AI Product (AIPS) Model Reference
This document defines the ontology-aligned conceptual AIPS 0.1 AI Product model after the Product Intrinsic Nature (PIN) and AI Product qualification realignment.
The AIPS 0.1 Core, AIPROD, and AIPDS JSON Schema, JSON-LD, RDF/Turtle, SHACL, and marketplace publish projection under static/spec/aips/0.1 are aligned to this model. Governance/risk computation is ontology-aligned through evidence-backed Risk Driver Assessments and a version-specific Governance Obligation Profile. AIPS-DPP remains scheduled for later ontology-aware generalization.
AIPS distinguishes:
- Logical Product Identity — the continuing Product across versions;
- AI Product Version — one immutable, governed Product version;
- AIPROD — the immutable semantic descriptor for that version;
- AI Qualification Evidence (AQE) — evidence that the version currently qualifies for the AI Product Family;
- AI Product Type (APT) — Primary or Derived intelligence provenance;
- AIPDS — deployment/execution realization;
- DPP — Product trust/passport projection.
The Product itself is not an immutable semantic artifact. The versioned descriptor and evidence artifacts are immutable.
1. Conceptual Overview
2. Logical Product Identity
ProductIdentity represents the continuing Product identity across versions.
A logical Product identity can persist even when a new version changes implementation, dependencies, APT, or — in exceptional cases — intrinsic Product Family qualification.
Example:
Fraud Assessment Product
├── v1 → AI Product / Primary / model-backed
└── v2 → re-evaluated after model removal
Historical versions retain their historical ontology and evidence.
3. AIProductVersion
AIProductVersion represents one immutable version of a Product that has qualified for the AI Product Family.
A published AI Product version must be:
- governed;
- self-describing;
- AI-qualified with evidence;
- APT-classified;
- capability-described;
- risk-evaluated;
- provenance-aware;
- version immutable.
The version should carry or link to:
| Property | Meaning |
|---|---|
version | Product version |
description | Human-readable semantic description |
intelligenceProposition | Distinct intelligence capability being productized |
aiProductOwner | Accountable AI Product Owner |
hasAIQualification | AI Qualification Evidence |
hasAIProductType | Primary or Derived |
hasIntelligenceCapability | One or more IC values |
hasConsumptionScope | Intended scope(s) of consumption |
hasPurpose | Purpose classification(s) |
hasExecutionMode | Interaction/execution modes |
hasIntelligenceEffect | Effects produced or caused |
hasAgencyAutonomy | Authority/autonomy classification |
hasEmbodiment | Relationship to physical world |
hasEffectiveRiskTier | Computed AIPS R0–R4 risk tier |
4. AI Qualification Evidence (AQE)
AIQualificationEvidence records why the exact Product version currently qualifies for the AI Product Family.
Mandatory semantics
AQE must establish:
- the Product version already satisfies BPS Product qualification;
- artificial intelligence is the Product's value-bearing essence;
- the Product owns a distinct intelligence proposition;
- the Product has at least one qualifying direct or transitive AI Primitive path;
- a governed Product boundary exists.
Qualification status
Conceptual enumeration:
CURRENTSTALEREVALIDATION_REQUIREDINVALID
A version with INVALID qualification cannot be represented as a current AI Product version.
5. AI Primitive Evidence
AIPrimitiveEvidence links the AI Product version to the intelligence-producing mechanism(s) that establish its qualification.
Possible primitive types include:
- Learned Model;
- Symbolic Reasoner;
- Knowledge Reasoner;
- Planner;
- Intelligent Search;
- qualifying Optimizer;
- Hybrid Intelligence Engine.
A model is one possible AI Primitive, not a universal requirement.
For a Derived AI Product the evidence may be transitive:
Derived AIP
→ Upstream AIP
→ Primary AIP
→ AI Primitive
The dependency path must be version-specific.
6. AI Product Type (APT)
hasAIProductType has exactly one value for a qualified AI Product version.
Primary
The Product's intelligence is constituted without a material upstream AI Product dependency, directly from AI Primitives together with governed data, knowledge, rules, tools, and controls.
Derived
The Product's distinct intelligence proposition materially depends upon one or more upstream AI Products.
A Model Product dependency does not by itself make the Product Derived.
Derivation topology
For Derived AI Products, a separate topology characteristic may describe:
SingleUpstreamComposite
Composite is not a peer APT.
7. Intelligence Capability (IC)
hasIntelligenceCapability is multi-valued.
Canonical conceptual capabilities include:
- Perception;
- Classification;
- Detection;
- Prediction;
- Generation;
- Recommendation / Ranking;
- Reasoning;
- Planning;
- Optimization;
- Control.
The earlier umbrella CapabilityType enumeration is deprecated because it mixed Product Type, intelligence mechanism, agency, and capability.
8. Orthogonal Ontology Dimensions
The AI Product model keeps these characteristics separate.
Consumption Scope (CS)
Examples: Local, Cross-Division, Global/Federated, External.
Purpose (PU)
Examples: Decision Support, Operational Automation, Compliance & Risk, Client Value Creation, Revenue Enablement, Innovation, Research/Exploration.
Interaction / Execution Mode (MO)
Examples: Batch, Real-time/Event-driven, Synchronous API, Interactive, Embedded, Continuous, Agentic/Goal-driven.
Intelligence Effect (IE)
Examples: Score, Classification, Prediction, Alert, Recommendation, Generated Content, Explanation/Evidence, Decision, Plan, Digital Action, Physical Action.
Agency / Autonomy (AR)
Conceptual scale: Inform, Recommend, Assist, Act with Approval, Autonomous Action.
Embodiment (EM)
Examples: Non-embodied, Physical-world Perceptive, Embodied/Actuating, Closed-loop Autonomous.
Risk Tier (RT)
AIPS retains computed R0–R4 risk classification. Risk does not determine APT.
9. Behavioral Summary
The existing ontology-aligned BehavioralSummary remains a detailed governance representation for topology, authority, autonomy, and control semantics.
It may include properties such as:
| Property | Meaning |
|---|---|
topologyScope | agent/system topology classification |
authorityScope | delegated authority boundary |
autonomyLevel | detailed autonomy semantics |
controlMode | human/system control relationship |
The higher-level AR/MO/EM dimensions provide normalized Product ontology classifications; BehavioralSummary provides detailed machine-enforceable behavioural semantics.
10. Governance & Risk Ontology
AIPS governance is computed from orthogonal Product ontology and contextual impact evidence rather than a single primary archetype.
EffectiveRiskTier
EffectiveRiskTier is a computed, provenance-aware artifact.
Conceptually:
| Property | Meaning |
|---|---|
tier | R0–R4 |
declaredRiskTier | optional declared/input tier |
riskDrivers | one or more evidence-backed RiskDriverAssessment records |
computationTimestamp | computation time |
computationMethod | governance-engine method/version |
escalationRequired | mandatory true for R4 |
A RiskDriverAssessment records a material governance driver, its contributed tier, rationale, and optional evidence reference. Driver kinds include agency, effect, embodiment risk, topology, domain sensitivity, decision finality, action authority, impact scale, reversibility, qualification currentness and dependency topology.
The normative hard floors include:
AutonomousAction→ R2 or higher;AutonomousAction + PhysicalAction→ R3 or higher;ClosedLoopAutonomous + PhysicalAction→ R3 or higher;AutonomousAction + EnterpriseOrchestrated→ R4.
GovernanceObligationProfile
Each AI Product version has a computed GovernanceObligationProfile derived from the Effective Risk Tier and the Product characteristics that created it.
The profile records:
- derived risk tier;
- governance driver dimensions;
- mandatory governance obligations;
- computation method/time;
- supporting evidence references.
Machine-checkable v0.1 minimums include:
- R2+ →
ChangeQualityMonitoring; - R3+ →
InterventionOverride+ImmutableAudit; - R4 →
ExecutiveEscalation; - autonomous/approved action →
ActionAuthorization; - autonomous physical action →
SafetyCase+EmergencyStopSafeState.
This replaces the legacy single PrimaryArchetype governance input. PrimaryArchetype remains compatibility vocabulary only.
11. AIPROD
AIPROD is the immutable semantic descriptor for one Product version.
It defines or links:
- Product identity/version;
- purpose and intelligence proposition;
- AI Qualification Evidence;
- APT and derivation topology;
- Intelligence Capabilities;
- CS / PU / MO / IE / AR / EM / RT characteristics;
- input/output or action/effect contracts;
- lineage/provenance;
- governance and prohibited uses;
- quality/evaluation semantics;
- discoverability metadata.
