Skip to main content

Ontology & Lifecycle Worked Examples

This page provides a single cross-reference for the canonical examples used throughout AIPS 0.1. The examples are intentionally diverse so that conformance does not collapse into “trained model behind an API”.

1. Primary learned-model AI Product — Fraud Detection​

Product Family: AI Product
APT: Primary
AI Primitive: learned fraud model
IC: Detection + Classification + Prediction
AR: Recommend
EM: Non-embodied

The Product consumes governed Data Products and a Model Product/primitive, but it has no material upstream AI Product dependency. It is therefore Primary, not Derived.

Machine-readable examples:

  • Core: static/spec/aips/0.1/examples/core/primary-fraud-detection.jsonld
  • AIPROD: static/spec/aips/0.1/examples/aiprod/primary-fraud-detection.aiprod.jsonld
  • AIPDS: static/spec/aips/0.1/examples/aipds/learned-fraud-detection.aipds.jsonld
  • DPP: static/spec/aips/dpp/0.1/examples/fraud-detection-lite.example.jsonld

2. Primary symbolic AI Product — Policy Reasoner​

Product Family: AI Product
APT: Primary
AI Primitive: symbolic reasoner
Learning Mode: NONE

This example proves that trained ML models and retraining are not universal requirements. Its evolution occurs through governed rule/knowledge changes and reasoner versioning.

Machine-readable examples:

  • Core: static/spec/aips/0.1/examples/core/symbolic-policy-reasoner.jsonld
  • AIPDS: static/spec/aips/0.1/examples/aipds/symbolic-policy-reasoner.aipds.jsonld
  • DPP: static/spec/aips/dpp/0.1/examples/symbolic-policy-reasoner.example.jsonld

3. Derived / Composite AI Product — Integrated Transaction Risk​

Fraud, sanctions and behavioural-risk AI Products are materially composed into a new governed intelligence proposition.

APT: Derived
Derivation Topology: Composite

The Product is not Derived merely because it calls multiple services; it is Derived because upstream AI Product intelligence is material to a new Productized intelligence proposition.

Machine-readable examples:

  • Core: static/spec/aips/0.1/examples/core/composite-derived-transaction-risk.jsonld
  • AIPROD: static/spec/aips/0.1/examples/aiprod/composite-derived-transaction-risk.aiprod.jsonld
  • DPP: static/spec/aips/dpp/0.1/examples/composite-derived-transaction-risk.example.jsonld

4. Physical AI — Autonomous Navigation Intelligence​

EM: Closed-loop Autonomous
IE: Physical Action
AR: Autonomous Action

This is an AI Product because the value-bearing essence is governed autonomous-navigation intelligence. A complete autonomous vehicle or robot may instead be a wider Compound Product in which Physical/Engineered and AI family contributions are co-essential.

Machine-readable examples:

  • Core: static/spec/aips/0.1/examples/core/autonomous-physical-navigation.jsonld
  • AIPDS: static/spec/aips/0.1/examples/aipds/physical-navigation.aipds.jsonld
  • DPP: static/spec/aips/dpp/0.1/examples/physical-navigation.example.jsonld

5. Boundary case — Downloadable Model Product​

A governed, versioned, downloadable Qwen-like model can be:

  • a Model Product;
  • its vendor's contextual Business Product;
  • a qualifying AI Primitive used by downstream AI Products.

It does not become an AIPS AI Product solely because it is productized.

Expected-negative Core boundary example:

static/spec/aips/0.1/examples/boundary/model-product-not-ai-product.jsonld

6. AI-enabled Service Product​

A streaming service can use recommendation AI extensively while remaining intrinsically a Service Product if ongoing service/access is its value-bearing essence. The recommendation capability may itself be a separately governed AI Product.

This illustrates the rule:

Uses AI ≠ AI Product.

7. Reclassification after AI Primitive removal​

Suppose Product A v1 ultimately depends on a Qwen Model Product, but A v2 replaces its only qualifying AI Primitive path with ordinary deterministic if/else logic.

  • A v1 remains historically an AI Product.
  • A v2 must be requalified and may leave the AI Product Family.
  • Downstream Products pinned to A v1 remain unchanged.
  • Downstream versions adopting A v2 must recompute their own transitive qualification.

Machine-readable AQE example:

static/spec/aips/0.1/examples/qualification/invalid-after-primitive-removal.jsonld

The lifecycle closure profile tests this behavior across AIPROD, AIPDS, DPP, Marketplace, COVP and PMDD projections.

8. Maturity is not qualification​

PMDD assesses maturity only after CURRENT AQE exists.

  • low maturity → Emerging AI Product;
  • high maturity → High-Maturity AI Product;
  • missing/invalid AI qualification → Not Assessable, not “AI Asset”.

This prevents maturity scoring from changing ontology.


Cross-example principle: Productness, AI Product qualification, APT, intelligence behavior, risk, trust and maturity are distinct claims and must remain independently evidenced.