16-I18 | Capability/primitive distinction | Intelligence Capability SHALL remain distinct from AI Primitive, AI Model and AI Product. | HYBRID |
| M16-I19 | Provision/execution distinction | Primitive eligibility/provision by P3 SHALL remain distinct from actual P4 invocation/execution and from Authorization. | HYBRID |
| M16-I20 | Mechanism neutrality | GEAIRA-aligned realization SHALL NOT require AI to be realized only through learned models, LLMs or one intelligence mechanism family. | ASSESSMENT |
40.3.4 Runtime, Agent, serving and Product invariants
| ID | Name | Canonical invariant | Mode |
|---|---|---|---|
| M16-I21 | Agent authority separation | Agent participation SHALL NOT by itself imply autonomy, Delegated Authority or Authorization. | HYBRID |
| M16-I22 | Tool authorization separation | Tool availability/use SHALL NOT by itself establish entitlement or Authorization for a consequential side effect. | HYBRID |
| M16-I23 | Execution identity | AI Execution SHALL remain occurrence-level and SHALL NOT be collapsed into Product, Agent, primitive or runtime-platform identity. | HYBRID |
| M16-I24 | Serving/Product separation | Serving, reuse, API exposure or publication SHALL NOT by itself establish Productness or AI Product qualification. | HYBRID |
| M16-I25 | Request/execution separation | Consumer/serving request identity SHALL remain distinguishable from AI Execution occurrence identity. | STRUCTURAL |
| M16-I26 | External provider accountability | External/managed-provider execution SHALL NOT automatically transfer enterprise accountability or semantic authority to the provider. | ASSESSMENT |
40.3.5 Decision, Authorization, action and human-oversight invariants
| ID | Name | Canonical invariant | Mode |
|---|---|---|---|
| M16-I27 | Effect/Decision distinction | A recommendation or decision-like Intelligence Effect SHALL NOT automatically become the governed operative enterprise Decision. | HYBRID |
| M16-I28 | Decision/Authorization distinction | Decision authority, Delegated Authority and point-of-use Authorization SHALL remain distinguishable where applicable. | HYBRID |
| M16-I29 | Human-presence distinction | Human availability/presence SHALL NOT be reported as completed review, approval, intervention or oversight without evidence that the required human act occurred. | HYBRID |
| M16-I30 | Action/enactment distinction | P6 action commitment/Authorization SHALL remain distinct from technical enactment by P4/P5/external mechanisms. | HYBRID |
| M16-I31 | Bounded autonomy | AutonomousAction or autonomous runtime behavior SHALL remain bounded by applicable authority, safety, currentness and intervention obligations. | ASSESSMENT |
| M16-I32 | Side-effect uncertainty | Potentially consequential side-effect uncertainty SHALL be preserved as INDETERMINATE/reconciliation-required where occurrence cannot be established safely; blind retry SHALL NOT erase uncertainty. | HYBRID |
40.3.6 Evidence, evaluation, assurance and lifecycle invariants
| ID | Name | Canonical invariant | Mode |
|---|---|---|---|
| M16-I33 | Telemetry/evidence distinction | Telemetry/observation SHALL NOT automatically be treated as sufficient Evidence. | HYBRID |
| M16-I34 | Evaluation/assurance distinction | Evaluation SHALL remain distinct from Assurance and an Evaluation Result SHALL NOT automatically constitute an Assurance Finding. | HYBRID |
| M16-I35 | Assurance/permission distinction | Assurance SHALL NOT automatically mean approval, Authorization, low risk or fitness for every context. | HYBRID |
| M16-I36 | Version/lifecycle/currentness distinction | Version identity, Lifecycle State, Currentness Assessment and freshness timestamp SHALL remain distinct. | STRUCTURAL |
| M16-I37 | Latest-not-current | The latest/newest version SHALL NOT automatically be treated as the governed current version. | HYBRID |
| M16-I38 | Historical truth | Supersession, revocation, reevaluation or later evidence SHALL NOT rewrite which versions, authority, context, contracts or outcomes governed a historical occurrence. | HYBRID |
| M16-I39 | Provenance/assurance distinction | Provenance SHALL support traceability but SHALL NOT by itself prove assurance, correctness, authorization or fitness. | HYBRID |
| M16-I40 | Evidence correlation without identity collapse | Material causal/evidence chains SHALL be reconstructible without requiring all occurrences to share one universal identity. | HYBRID |
40.3.7 Interaction, interoperability, topology, pattern and Profile invariants
| ID | Name | Canonical invariant | Mode |
|---|---|---|---|
| M16-I41 | Explicit material contracts | Material cross-plane hand-offs SHALL have explicit architectural interaction/contract obligations; an Interaction Contract SHALL NOT be equated automatically with one API/protocol. | HYBRID |
| M16-I42 | No authority transitivity | Authority, policy, approval or entitlement SHALL NOT become automatically transitive merely because information crosses an architecture boundary. | ASSESSMENT |
| M16-I43 | Semantic compatibility beyond schema | Protocol/schema compatibility SHALL NOT be treated as sufficient proof of semantic, capability, effect, governance or assurance compatibility. | HYBRID |
| M16-I44 | Adjacent-authority independence | M12 mapping/co-realization SHALL NOT transfer semantic authority or automatically establish conformance to an adjacent architecture/specification. | HYBRID |
| M16-I45 | Topology independence | GEAIRA conformance SHALL remain independent of centralized, distributed, federated, edge, embedded, provider or hybrid M13 topology choice. | ASSESSMENT |
| M16-I46 | Shared-platform non-merger | Shared implementation/platform co-realization SHALL NOT merge GEDRA, GEARA, GEAIRA or other architecture jurisdictions. | HYBRID |
| M16-I47 | Projection-not-ownership | M14 patterns and M15 views/flows SHALL remain projections/compositions of canonical architecture and SHALL NOT create new authority, ABB identity, Product identity or universal lifecycle order. | HYBRID |
| M16-I48 | Profile monotonicity | A GEAIRA Profile MAY select, tighten, contextualize, specialize and extend the baseline but SHALL NOT weaken, waive or incompatibly redefine an applicable core invariant. | HYBRID |
GEAIRA-M16-R09 - Core invariant status: M16-I01 through M16-I48 are core GEAIRA architecture invariants for 0.1 and SHALL be evaluated when applicable to a conformance subject.
GEAIRA-M16-R10 - Applicability is contextual, not discretionary: An invariant MAY be marked not applicable only where its stated architectural condition genuinely does not arise in the subject scope; not-applicable SHALL NOT be used as a waiver mechanism.
GEAIRA-M16-R11 - Applicability evidence: Every not-applicable result for a material invariant SHALL include an explicit rationale tied to the assessed subject and scope.
GEAIRA-M16-R12 - Violation consequence: Violation of an applicable core invariant SHALL make the relevant core conformance claim NONCONFORMANT regardless of compensating strengths elsewhere.