AIPROD is marketplace- and registry-facing.
12. Machine-readable bindings
The aligned AIPS 0.1 bindings use the following core properties:
| Concept | Core property / construct | AIPROD property |
|---|---|---|
| AI Product version | aips:AIProductVersion | aips:AIPROD descriptor |
| Intelligence proposition | aips:intelligenceProposition | aips:intelligenceProposition |
| AI Qualification Evidence | aips:hasAIQualification | aips:hasAIQualification |
| AI Product Type | aips:hasAIProductType | aips:declaresAIProductType |
| Derivation topology | aips:hasDerivationTopology | aips:declaresDerivationTopology |
| Intelligence Capability | aips:hasIntelligenceCapability | aips:declaresIntelligenceCapability |
| Consumption Scope | aips:hasConsumptionScope | aips:declaresConsumptionScope |
| Purpose | aips:hasPurpose | aips:declaresPurpose |
| Execution Mode | aips:hasExecutionMode | aips:declaresExecutionMode |
| Intelligence Effect | aips:hasIntelligenceEffect | aips:declaresIntelligenceEffect |
| Agency / Autonomy | aips:hasAgencyAutonomy | aips:declaresAgencyAutonomy |
| Embodiment | aips:hasEmbodiment | aips:declaresEmbodiment |
| Effective risk | aips:hasEffectiveRiskTier | aips:exposesEffectiveRiskTier |
| Governance obligations | aips:hasGovernanceProfile | aips:exposesGovernanceProfile |
| Topology scope | aips:hasTopologyScope | aips:declaresTopologyScope |
PrimaryArchetype and the earlier umbrella capability vocabulary remain present only as deprecated 0.1 compatibility terms. They are no longer the normative AI Product ontology classification model.
The conformance examples under static/spec/aips/0.1/examples exercise Primary, Derived, Composite-Derived, symbolic AI, Model Product boundary, and qualification currentness/invalidation.
13. AIPDS
AIPDS describes execution/deployment realization.
It may include:
- deployment environments;
- runtime configuration;
- serving/execution modes;
- resource profiles;
- promotion gates;
- model training/retraining only where applicable;
- reasoning/planning/agent/physical execution details where applicable;
governanceProfileRefand operationalGovernanceControlBindingmappings.
AIPDS must not assume that every AI Product owns a trained model or retraining lifecycle.
14. DPP
The Digital Product Passport binds Product-version trust evidence.
The ontology-aligned DPP should eventually bind or reference:
- Product identity/version;
- AIPROD integrity;
- AI Qualification Evidence/currentness;
- APT;
- effective risk tier;
- issuer identity;
- evidence references;
- supersession/revocation state.
Model cards and training evidence become conditional modules where relevant rather than universal AI Product requirements.
15. Version Immutability Doctrine
For each published AI Product version:
- AIPROD is immutable;
- AI Qualification Evidence is immutable for that evaluated version/dependency state;
- APT is historically preserved;
- risk computation evidence is historically preserved;
- DPP issuance state is historically traceable.
A material change creates a new Product version/artifact with provenance linkage rather than rewriting history in place.
Qualification currentness can change as the ecosystem evolves, but historical evidence must remain available.
16. Downstream Classification Impact
When an upstream dependency changes in a way that may alter AI qualification or APT, AIPS/HDIP should support graph-based impact analysis.
A newly published upstream version does not mutate downstream versions still bound to an older valid dependency.
17. Separation of Concerns
The core Product model does not embed all deployment, commercial, marketplace, or cryptographic details.
Those belong to specialized artifacts:
- AIPROD — semantic descriptor;
- AIPDS — deployment/execution specification;
- DPP — trust/passport envelope;
- catalog/marketplace projections — discoverability and acquisition views.
18. Architectural Positioning
AIPS is intended to provide a computable, registry-driven framework for AI Product semantics, qualification, governance, deployment, and trust.
The ontology-aligned model is built on four separations:
- Productness ≠ AI Product qualification;
- Model Product ≠ AI Product;
- APT ≠ Intelligence Capability ≠ Agency ≠ Embodiment ≠ Risk;
- Logical Product identity ≠ immutable Product-version descriptor.
These separations allow AIPS to govern learned, symbolic, agentic, derived, and physical AI without forcing every AI Product into a model-centric lifecycle.