Versioning & Lifecycle
AIPDS is a version-specific operational realization of an AI Product version. Runtime changes must therefore be traceable to the Product version, AIPROD semantics, AI qualification evidence, and deployment configuration they realize.
Distinct Versioned Objects
AIPS distinguishes:
Product Identity
↓
AI Product Version
├── AIPROD versioned semantic descriptor
└── AIPDS versioned deployment descriptor
A runtime deployment instance is not the same thing as the logical Product version or its immutable descriptors.
Versioning Triggers
A new Product/AIPROD/AIPDS version may be required by material changes to:
- Product semantics or intelligence proposition;
- AI Product Type or derivation topology;
- qualifying AI Primitive;
- upstream AI Product dependency;
- model/rule/knowledge constitution;
- input/effect contracts;
- agency/autonomy or embodiment;
- execution mode;
- risk posture;
- safety or governance boundary.
Routine runtime scaling need not create a new Product version when semantics and contracts remain unchanged.
Learning and Evolution Versions
Where retraining occurs, the deployment must link to the resulting model/version and training evidence.
Where the Product does not train, equivalent provenance applies to:
- rule versions;
- knowledge/ontology versions;
- planner configuration;
- prompts/policies/guardrails;
- upstream AI Product versions;
- executable runtime versions.
AI Qualification Currentness
A deployment must not assume that historical AI Product classification remains valid forever.
If a new Product version removes or invalidates the only qualifying AI Primitive path, the new version may cease to qualify as an AI Product. Historical versions retain their historical classification.
AIPDS must therefore preserve explicit bindings rather than silently following ontology-significant dependency changes.
Lifecycle States
Operational lifecycle states may include:
- Experimental
- Development
- Production
- Deprecated
- Retired
Production deployment additionally requires current governance/qualification evidence appropriate to the applicable AIPS rules.
Principle: Product identity may continue across versions while AI qualification, APT, constitution, and deployment realization change. Versioning must preserve that history rather than mutating it away.