Lifecycle, Versioning & Qualification Currentness
An AI Product has a continuing logical Product identity and a sequence of immutable Product versions. Each published Product version carries its own AIPROD, AI Product qualification evidence, APT, governance classifications, and lineage.
This prevents a later implementation change from rewriting the historical ontology of earlier Product versions.
Lifecycle Stages
AI Products should declare a lifecycle stage such as:
- Experimental — prototype/research usage;
- Beta — controlled early adoption;
- Production — governed supported usage;
- Deprecated — still available but scheduled for retirement;
- Retired — no longer available for new consumption.
Lifecycle stage is independent from AI Product Family qualification.
Versioning Requirements
AIPS recommends semantic versioning (MAJOR.MINOR.PATCH) unless a governed organizational scheme is declared.
Version metadata should include:
- release date;
- lifecycle stage;
- compatibility notes;
- dependency changes;
- qualification status;
- APT changes, if any;
- deprecation/retirement information.
Published semantic descriptors are immutable. Material change creates a new Product version/AIPROD artifact rather than mutating historical evidence in place.
AI Qualification Is Version-specific
A logical Product may retain identity while later versions change their intrinsic constitution.
Example:
If v2 has no other qualifying AI Primitive path and no longer satisfies the AI Product definition, v2 must not inherit v1's AI Product Family classification.
The semantic business proposition may remain similar even while Product Family qualification changes.
Qualification Currentness
Conceptual states are:
- CURRENT
- STALE
- REVALIDATION_REQUIRED
- INVALID
Currentness should be recalculated when relevant Product constitution changes.
Typical triggers include:
- AI Primitive removal/replacement;
- upstream AI Product version migration;
- material dependency graph change;
- APT-changing changes;
- change to the intelligence proposition or value-bearing essence.
APT Evolution
APT may change across Product versions.
Examples:
v1: Primary → direct model primitive
v2: Derived → material upstream AI Product dependency
or:
v1: Derived → upstream AI Product
v2: Primary → upstream AI Product removed; direct primitive introduced
APT change must be explicitly represented in the new AIPROD and supported by lineage evidence.
AIPS does not currently mandate that every APT change maps mechanically to a particular SemVer level; the Product Owner and governance rules must assess whether the change is materially breaking for consumers and semantics.
Downstream Impact
Versioned dependency binding prevents a newly published upstream version from silently changing downstream ontology.
A mature registry/knowledge graph should support:
- change event detection;
- affected dependency traversal;
- AI qualification re-evaluation for adopting downstream versions;
- APT and governance impact analysis;
- owner notification and controlled migration.
Retirement Obligations
When a Product version is retired:
- consumers are notified through catalog and contract channels;
- AIPROD, qualification evidence, and lineage remain available for audit;
- successor Product/version references are recorded where relevant;
- downstream dependencies are identified before final retirement.
Summary
- Logical Product identity and Product-version ontology are distinct.
- Historical Product versions retain immutable qualification and APT evidence.
- New versions must revalidate AI Product Family membership when constitution changes materially.
- Currentness enables automated downstream impact analysis.
Principle: The Product may evolve; the historical truth of each Product version must not.
Lifecycle closure across artifacts
A Product version is not considered lifecycle-consistent merely because its AIPROD is immutable. AQE, AIPROD, AIPDS, AIPS-DPP, Marketplace, COVP, PMDD and PDEP evidence must continue to describe the same Product reality. See Lifecycle Closure & Cross-Artifact Consistency.
A new version may change AI Product Family qualification or APT without rewriting the historical semantics of earlier versions. Downstream impact follows actual material/version bindings.