GEAIRA-M16-R13 - Indeterminate consequence: Where an applicable invariant cannot be assessed because required evidence is unavailable or materially ambiguous, the conformance result SHALL remain INDETERMINATE rather than be assumed conformant.
GEAIRA-M16-R14 - No score substitution: An aggregate score, maturity rating or percentage SHALL NOT override a failed or indeterminate applicable core invariant.
GEAIRA-M16-R15 - Stronger local rules allowed: A Profile or local architecture MAY impose stricter rules than an invariant where doing so does not redefine its meaning or weaken another core invariant.
GEAIRA-M16-R16 - Invariant traceability: Each invariant assessment SHOULD retain traceability to relevant M01-M15 architecture elements, interactions, evidence and Profile rules.
40.4 Conformance Subject model
A Conformance Subject is the bounded architecture artifact or realized architecture for which a GEAIRA claim is made.
Supported subject classes include:
- Enterprise Architecture - enterprise-wide or multi-domain GEAIRA specialization.
- Domain Architecture - bounded business/technology domain architecture.
- Solution Architecture - architecture of a specific solution, Product or use case.
- GEAIRA Profile - governed specialization of the GEAIRA baseline.
- Realization Mapping - mapping of concrete implementation units to GEAIRA responsibilities/ABBs/contracts.
- Reference/Pattern Composition - a documented reusable composition of GEAIRA patterns/views for a defined context.
A subject claim SHALL define its scope explicitly. A solution is not required to instantiate every plane/ABB/pattern, only those applicable to its actual architecture and claimed scope; however, non-instantiation does not permit violation of applicable invariants.
GEAIRA-M16-R17 - Bounded claim scope: Every conformance claim SHALL identify the assessed subject, version, organizational/solution scope and relevant architecture boundary.
GEAIRA-M16-R18 - No universal-element requirement: Core conformance SHALL NOT require every solution to instantiate all seven planes, twenty-one ABBs, thirty-three interactions, eight realization patterns or fifteen architecture patterns.
GEAIRA-M16-R19 - Applicable responsibility completeness: A conformance subject SHALL account for every material GEAIRA responsibility that is applicable to the claimed scope, whether realized explicitly, embedded, delegated to an adjacent authority, human-mediated or provided by a shared platform.
GEAIRA-M16-R20 - Omission requires rationale: Where a normally relevant plane responsibility, control point or interaction is omitted from the subject, the omission SHALL be justified by scope/applicability rather than by implementation convenience.
GEAIRA-M16-R21 - Embedded realization permitted: Embedded, co-located or shared realization SHALL be conformant where canonical responsibility, authority and evidence remain traceable.
GEAIRA-M16-R22 - Adjacent delegation traceability: Where a responsibility is discharged through GEDRA, GEARA, AIPS/AIPDS, HDIP or another adjacent authority/realization, the mapping SHALL identify the responsible external construct and preserve M12 jurisdiction.
GEAIRA-M16-R23 - Claim granularity: An organization MAY issue distinct conformance claims for enterprise, domain, solution, Profile and realization scopes; one claim SHALL NOT automatically imply the others.
GEAIRA-M16-R24 - Conformance inheritance prohibited: Conformance of a platform, framework or parent architecture SHALL NOT automatically confer conformance on every solution or Product using it.
40.5 Conformance claim types
M16 defines three primary claim types:
40.5.1 Core Architecture Conformance
A claim that a Conformance Subject preserves all applicable M16 core invariants and the applicable responsibilities/contracts of the GEAIRA 0.1 baseline.
40.5.2 Profile Conformance
A claim that a Conformance Subject conforms to the GEAIRA core and to a named governed Profile whose rules are themselves M16-conformant.
40.5.3 Realization Mapping Conformance
A claim that a concrete implementation mapping faithfully realizes the claimed GEAIRA/Profile architecture, including traceability from realization units to responsibilities, ABBs, interactions/contracts, control points and evidence obligations.
These claim types are cumulative where relevant:
GEAIRA Core
|
+--> governed Profile (optional)
|
+--> enterprise/domain/solution architecture
|
+--> concrete realization mapping
A Profile or implementation MAY be useful without a conformance claim, but it SHALL NOT use GEAIRA-conformant language unless the corresponding claim has been assessed.
GEAIRA-M16-R25 - Core before Profile: A Profile Conformance claim SHALL include Core Architecture Conformance; a Profile cannot replace the core.
GEAIRA-M16-R26 - Architecture before realization: A Realization Mapping Conformance claim SHALL reference the GEAIRA or GEAIRA-Profile architecture it purports to realize.
GEAIRA-M16-R27 - Named baseline: Every conformance claim SHALL identify the GEAIRA version and any Profile version against which it was assessed.
GEAIRA-M16-R28 - No implied adjacent conformance: A GEAIRA claim SHALL NOT be presented as GEDRA, GEARA, AIPS, AIPDS, W3C PROV or HDIP conformance unless those claims are independently established.
GEAIRA-M16-R29 - Product claims remain external: GEAIRA conformance SHALL NOT be used as proof that something is a Product, AI Product or a particular AIPS classification.
GEAIRA-M16-R30 - Pattern/view use not a claim type: Use of an M14 pattern or M15 view/flow SHALL NOT by itself constitute a separate conformance claim or proof of core conformance.
GEAIRA-M16-R31 - Profile claim naming: A Profile claim SHALL name the Profile explicitly; generic GEAIRA-conformant wording SHALL NOT conceal material Profile-specific constraints.
GEAIRA-M16-R32 - Scope-sensitive claims: If only part of a broader system has been assessed, the claim SHALL identify that boundary and SHALL NOT imply assessment of unassessed architecture outside it.
40.6 Conformance outcomes
M16 uses three normative assessment outcomes for an actual claim:
| Outcome | Meaning |
|---|---|
CONFORMANT | All applicable core/Profile obligations assessed and satisfied |
NONCONFORMANT | One or more applicable required obligations/invariants are violated |
INDETERMINATE | Required evidence or applicability judgement is materially unresolved |
NOT_APPLICABLE is an invariant-level applicability result, not an overall conformance outcome.
NOT_ASSESSED may be used internally to indicate pending work but is not a valid positive conformance result.
M16 deliberately does not define PARTIALLY_CONFORMANT as a normative positive claim. A subject may report partial implementation progress, but a material failed core invariant remains nonconformance for the claimed scope.
GEAIRA-M16-R33 - Positive claim threshold: CONFORMANT SHALL be issued only where all applicable mandatory core/Profile obligations have been assessed and satisfied.
GEAIRA-M16-R34 - No partial-positive label: PARTIALLY_CONFORMANT, maturity percentages or similar labels SHALL NOT be used as substitutes for the normative M16 conformance outcome.
