Skip to main content

Lifecycle Closure & Cross-Artifact Consistency

AIPS lifecycle closure proves that the governed artifacts describing an AI Product version still refer to the same Product reality.

A lifecycle is not closed merely because AIPROD, AIPDS, AIPS-DPP, Marketplace, COVP and PMDD artifacts all exist. They must agree on the Product version, AI qualification, APT, semantic characteristics, risk/governance posture and trust currentness that they project.

Closure principle: no artifact may silently redefine ontology that belongs to another authoritative artifact.

Authoritative flow​

The arrows do not imply that every downstream artifact owns the upstream semantics. They indicate dependency and projection.

What must remain consistent​

For a Product version, AIPS expects consistency across at least:

  • identity and version — every artifact must resolve to the intended Product version;
  • AI qualification — qualification status and evidence path must not be contradicted downstream;
  • APT and derivation topology — AIPROD, DPP and Marketplace must project the compiled classification;
  • IC / CS / PU / MO / IE / AR / EM / topology — projections may omit non-required detail but must not contradict authoritative semantics;
  • risk and governance — DPP and deployment controls must represent the compiled Governance Obligation Profile;
  • deployment binding — AIPDS must deploy the Product version represented by AIPROD/AQE;
  • trust currentness — Marketplace eligibility must agree with DPP/PDEP currentness;
  • consumer evidence — COVP must bind to the exact Product version and marketplace projection it observes;
  • maturity — PMDD assesses the already-qualified Product version and must not substitute a different AQE/governance posture.

History is immutable; currentness is recomputed​

A new Product version never rewrites the historical ontology of an older version. If Product A v1 qualified as an AI Product because it depended on a Qwen Model Product, and A v2 replaces that path with ordinary deterministic if/else logic, A v1 remains historically an AI Product.

A v2 must be assessed from its own constitution. Downstream Product versions are affected only where their actual version bindings or material dependency paths are affected.

See: