# AIPS 0.1 Ontology Realignment Examples

These examples exercise the machine-readable AI Product ontology and the mechanism-neutral AIPDS deployment model.

The suite intentionally contains both positive and negative boundaries.

## Core / AIPROD coverage

- Primary AI Product with a learned-model primitive;
- Derived AI Product with a single upstream AI Product;
- Composite Derived AI Product;
- symbolic AI Product with no ML model dependency;
- qualification currentness and revalidation;
- Model Product boundary;
- qualification invalidation after the last primitive path disappears;
- AIPROD authoring examples;
- marketplace publish projection.

## AIPDS coverage

- learned-model Product with triggered retraining;
- symbolic reasoner with `learningMode = NONE` and no training specification;
- embedded/continuous Physical AI deployment with safety binding;
- expected-negative deployment that omits the mandatory evolution declaration.

The symbolic AIPDS case is an important boundary test: **a valid AI Product deployment does not universally require model training, retraining, or a model-serving interface.**

Run:

```bash
python3 scripts/validate-aips-0.1.py
```

JSON Schema provides structural validation. RDF/Turtle and JSON-LD parsing validates serialization integrity. SHACL remains the semantic constraint language; when `pyshacl` is available the helper also executes Core, AIPROD, and AIPDS SHACL shapes.


## Stage 5 governance / risk conformance

Stage 5 adds governance-aware examples:

- `core/autonomous-physical-navigation.jsonld` — valid R3 autonomous Physical AI with safety/action obligations;
- `core/invalid-autonomous-physical-r2.jsonld` — expected invalid under-tiered autonomous Physical AI;
- `core/invalid-enterprise-autonomous-r3.jsonld` — expected invalid enterprise-orchestrated autonomous Product that is not R4;
- AIPDS examples now bind a `governanceProfileRef` and one or more `governanceControlBindings`.

These cases prove that risk and governance are derived from ontology/context evidence rather than a legacy single archetype.