GEAIRA-M16-R35 - Indeterminate is not pass: INDETERMINATE SHALL NOT be represented as conformant merely because no violation has yet been proven.
GEAIRA-M16-R36 - Not-applicable is local: NOT_APPLICABLE SHALL be recorded per invariant/obligation with rationale and SHALL NOT be used as a blanket subject-level exemption.
GEAIRA-M16-R37 - Reassessment on material change: A material change affecting authority, semantics, capability, Product/Agent identity, contracts, topology, currentness, safety, evidence or Profile applicability SHOULD trigger scoped reassessment.
GEAIRA-M16-R38 - Historical conformance truth: Reassessment SHALL NOT rewrite the historical fact that a subject was assessed against a specific version/Profile/evidence set at an earlier time.
GEAIRA-M16-R39 - Expired evidence handling: Where required conformance evidence becomes non-current or invalid, current conformance status SHALL be reassessed rather than silently inherited from the historical assessment.
GEAIRA-M16-R40 - Claim revocation traceability: Withdrawal, expiration or supersession of a conformance claim SHALL remain historically traceable.
40.7 Minimum Conformance Evidence Package
A positive GEAIRA claim SHOULD be supported by a Conformance Evidence Package (CEP) containing, at minimum:
- subject identifier and version;
- claimed scope and architecture boundary;
- GEAIRA baseline version;
- Profile identifier/version where applicable;
- applicable-plane/responsibility assessment;
- ABB/embedded-realization mapping where applicable;
- M11 interaction/contract mapping for material boundaries;
- M12 adjacent-authority mappings;
- M13 realization/topology declaration;
- M14 pattern composition where used;
- M15 view/flow references where used for review;
- M16 invariant applicability/result matrix;
- material control-point and human-oversight evidence;
- evidence/provenance/currentness references;
- known unresolved items and resulting conformance status;
- accountable assessor/authority and assessment timestamp/version.
The CEP is a conformance record, not a replacement for the architecture itself.
GEAIRA-M16-R41 - Evidence package required: A positive conformance claim SHALL have enough evidence to reconstruct what subject/version/scope was assessed and how applicable invariants were satisfied.
GEAIRA-M16-R42 - Evidence references over duplication: A CEP MAY reference authoritative architecture/evidence records rather than duplicate them, provided references are stable, access-governed and version/currentness aware.
GEAIRA-M16-R43 - Human assessment accountability: Where HYBRID or ASSESSMENT invariants materially determine conformance, the CEP SHALL identify the accountable assessment authority or review mechanism.
GEAIRA-M16-R44 - Evidence integrity: Material conformance evidence SHALL preserve provenance, version/currentness and tamper/alteration expectations appropriate to the claim.
GEAIRA-M16-R45 - No telemetry-only conformance: Operational telemetry alone SHALL NOT be considered sufficient evidence of architecture conformance where semantic, authority, human-oversight or architecture-judgement obligations also apply.
GEAIRA-M16-R46 - Evidence proportionality: The evidence package MAY be proportionate to subject scope/risk, but proportionality SHALL NOT omit evidence needed to decide an applicable core invariant.
GEAIRA-M16-R47 - Evidence historical binding: A CEP SHALL retain the architecture/Profile/evidence versions actually used for the assessment rather than resolve all references dynamically to latest versions.
GEAIRA-M16-R48 - Machine formalization boundary: A future executable GEAIRA formalization MAY automate CEP structure and safely machine-detectable invariant checks but SHALL NOT convert assessment-only architectural judgement into false machine certainty.
40.8 GEAIRA Profile model
A GEAIRA Profile is a governed specialization of the generic architecture for a defined context such as an enterprise, business domain, jurisdiction, platform family, AI risk posture, deployment class or operating model.
A Profile may narrow choice and strengthen obligation while preserving core meaning.
M16 defines the following conceptual Profile fields:
| Field | Purpose |
|---|---|
profileId | Stable Profile identity |
profileVersion | Profile version |
title | Human-readable title |
baselineRef | GEAIRA baseline/version being specialized |
scope | Enterprise/domain/solution/context covered |
jurisdictionContext | Regulatory/organizational context where relevant |
targetSubjectTypes | Subject types expected to use the Profile |
selectedElements | Selected conditional/optional responsibilities, ABBs, interactions, patterns or views |
tightenedObligations | Stronger obligations/thresholds/control requirements |
contextualMappings | Local terminology/architecture mappings to canonical GEAIRA identities |
specializations | Specialized GEAIRA responsibilities/ABBs/contracts preserving parent traceability |
extensions | Profile-owned additional elements |
realizationConstraints | Permitted/required M13 topology or implementation constraints where context-specific |
evidenceRequirements | Additional conformance/assurance evidence |
compatibilityAndSupersession | Profile version compatibility/currentness rules |
authorityAndChangeControl | Accountable Profile owner/change process |
A Profile extension SHOULD use a namespace/identity space distinguishable from core GEAIRA identifiers.
GEAIRA-M16-R49 - Profile baseline binding: Every Profile SHALL identify the GEAIRA baseline/version it specializes.
GEAIRA-M16-R50 - Profile traceability: Every Profile specialization of a GEAIRA element SHALL retain explicit traceability to the canonical element being specialized.
GEAIRA-M16-R51 - Profile extension identity: Profile-owned extensions SHALL use distinguishable identifiers/namespaces and SHALL NOT masquerade as new canonical GEAIRA 0.1 elements.
GEAIRA-M16-R52 - Profile authority: A Profile SHALL identify the accountable authority for Profile definition, approval, versioning and change.
GEAIRA-M16-R53 - Profile currentness: Profile version/currentness SHALL be explicit; use of the latest Profile version SHALL NOT automatically rewrite which version governed historical architectures/assessments.
GEAIRA-M16-R54 - Profile composition: Profiles MAY compose with adjacent GEDRA/GEARA/AIPS/HDIP Profiles, but each architecture/specification authority and conformance claim SHALL remain separately traceable.
GEAIRA-M16-R55 - Nested Profiles: A Profile MAY specialize another conformant GEAIRA Profile, provided the chain to the GEAIRA baseline remains explicit and no inherited core/Profile obligation is weakened.
GEAIRA-M16-R56 - Profile applicability: A Profile SHALL define when it applies; applicability SHALL NOT be inferred merely from technology/platform use unless the Profile explicitly scopes itself that way.
40.9 Permitted Profile operations
A conformant Profile MAY:
- Select conditional/optional GEAIRA responsibilities/patterns that become mandatory in the Profile context.
- Tighten an obligation, threshold, review requirement, safety boundary, evidence requirement or currentness constraint.
- Contextualize generic GEAIRA terms through explicit local mappings.
- Specialize a responsibility, ABB, interaction contract, control point, pattern, view or flow while retaining parent traceability.
