AIPS 0.1 Conformance Examples
The ontology-realigned AIPS 0.1 machine bindings include executable examples under static/spec/aips/0.1/examples.
| Case | Expected result | What it proves |
|---|---|---|
| Primary fraud detection | Valid AI Product Version | Direct learned-model primitive can support APT = Primary |
| Derived transaction intervention | Valid AI Product Version | Material upstream AI Product dependency yields APT = Derived |
| Composite transaction risk | Valid AI Product Version | Multiple upstream AI Products remain APT = Derived, Topology = Composite |
| Symbolic policy reasoner | Valid AI Product Version | ML model dependency is not universally required |
| Qualification revalidation | Valid structural state | Qualification currentness is version-specific and can require revalidation |
| Autonomous physical navigation | Valid AI Product Version | Autonomous physical action is R3+ and carries action/safety governance obligations |
| Autonomous physical R2 | Expected invalid | Physical autonomous action cannot be under-tiered or omit safety/action obligations |
| Enterprise autonomous R3 | Expected invalid | Enterprise-orchestrated autonomous action requires R4 + escalation obligations |
| Model Product boundary | Expected invalid as AI Product Version | Productized model does not automatically qualify as an AI Product |
| Primitive-removal invalidation | Valid AQE record | INVALID qualification can record loss of the last qualifying primitive path |
| AIPROD examples | Valid | Semantic descriptor carries AQE, APT and orthogonal ontology dimensions |
| Publish projection | Valid | Marketplace surface exposes ontology-aligned classifications |
| Learned-model AIPDS | Valid | Retraining remains supported where it genuinely applies |
| Symbolic AIPDS | Valid | learningMode = NONE and no training specification is valid |
| Physical-navigation AIPDS | Valid | Embedded + Continuous Physical AI carries safety and governance-control bindings |
| Missing-evolution AIPDS | Expected invalid | Every deployment must explicitly state its learning/evolution posture |
| PDEP Primary fraud Stage 10 | Valid lifecycle record | Promotion binds current AQE/APT, governance, AIPROD, AIPDS and DPP |
| PDEP Composite Derived Stage 11 | Valid lifecycle/currentness state | Continuous Assurance may require revalidation without rewriting historical Product evidence |
| PDEP missing definitive qualification | Expected invalid | Stage 10 cannot bypass Stage 7 definitive AQE/APT compilation |
| PDEP stale promotion | Expected invalid | Promotion requires current qualification, dependencies, governance and assurance |
| Lifecycle current Primary closure | Valid lifecycle consistency record | AQE/AIPROD/AIPDS/DPP/Marketplace/COVP/PMDD/PDEP can close as one coherent Product version |
| Primitive-removal reclassification | Valid conditional lifecycle record | New non-AI version does not rewrite historical AI Product classification; only adopting downstream versions revalidate |
| Lifecycle failed-check closure | Expected invalid | CONSISTENT cannot coexist with failed cross-artifact checks |
| Cross-artifact DPP version drift | Expected inconsistent bundle | An individually valid DPP can still contradict the Product version used by the rest of the lifecycle |
| Cross-artifact Marketplace APT drift | Expected inconsistent bundle | Marketplace cannot invent a different APT from the compiled/AIPROD/DPP ontology |
Validation helper
Run from the repository root:
python3 scripts/validate-aips-0.1.py
The helper validates Core, AIPROD, AIPDS, publish, AIPS-DPP, PDEP Spec C, Marketplace, COVP, PMDD and Lifecycle Closure JSON Schemas; parses their JSON-LD and Turtle artifacts; and parses all SHACL shapes. If pyshacl is installed, it executes the combined semantic constraints against positive RDF examples.
The negative cases are intentional ontology/deployment boundary tests and therefore count as passing when they fail their target JSON Schema as expected.