AIPS DPP — Conformance Levels
1. Conformance model
An AIPS DPP is conformant only when it satisfies both:
- BPS DPP Core conformance — universal passport structure, identity and integrity.
- AIPS DPP conformance — AI Product qualification and trust evidence.
AIPS DPP deliberately separates universal trust evidence from mechanism-specific evidence.
2. Lite and Full
| Class | Purpose | Required posture |
|---|---|---|
| AIPS-DPP-Lite | marketplace discovery and pre-onboarding | mandatory trust summaries inline |
| AIPS-DPP-Full | audit, regulatory review and deep onboarding | same inline summaries plus ByRef evidence as required |
3. Mandatory sections
| Section | Lite | Full | Meaning |
|---|---|---|---|
| AI Qualification | Required | Required | proves current AI Product Family qualification |
| AI Ontology Summary | Required | Required | APT and orthogonal classification posture |
| Effective Risk Tier | Required | Required | R0-R4 plus risk-driver evidence |
| Governance Obligation Profile | Required | Required | current control obligations |
| Trust Currentness | Required | Required | qualification/dependency/governance/assurance currentness |
| AI Product Assurance | Required | Required | product-level assurance summary/evidence |
4. Conditional evidence
| Section | Rule |
|---|---|
| Model Card | required when qualification evidence contains a LEARNED_MODEL primitive |
| Training Data | conditional on training/learning provenance being applicable |
| Training Pipeline | conditional on owned/material training or retraining activity |
| Risk Register | policy-dependent human-readable evidence |
| Detailed assurance | Full view SHOULD expose ByRef evidence where available |
A Product using a SYMBOLIC_REASONER primitive is not invalid merely because it has no model or training artifacts.
5. Semantic consistency rules
APT=DerivedMUST declare derivation topology.APT=PrimaryMUST NOT declare derivation topology.- a
Validpassport MUST NOT carryAI Qualification = INVALID. - R4 MUST set
escalationRequired=true. - the Governance Obligation Profile MUST be derived from the same effective risk tier exposed by the passport.
- a learned-model qualification path MUST have Model Card evidence.
6. Validation stack
- JSON Schema — structural authoring/conformance.
- JSON-LD/RDF parsing — semantic binding.
- BPS SHACL — universal passport rules.
- AIPS SHACL — AI-specific semantic constraints.
- hash/signature verification — integrity and issuer authenticity.
- evidence dereferencing — Full-profile verification where authorized.
7. Qualification currentness
AI Product Family membership is version-specific and evidence-backed. If an upstream change invalidates the only qualifying AI Primitive path, the passport's qualification evidence becomes stale/revalidation-required/invalid as appropriate. A historical passport remains an immutable record of the state at issuance.