- Extend GEAIRA with Profile-owned responsibilities, ABBs, contracts, patterns, views, controls or evidence requirements needed by the context.
- Bind generic responsibilities to local/enterprise standards, policies, registries, platforms or operating-model constructs.
- Restrict the set of acceptable M13 realization patterns or require additional topology constraints.
- Require stronger governance, security, privacy, human oversight, resilience, evidence, evaluation or assurance obligations.
- Define local realization mappings such as HDIP-specific capabilities while preserving generic architecture identity.
For example:
GEAIRA core:
P6 point-of-use Authorization where applicable
High-impact enterprise Profile:
+ mandatory human approval above a defined impact boundary
+ explicit emergency stop
+ immutable evidence retention
+ local Authorization control binding
The Profile has strengthened the core; it has not redefined Authorization.
GEAIRA-M16-R57 - Selection may strengthen: A Profile MAY promote a conditional or optional GEAIRA responsibility/pattern to mandatory for the Profile context.
GEAIRA-M16-R58 - Tightening is monotonic: Profile tightening SHALL increase or preserve the core obligation; it SHALL NOT reduce the protection, distinction, traceability or accountability required by an applicable invariant.
GEAIRA-M16-R59 - Contextual names require mapping: A Profile MAY use local terminology, but local names SHALL map explicitly to canonical GEAIRA identities where they specialize or realize them.
GEAIRA-M16-R60 - Specialized ABB discipline: A Profile-specific ABB specialization SHALL satisfy M02 ABB admission discipline and SHALL NOT be created merely because a local product/service exists.
GEAIRA-M16-R61 - Local technology binding: A Profile MAY require a technology/platform as a local implementation constraint, but SHALL identify that requirement as Profile-specific rather than intrinsic GEAIRA architecture.
GEAIRA-M16-R62 - Stronger evidence allowed: Profiles MAY require stricter or more frequent evidence/evaluation/assurance than the generic baseline.
GEAIRA-M16-R63 - Topology restriction allowed: Profiles MAY restrict acceptable M13 realization patterns for latency, sovereignty, safety, resilience or operating-model reasons without redefining GEAIRA core topology neutrality.
GEAIRA-M16-R64 - Profile extension composition: Profile-owned extensions MAY participate in M11-like contracts and M14/M15 compositions, provided core GEAIRA identities and authority boundaries remain unambiguous.
40.10 Prohibited Profile operations
A Profile SHALL NOT:
- waive or weaken an applicable M16 core invariant;
- redefine GEAIRM or another external authority's semantic concept incompatibly;
- relabel a local service/registry/store as a canonical GEAIRA ABB without satisfying ABB admission/traceability;
- collapse Product, Agent, Primitive, Execution, Decision or Authorization identities for convenience;
- make P1 governing context substitute for P6 point-of-use Authorization where P6 control is applicable;
- make retrieval imply grounding/correctness;
- make telemetry imply Evidence/Assurance;
- make human presence imply approval;
- make the newest version automatically current or rewrite historical truth;
- treat topology/platform choice as proof of architecture conformance;
- transfer semantic/architecture authority merely through co-location or shared platform realization;
- suppress failure/indeterminate/revocation behavior needed by the core architecture;
- use
not applicablesimply to evade an inconvenient core obligation; - introduce a Profile waiver that is then reported as core conformance.
A local architecture may intentionally deviate from GEAIRA. M16 requires that such deviation be reported honestly as nonconformance or as outside the claim scope; it does not prohibit organizations from making local design choices.
GEAIRA-M16-R65 - No waiver semantics: GEAIRA 0.1 defines no Profile mechanism that converts a violated applicable core invariant into conformant status.
GEAIRA-M16-R66 - Deviations remain deviations: A documented local deviation MAY be recorded for governance purposes, but SHALL remain NONCONFORMANT for the affected claim scope until resolved or the claim scope legitimately changes.
GEAIRA-M16-R67 - No authority laundering by Profile: A Profile SHALL NOT transfer or manufacture Product, Agent, data, analytics, provenance, policy or AI semantic authority that belongs to another specification/authority.
GEAIRA-M16-R68 - No implementation promotion: A Profile SHALL NOT promote a vendor/product/service/framework/registry/store into canonical GEAIRA status merely because it is mandated locally.
GEAIRA-M16-R69 - No history rewrite: Profile upgrades, migration or reassessment SHALL preserve the Profile/baseline versions and material facts that governed historical occurrences and historical conformance assessments.
GEAIRA-M16-R70 - No hidden weakening through mappings: Contextual mapping, aliasing or specialization SHALL NOT change the meaning of a core distinction while claiming semantic equivalence.
GEAIRA-M16-R71 - Conformance language discipline: A subject with known applicable invariant violations SHALL NOT be described as GEAIRA-conformant merely because the violations are accepted by local governance.
GEAIRA-M16-R72 - Future formalization fidelity: Any later GEAIRA machine-readable/executable conformance programme SHALL preserve M16's separation between structural checks, hybrid assessment and architecture judgement and SHALL NOT weaken the human-readable normative baseline.
40.11 Conformance assessment workflow
A GEAIRA 0.1 conformance assessment SHOULD follow this sequence:
1. Identify Conformance Subject, version, scope and claim type.
2. Identify GEAIRA baseline and applicable Profile(s).
3. Determine applicable P1-P7 responsibilities and X1-X8 overlays.
4. Trace canonical/specialized ABB and embedded realization responsibility.
5. Identify material M11 interactions/contracts and control points.
6. Identify M12 adjacent authorities and preserve jurisdiction.
7. Declare actual M13 realization/topology pattern(s).
8. Record M14 pattern composition and M15 views/flows used for review.
9. Assess M16-I01..I48 applicability.
10. Gather structural, hybrid and assessment evidence.
11. Record each invariant result with rationale/evidence.
12. Determine overall claim outcome: CONFORMANT, NONCONFORMANT or INDETERMINATE.
13. Record accountable assessor/authority and assessment version/time.
14. Preserve the evidence package for historical traceability.
15. Reassess when material semantic, authority, Profile, topology, control or currentness change occurs.
M16 does not require a centralized certification body. Enterprises may define governance mechanisms appropriate to their context, but conformance claims must remain reproducible and attributable.
40.12 Conformance anti-patterns
M16 freezes the following eighteen conformance/Profile anti-patterns:
- Diagram Match Conformance - declaring conformance because a solution diagram resembles M15.
- Technology Badge Conformance - declaring conformance because a preferred AI platform/framework is used.
- Terminology Conformance - renaming local components with GEAIRA labels without preserving responsibility semantics.
- Percentage Conformance - using an aggregate score to hide a failed core invariant.
- Waiver-to-Pass - accepting a violated core invariant and still reporting core conformance.
- Not-Applicable Shopping - marking difficult obligations N/A without scope-based rationale.
