AIPS 0.1 Migration & Deprecation Guide
This guide maps pre-realignment AIPS concepts to the ontology-aligned AIPS 0.1 model.
Migration map
| Earlier concept / assumption | AIPS 0.1 alignment |
|---|---|
| Packaged AI artifact automatically becomes an AI Product | Productness and AI Product qualification are separate; PIN + AQE determine AI Product Family membership |
| Raw model checkpoint can become an AI Product by adding metadata/interfaces | A productized learned model is ordinarily a Model Product; it becomes an AI Product only if the Product's own value-bearing essence satisfies AI Product qualification |
PrimaryArchetype = INFERENCE / GENERATIVE / AGENTIC | Deprecated as an umbrella classifier; use APT + IC + MO + IE + AR + EM + topology + risk |
| Capability Type mixes Model, Agent, Generative, Symbolic, AGI/ASI | Replace with orthogonal dimensions; APT is Primary / Derived, while IC describes intelligence capability |
| Universal Learning & Inference Contract (LIC) | IBEC — Intelligence Behavior & Evolution Contract is the general concept; LIC remains compatibility terminology for learning/inference-specific cases |
| Every AI Product retrains / continuously learns | learningMode = NONE is valid; evolution posture is mandatory, retraining is conditional |
| Inference serving is universal | Execution/interaction is universal; inference serving is one execution specialization |
| Model drift is the universal monitoring model | Use mechanism-appropriate change/quality monitoring plus qualification, dependency, governance and assurance currentness |
| Model Card/training evidence required for every AIPS-DPP | Product trust evidence is universal; Model Card/training evidence is conditional on constitution/applicability |
| AI Product vs Data Product distinguished by dynamic vs static behavior | Distinguish by Product Intrinsic Nature / value-bearing essence |
L0 = AI Asset in PMDD | L0 = Emerging AI Product; candidates without CURRENT AQE are Not Assessable |
| Composite is a peer AI Product Type | Composite is a Derived topology, not APT03 |
| Orchestration of multiple AI Products automatically creates another AI Product | A new distinct governed intelligence proposition is required for a Derived AI Product |
| Marketplace label establishes Product family | Marketplace projects verified semantics; it cannot create Product Family membership |
| AI Product classification is permanent | Qualification is version-specific and can become stale/invalid; historical versions retain their historical semantics |
| 13-stage interactive Spec C | Normative PDEP Spec C is 12 stages: Stage 0–11 |
Compatibility policy
Deprecated vocabulary MAY remain in serialized or documented compatibility surfaces during the AIPS 0.1 transition, but new implementations SHOULD emit the ontology-aligned constructs.
Where both old and new representations exist, the ontology-aligned construct is authoritative.
PrimaryArchetype
- Status: deprecated compatibility vocabulary.
- Do not use it to compute APT or governance obligations.
- Replace with IC/AR/EM/IE/topology/risk-driver evidence as appropriate.
LIC
- Status: specialized compatibility terminology.
- Use IBEC as the general lifecycle contract.
- LIC may describe the learning/inference subset where that terminology remains useful.
Historical filenames and routes
Some document/component filenames retain historical names to preserve stable routes, even where their semantic title has changed. For example, the AIPDS pages historically named around inference/retraining now document broader execution/evolution semantics. Route compatibility does not imply legacy semantics remain normative.
Migration sequence for an existing implementation
- Establish explicit Product identity and Product-version boundaries.
- Add version-specific AI Qualification Evidence and AI Primitive evidence.
- Resolve APT (
Primary/Derived) independently of IC/MO/IE/AR/EM/risk. - Replace archetype-driven risk logic with Risk Driver Assessment + Governance Obligation Profile.
- Generalize deployment contracts from mandatory inference/retraining to Execution + Evolution specifications.
- Reissue AIPS-DPP as an AI Product Trust Passport with conditional model evidence.
- Align Marketplace listing kind/currentness and consumption-port projection.
- Add COVP evidence without allowing usage/value to rewrite ontology.
- Migrate PMDD to maturity-only semantics and applicability-aware scoring.
- Enforce lifecycle closure and cross-artifact consistency before promotion/publication.