- Inherited Platform Conformance - assuming every workload is conformant because its shared platform is conformant.
- Profile Rewrites Core - using a Profile to redefine a core GEAIRA distinction.
- Profile Owns External Semantics - copying BPS/AIPS/GEDRM/GEARM/Agent/PROV semantics into a local Profile as owned terms.
- Vendor Becomes ABB - turning a local product/service into a canonical ABB by naming convention.
- Structural-Only Assurance - treating schema/configuration validity as proof of architecture judgement.
- Telemetry-Only Evidence Package - using logs/metrics as the entire conformance basis.
- Latest-Only Assessment - losing the versions that governed historical conformance.
- One Global Claim - issuing one enterprise-wide claim that obscures unassessed domains/solutions.
- Adjacent Conformance Leakage - inferring GEDRA/GEARA/AIPS/HDIP conformance from GEAIRA conformance or vice versa.
- Pattern Means Conformant - claiming alignment because an M14 pattern is implemented.
- Maturity Equals Conformance - using capability maturity as a substitute for invariant satisfaction.
- Automated Judgement Theatre - presenting an automated rule engine as definitive proof of HYBRID/ASSESSMENT invariants.
40.13 Frozen M16 decisions
M16 freezes:
48 core Architecture Invariants (M16-I01..M16-I48)
3 invariant assessment modes (STRUCTURAL, HYBRID, ASSESSMENT)
6 Conformance Subject classes
3 primary conformance claim types
3 normative overall outcomes
16 minimum Conformance Evidence Package dimensions
16 conceptual Profile fields
9 permitted Profile operation classes
14 prohibited Profile operation classes
18 conformance/Profile anti-patterns
72 normative M16 rules
The following principles are now explicit:
conformance != diagram similarity
conformance != technology choice
conformance != maturity score
Profile != waiver
Profile tightening != semantic redefinition
not applicable != exemption by convenience
machine-checkable != architecture judgement
adjacent mapping != adjacent conformance
shared platform != inherited conformance
historical reassessment != historical rewrite
M16 introduces no new GEAIRM semantic concept/relationship and no new GEAIRA plane, canonical ABB, M11 interaction, M12 boundary, M13 realization pattern, M14 architecture pattern or M15 canonical view/flow.
41. M16 Completion State
GEAIRA-M16 - COMPLETE.
M16 establishes:
- forty-eight frozen core architecture invariants;
- explicit STRUCTURAL, HYBRID and ASSESSMENT checkability classes;
- a bounded Conformance Subject model;
- Core, Profile and Realization Mapping conformance claim types;
CONFORMANT,NONCONFORMANTandINDETERMINATEnormative outcomes;- a sixteen-dimension minimum Conformance Evidence Package;
- a sixteen-field governed Profile model;
- permitted monotonic specialization operations and explicit prohibited weakening/waiver operations;
- an eighteen-item conformance/Profile anti-pattern catalogue;
- 72 normative M16 rules.
M16 therefore provides a normative basis from which a later executable GEAIRA conformance programme may safely automate structural portions without pretending to automate architectural judgement.
Next slice
GEAIRA-M17 - Consolidation & 0.1 Baseline Freeze
M17 will reconcile M01-M16 end to end, verify all frozen identifiers/counts and cross-slice boundaries, close residual contradictions or duplication, establish the canonical human-readable GEAIRA 0.1 baseline, and define the hand-off into any subsequent machine-readable/executable architecture formalization programme.
42. GEAIRA-M17 - Consolidation & 0.1 Baseline Freeze
42.1 Purpose
GEAIRA-M17 closes the human-readable GEAIRA 0.1 architecture programme by reconciling M01-M16 as one coherent normative baseline.
M17 does not introduce a new architecture layer. It freezes what the programme has already established, resolves canonical counts from the specification itself, records the final authority and identity boundaries, defines the change policy for the frozen baseline, and establishes the fidelity contract for any later machine-readable or executable formalization.
The M17 closure principle is:
freeze meaning before formalizing representation
A later registry, ontology, schema, SHACL layer, Profile schema, validator, testcase suite or release package may encode the GEAIRA 0.1 baseline. It MUST NOT silently redesign it.
GEAIRA-M17-R01 - Human-readable baseline authority: The M01-M17 human-readable GEAIRA 0.1 specification SHALL be the normative architecture baseline against which subsequent GEAIRA 0.1 machine-readable and executable artifacts are derived.
GEAIRA-M17-R02 - Closure not redesign: M17 SHALL consolidate and freeze M01-M16 and SHALL NOT use programme closure as a reason to introduce a new plane, overlay, canonical ABB, canonical interaction, adjacent boundary, realization pattern, architecture pattern, Reference View, flow or semantic authority.
GEAIRA-M17-R03 - Specification-derived counts: Canonical baseline counts SHALL be derived from the normative specification identifiers and not from conversational summaries, diagrams or secondary documentation.
GEAIRA-M17-R04 - Human-readable precedence: Where a later GEAIRA 0.1 machine artifact conflicts with the frozen human-readable baseline, the human-readable baseline SHALL prevail until a governed correction or version change is issued.
GEAIRA-M17-R05 - No semantic backflow: GEAIRA executable formalization SHALL NOT retroactively redefine GEAIRM, BPS/PIN, AIPS, GEDRM, GEARM, Agent Ontology or W3C PROV semantics.
GEAIRA-M17-R06 - No implementation backflow: Enterprise-specific realization choices such as HDIP capabilities, vendor platforms, services, registries, stores or protocols SHALL NOT be promoted into GEAIRA core merely because they realize frozen responsibilities.
GEAIRA-M17-R07 - Publication status separation: Freezing the human-readable 0.1 architecture baseline SHALL NOT by itself imply final publication or Released status; publication/promotion remains a separate governed decision.
GEAIRA-M17-R08 - Baseline identity: The closed M17 subject SHALL be identified as the GEAIRA 0.1 human-readable baseline and SHALL remain distinguishable from later executable release candidates or published releases.
42.2 Canonical 0.1 baseline inventory
M17 freezes the following architecture inventory:
| Baseline subject | Frozen count |
|---|---|
| Architecture Planes | 7 |
| Cross-Cutting Overlays | 8 |
| Canonical plane Architectural Responsibilities | 96 |
| Canonical ABBs | 21 |
| Plane-level interaction declarations from P1-P7 | 70 |
| M11 interaction families | 6 |
| M11 canonical inter-plane interactions | 33 |
| Plane-local logical control-point classes | 20 |
| M11 cross-plane control-point classes | 4 |
| M12 adjacent boundaries | 9 |
| M13 deployment-neutral realization patterns | 8 |
| M14 canonical architecture patterns | 15 |
| M14 paired boundary counterexamples | 15 |
| M15 canonical Reference Views | 10 |
| M15 representative flows | 10 |
| M16 core architecture invariants | 48 |
| M01-M16 normative rules | 853 |
The seven Plane responsibility/ABB counts are:
| Plane | Canonical responsibilities | Canonical ABBs |
|---|---|---|
| P1 - AI Intent, Purpose & Authority | 12 | 2 |
| P2 - Knowledge, Context & Grounding | 13 | 3 |
| P3 - Intelligence Capability & Primitive | 13 | 3 |
| P4 - AI Execution, Agentic & Environment | 14 | 3 |
| P5 - AI Serving, Product & Integration | 14 | 3 |
| P6 - Decision, Action & Human Oversight | 14 | 3 |
| P7 - Evaluation, Assurance, Observability & Evolution | 16 | 4 |
| Total | 96 | 21 |
The seven source interaction counts are:
P1 8
P2 9
P3 9
P4 11
P5 10
P6 12
P7 11
-------
70
M11 reconciles these source declarations into 33 canonical inter-plane interactions rather than duplicating source and target descriptions.
GEAIRA-M17-R09 - Plane inventory freeze: P1-P7 SHALL remain the seven canonical GEAIRA 0.1 Planes.
GEAIRA-M17-R10 - Overlay inventory freeze: X1-X8 SHALL remain the eight canonical GEAIRA 0.1 Cross-Cutting Overlays.
GEAIRA-M17-R11 - Responsibility inventory freeze: The 96 P1-P7 canonical Architectural Responsibility identifiers established in M04-M10 SHALL remain the canonical plane responsibility inventory for GEAIRA 0.1.
GEAIRA-M17-R12 - ABB inventory freeze: The 21 P1-P7 canonical ABB identifiers established in M04-M10 SHALL remain the canonical ABB inventory for GEAIRA 0.1.
GEAIRA-M17-R13 - Source interaction inventory freeze: The 70 P1-P7 plane-level interaction declarations established in M04-M10 SHALL remain traceable source declarations for the M11 consolidation.
GEAIRA-M17-R14 - M11 interaction inventory freeze: M11-INT-01 through M11-INT-33 SHALL remain the canonical inter-plane interaction catalogue for GEAIRA 0.1.
GEAIRA-M17-R15 - Pattern/view/invariant inventory freeze: M13-RP-01..08, M14-P01..P15, M15-V01..V10, M15-FLOW-01..10 and M16-I01..I48 SHALL retain their frozen identities.
GEAIRA-M17-R16 - Identifier stability: A frozen canonical identifier SHALL NOT be silently reused for a materially different meaning in GEAIRA 0.1.
42.3 Frozen architecture authority model
GEAIRA 0.1 remains one member of a deliberately federated specification landscape.
| Authority/specification | Canonical jurisdiction relative to GEAIRA |
|---|---|
| GEAIRM | Generic enterprise-AI semantics consumed by GEAIRA |
| GEAIRA | Generic enterprise-AI architectural responsibility and interaction structure |
| BPS / PIN | Generic Product ontology and Product Intrinsic Nature |
| AIPS | AI Product qualification, AI Product semantics, Intelligence Capability/Primitive and AI risk/governance semantics |
| AIPDS | Operational/deployment specialization for a qualified AI Product version |
| KivaNura Agent Ontology | Agent identity and taxonomy |
| GEDRM / GEDRA | Generic Data semantics / Data reference architecture |
| GEARM / GEARA | Generic Analytics/Information semantics / reference architecture |
| W3C PROV | Generic provenance semantics |
| HDIP | Enterprise-specific target realization/profile that may co-realize multiple generic architectures |
The closure model is therefore:
semantic authority != architecture authority != realization authority
and:
mapping != ownership transfer
co-location != authority transfer
shared platform != architecture merger
GEAIRA-M17-R17 - Semantic foundation boundary: GEAIRM SHALL remain the semantic foundation for generic enterprise-AI concepts used by GEAIRA; GEAIRA SHALL NOT become a competing semantic ontology.
GEAIRA-M17-R18 - Product authority boundary: BPS/PIN SHALL remain authoritative for generic Product semantics and AIPS SHALL remain authoritative for AI Product qualification and AI Product-specific semantics.
GEAIRA-M17-R19 - Agent authority boundary: GEAIRA Agent-related responsibilities SHALL reference rather than replace the Agent identity/taxonomy authority established outside GEAIRA.
GEAIRA-M17-R20 - Data and analytics authority boundary: GEDRM/GEDRA and GEARM/GEARA SHALL remain independent peer authorities for Data and Analytics/Information concerns intersecting GEAIRA.
GEAIRA-M17-R21 - Provenance authority boundary: W3C PROV MAY provide generic provenance semantics used by GEAIRA evidence/provenance responsibilities; use of PROV SHALL NOT by itself establish GEAIRA Evidence or Assurance sufficiency.
GEAIRA-M17-R22 - AIPDS boundary: AIPDS MAY specialize how a qualified AI Product version realizes GEAIRA responsibilities, but SHALL NOT be treated as the generic GEAIRA architecture itself.
GEAIRA-M17-R23 - HDIP boundary: HDIP MAY co-realize GEAIRA, GEDRA, GEARA and AIPS responsibilities, but HDIP-specific planes, layers, registries and services SHALL remain enterprise/profile realization constructs unless separately admitted into a generic specification.
GEAIRA-M17-R24 - Independent conformance: Conformance to one adjacent specification/architecture SHALL NOT automatically confer conformance to another unless a governed compound Profile explicitly assesses each applicable authority.
42.4 Canonical distinction ledger
M17 freezes the following interpretation guardrails as a consolidated architecture ledger:
GEAIRM semantic relationship != GEAIRA architecture component
mandatory responsibility != mandatory standalone ABB != mandatory standalone component
Plane != tier != team != runtime sequence
Cross-Cutting Overlay != mandatory shared service/control plane
Governing intent != runtime Prompt/Instruction
P1 governing envelope != P6 point-of-use Authorization
AI Context != Environment
retrieval != grounding != correctness
Memory != learning/retraining
Intelligence Capability != AI Primitive
AI Primitive != AI Model
primitive provision != AI Execution
Agent != AI Primitive != AI Execution != AI Product
Agent participation != authority
Tool use != Authorization
serving != Productness != AI Product qualification
Consumer Request != AI Execution
recommendation/decision-like effect != operative Decision
Delegated Authority != Authorization
human presence != completed review/approval
Action commitment != technical enactment
telemetry != Evidence
evaluation != Assurance
Assurance != approval/Authorization
Lifecycle State != Currentness Assessment
latest version != governed current version
schema/protocol compatibility != semantic compatibility
M13 realization pattern != M14 architecture pattern
Reference View != architecture ownership
representative flow != universal lifecycle
Profile != waiver
shared platform != inherited conformance
GEAIRA-M17-R25 - Distinction preservation: Subsequent representations, Profiles and realizations SHALL preserve the canonical distinctions consolidated in this ledger.
GEAIRA-M17-R26 - Responsibility/component separation: Machine formalization SHALL preserve the M02 rule that responsibility existence does not prove the need for a standalone ABB or implementation component.
GEAIRA-M17-R27 - Plane neutrality: Plane numbering SHALL remain organizational and SHALL NOT be encoded as a mandatory sequential runtime pipeline.
GEAIRA-M17-R28 - Overlay orthogonality: Cross-Cutting Overlays SHALL remain orthogonal concerns and SHALL NOT be formalized as additional P8+ Planes or mandatory centralized services.
GEAIRA-M17-R29 - Context/grounding separation: Retrieval, grounding-source qualification, context assembly and correctness/fitness claims SHALL remain independently representable.
GEAIRA-M17-R30 - Capability/primitive/execution separation: Intelligence Capability, AI Primitive and AI Execution SHALL remain distinct identities/roles through formalization and realization.
GEAIRA-M17-R31 - Decision/authority separation: Recommendation, Decision, Delegated Authority, Authorization, Action and human oversight SHALL remain separately traceable where applicable.
GEAIRA-M17-R32 - Evidence/assurance separation: Observation, telemetry, Evidence, Evaluation Result, Assurance Finding and Currentness Assessment SHALL NOT be collapsed into one generic status concept.
42.5 Interaction, control and failure-semantics closure
M11 is the canonical cross-plane contract layer. M04-M10 plane interactions remain source declarations; M11-INT-01..33 are the consolidated inter-plane interaction identities.
An interaction contract remains conceptual and may be realized through synchronous calls, asynchronous messages, streams, files, shared governed state, embedded communication, human workflow or another suitable mechanism.
M17 also freezes the cross-programme failure principle:
delivered != accepted != completed != side effect committed != outcome observed
and:
timeout + uncertain side effect -> INDETERMINATE / reconciliation required
Blind retry is not a generic safe default for consequential operations.
GEAIRA-M17-R33 - M11 contract authority: M11 SHALL remain the canonical GEAIRA 0.1 cross-plane interaction/contract model.
GEAIRA-M17-R34 - Source-to-canonical traceability: Each material plane-level interaction used by a realization SHALL remain traceable to the applicable M11 interaction or to an explicitly retained plane-internal/external boundary.
GEAIRA-M17-R35 - Protocol neutrality: Canonical interactions SHALL NOT be redefined as REST, gRPC, event bus, shared database, file transfer or any other mandatory protocol realization.
GEAIRA-M17-R36 - Authority non-transitivity: Authority, permission and accountability SHALL NOT be assumed to transitively propagate merely because data, context, instructions or results cross an interaction boundary.
GEAIRA-M17-R37 - Evidence propagation: Evidence/provenance obligations crossing boundaries SHALL preserve occurrence identity, source identity, relevant version/currentness and material transformation/correlation context.
GEAIRA-M17-R38 - Outcome semantic fidelity: Implementations SHALL preserve material distinctions among accepted, completed, rejected, deferred, conditional, partial, failed, indeterminate, cancelled/revoked, expired/stale and not-applicable outcomes where applicable.
GEAIRA-M17-R39 - Consequential retry discipline: A realization SHALL NOT blindly retry a consequential interaction when prior side-effect state is materially indeterminate.
GEAIRA-M17-R40 - Control-point realization neutrality: Logical control points MAY be centralized, distributed, embedded, human-mediated or co-located and SHALL NOT imply standalone control services.
42.6 Historical truth, version and currentness closure
M17 freezes one historical-truth principle across all planes and adjacent boundaries:
later change constrains future interpretation/use
!=
rewrite of what historically occurred
Examples include:
- a context source later being superseded;
- an AI Primitive later becoming non-current;
- a serving contract later being deprecated;
- an Authorization later being revoked;
- an Evaluation criterion later changing;
- a Profile later being superseded;
- an architecture mapping later migrating topology.
Historical records retain the versions, facts and governing state that actually applied at the time of the occurrence or assessment.
GEAIRA-M17-R41 - Historical truth preservation: Later lifecycle, policy, Profile, topology, contract, evidence or currentness change SHALL NOT mutate the factual identity/version history of an earlier occurrence.
GEAIRA-M17-R42 - Currentness contextuality: Currentness SHALL remain purpose/scope/context dependent and SHALL NOT be reduced to latest-version or latest-timestamp semantics.
GEAIRA-M17-R43 - Supersession discipline: Supersession SHALL affect future preference/eligibility as governed but SHALL NOT imply deletion or invalidation of historical references.
GEAIRA-M17-R44 - Contract history: Historical interactions SHALL retain the material contract/version that governed them even after contract evolution.
GEAIRA-M17-R45 - Authorization history: Revocation or expiry MAY prevent future action but SHALL NOT erase a historically valid Authorization occurrence.
GEAIRA-M17-R46 - Assurance history: A later Assurance Finding or criterion change MAY alter current fitness but SHALL NOT rewrite the historical Evaluation/Assurance result that was produced under earlier evidence and criteria.
GEAIRA-M17-R47 - Profile history: Historical conformance assessments SHALL retain the GEAIRA baseline/Profile versions against which they were assessed.
GEAIRA-M17-R48 - Topology history: Deployment migration among M13 realization patterns SHALL preserve material historical realization/topology facts needed for evidence, incident analysis and audit.
42.7 Pattern, view, flow and conformance closure
M13, M14, M15 and M16 form distinct layers:
M13 - where/how responsibilities may be realized
M14 - reusable architecture problem/solution arrangements
M15 - stakeholder projections and representative traversals
M16 - invariant/conformance/Profile rules
None is a substitute for the others.
GEAIRA-M17-R49 - Realization/pattern separation: M13 realization topology SHALL remain distinct from M14 architecture problem/solution patterns.
GEAIRA-M17-R50 - Pattern identity discipline: Implementing an M14 pattern SHALL NOT by itself create Product, Agent, Primitive, capability, authority or conformance identity.
GEAIRA-M17-R51 - View projection discipline: M15 Reference Views SHALL remain projections of canonical architecture and SHALL NOT own duplicate ABB or interaction identities.
GEAIRA-M17-R52 - Flow non-universality: M15 representative flows SHALL NOT be treated as one mandatory lifecycle or runtime sequence.
GEAIRA-M17-R53 - Invariant supremacy for conformance: Applicable M16 core invariants SHALL remain the normative basis for GEAIRA conformance claims rather than aggregate maturity/coverage scores.
GEAIRA-M17-R54 - No waiver-to-pass: Profiles and local governance decisions SHALL NOT convert an applicable violated core invariant into a conformant result.
GEAIRA-M17-R55 - Indeterminate honesty: Materially insufficient evidence SHALL yield INDETERMINATE rather than assumed conformity.
GEAIRA-M17-R56 - Scope-sensitive conformance: A subject need not instantiate every GEAIRA plane/ABB/pattern, but every applicable responsibility and invariant within the stated claim scope SHALL be accounted for and traceable.
42.8 0.1 change-control and errata policy
After M17, the human-readable GEAIRA 0.1 baseline is frozen for formalization.
Permitted non-semantic corrections MAY include:
- spelling, grammar and formatting fixes;
- broken cross-reference fixes;
- correction of an obviously erroneous secondary count where canonical identifiers already determine the answer;
- clarification that does not change normative meaning, applicability or identity.
A change is normative/material when it alters, removes or redefines a frozen identity, responsibility, distinction, interaction obligation, invariant, authority boundary, applicability rule or conformance result.
Material changes require governed versioning rather than silent mutation.
GEAIRA-M17-R57 - Frozen-core change discipline: After M17, a material change to a frozen GEAIRA 0.1 core element SHALL require a governed amendment/version decision and SHALL NOT be hidden inside machine-artifact generation.
GEAIRA-M17-R58 - Editorial errata: Non-semantic editorial corrections MAY be issued against 0.1 provided they do not change canonical identity, normative meaning or conformance outcome.
GEAIRA-M17-R59 - Count correction rule: A secondary prose count MAY be corrected as editorial errata when the canonical identifier inventory already unambiguously determines the count and no normative identity/meaning changes.
GEAIRA-M17-R60 - Material-defect handling: If executable formalization exposes a genuine contradiction or material baseline defect, the issue SHALL be recorded and resolved through the governed human-readable baseline/change process before machine artifacts claim the corrected meaning.
42.9 Hand-off to machine-readable/executable formalization
M17 does not itself create GEAIRA machine-readable architecture artifacts. It freezes the source from which they may now be built.
A subsequent executable programme SHOULD, at minimum, be able to represent and reconcile:
- the GEAIRA architecture element registry, including Planes, Overlays, Responsibilities, ABBs, control-point classes, boundaries, realization patterns, architecture patterns, Reference Views, flows, invariants and Profile/conformance constructs;
- architecture relationships and traceability among those elements;
- the M11 canonical interaction catalogue and contract semantics;
- the M12 external/adjacent authority references without ownership transfer;
- the M16 invariant catalogue and checkability classes;
- structural schemas and Profile constraints;
- examples and positive/negative boundary testcases;
- deterministic structural conformance checks where safe;
- explicit human/hybrid assessment obligations where automation is insufficient;
- cross-artifact reconciliation, regression and release-closure controls.
A later executable programme MAY use a staged sequence similar to proven KivaNura specification programmes, but the artifact sequence itself is not part of the generic GEAIRA architecture.
GEAIRA-M17-R61 - Formalization derivation: Every canonical machine-readable GEAIRA 0.1 element SHALL trace to a frozen human-readable M01-M17 authority or to an explicitly identified machine-support construct that does not redefine architecture meaning.
GEAIRA-M17-R62 - Conservative automation: Executable validation SHALL automate only those obligations that can be safely determined from available machine evidence and SHALL preserve M16 STRUCTURAL/HYBRID/ASSESSMENT distinctions.
GEAIRA-M17-R63 - Cross-artifact reconciliation: A future executable GEAIRA 0.1 release candidate SHALL include deterministic reconciliation proving that canonical identifiers, relationships, invariant references and frozen counts remain consistent across machine artifacts and the human-readable baseline.
GEAIRA-M17-R64 - Release closure separation: Completion of machine formalization SHALL still require a separate governed release-closure/publication decision; M17 alone closes the human-readable architecture programme, not the entire publication lifecycle.
42.10 M01-M17 normative-rule ledger
M17 freezes the following rule-count ledger directly from canonical rule identifiers:
| Slice | Normative rules |
|---|---|
| M01 | 29 |
| M02 | 42 |
| M03 | 42 |
| M04 | 48 |
| M05 | 52 |
| M06 | 52 |
| M07 | 56 |
| M08 | 40 |
| M09 | 50 |
| M10 | 56 |
| M11 | 59 |
| M12 | 65 |
| M13 | 60 |
| M14 | 66 |
| M15 | 64 |
| M16 | 72 |
| M17 | 64 |
| Total | 917 |
The M11 value includes the explicitly suffixed canonical rule GEAIRA-M11-R12A in addition to the numeric R01-R58 sequence; canonical rule counts therefore use complete rule identifiers rather than numeric-only parsing.
42.11 Frozen GEAIRA 0.1 human-readable baseline
The resulting canonical 0.1 architecture is summarized as:
GEAIRM semantics
|
v
GEAIRA responsibility-first architecture
|
+-- 7 Planes
+-- 8 Cross-Cutting Overlays
+-- 96 canonical plane responsibilities
+-- 21 canonical ABBs
+-- 33 canonical inter-plane interactions / 6 families
+-- 9 adjacent architecture/authority boundaries
+-- 8 deployment-neutral realization patterns
+-- 15 canonical architecture patterns
+-- 10 canonical Reference Views
+-- 10 representative flows
+-- 48 core architecture invariants
|
v
Profiles / enterprise-domain-solution architectures
|
v
Concrete realizations such as HDIP-aligned implementations
The architecture remains technology-neutral and realization-neutral while preserving strong semantic, authority, control, evidence, lifecycle and conformance boundaries.
M17 introduces no new GEAIRM semantic concept/relationship and no new GEAIRA plane, overlay, canonical ABB, M11 interaction, M12 boundary, M13 realization pattern, M14 architecture pattern, M15 canonical view/flow or M16 core invariant.
43. M17 Completion State
GEAIRA-M17 - COMPLETE.
GEAIRA 0.1 human-readable architecture baseline - FROZEN.
M17 establishes:
- a canonical specification-derived inventory across M01-M16;
- 96 frozen plane Architectural Responsibilities and 21 frozen canonical ABBs;
- confirmation that 70 source plane interactions reconcile into the 33 M11 canonical inter-plane interactions;
- explicit closure of the semantic/architecture/realization authority model;
- a consolidated distinction ledger across intent, context, mechanisms, runtime, Product, Decision/Authorization, evidence and lifecycle/currentness;
- a frozen historical-truth and failure/outcome discipline;
- an explicit 0.1 errata/change-control policy;
- a machine-formalization fidelity contract;
- 64 normative M17 rules;
- 917 total normative M01-M17 rules.
The M01-M17 human-readable architecture programme is therefore closed for GEAIRA 0.1. Any subsequent machine-readable/executable programme is downstream formalization of this frozen baseline and is not permitted to silently redesign it.
Next programme
GEAIRA 0.1 machine-readable / executable formalization - to be initiated separately from the frozen M17 baseline.