Skip to main content

Taxonomy

This taxonomy defines the classification system that underpins the
Base Product Specification (BPS), AI Product Deployment Specification (AIPDS),
and AI Product Specification (AIPROD).

It provides common categories for describing products in terms of
capabilities, lifecycle, risk, governance, and ecosystem alignment.

1. Capability Taxonomy​

AI Products must declare their capability type(s).
This classification ensures discoverability, comparability, and composability.

  • Predictive → Forecasting, scoring, classification.
  • Descriptive → Clustering, summarization, analysis.
  • Prescriptive → Recommending actions or interventions.
  • Generative → Producing novel outputs (text, images, code, audio, 3D).
  • Agentic → Acting autonomously or semi-autonomously to achieve goals.
  • Symbolic / Reasoning → Logic, rules, knowledge graphs, neuro-symbolic systems.
  • Hybrid → Explicit combination of two or more capability types.

2. Lifecycle Taxonomy​

Products progress through stages of maturity, each with distinct governance requirements.

  • Experimental → Research or prototype stage.
  • Beta → Limited release, partial SLAs.
  • Production → Fully supported, with declared SLAs.
  • Deprecated → Maintained but not recommended for new use.
  • Retired → Withdrawn, metadata persists for audit.

3. Risk Taxonomy​

Risk levels align with emerging regulatory frameworks (e.g., EU AI Act).

  • Minimal Risk → Trivial applications (e.g., spam filters).
  • Limited Risk → Low-stakes applications, transparency obligations apply.
  • High Risk → Significant implications for health, safety, rights, or finance.
  • Unacceptable Risk → Prohibited applications (e.g., social scoring, lethal autonomous weapons).

4. Governance Taxonomy​

Products must align with governance classifications to ensure accountability.

  • Compliance Scope
    • GDPR, HIPAA, EU AI Act, ISO/IEC 42001, NIST AI RMF.
  • Policy Obligations
    • Data residency, privacy, fairness, safety, bias monitoring.
  • Enforcement Mechanisms
    • Automated (technical controls), manual (audits, approvals).

5. Discovery & Cataloging Taxonomy​

AI Products must be discoverable via trustworthy mechanisms.

  • Registration Type
    • Internal catalog, cross-organization registry, public marketplace.
  • Identifier Type
    • URN, DOI, UUID, URI.
  • Discovery Methods
    • Keyword search, semantic/embedding search, AI-assisted recommendation.

6. Economic Taxonomy​

AI Products must declare their economic models for sustainable self-service.

  • Pricing Models

    • Fixed (subscription, license).
    • Usage-based (per API call, per token, per inference).
    • Hybrid (subscription + usage).
    • Freemium (tiered access).
    • Tokenized (credits, blockchain-based).

  • Resource Metrics

    • Compute (CPU/GPU hours).
    • Storage (GB/TB).
    • Networking (bandwidth).
    • Energy efficiency / carbon impact.

7. Observability & Transparency Taxonomy​

Products must expose signals for audit and trust.

  • Input Traces → Logged or sampled inputs.
  • Intermediate Representations → Embeddings, feature maps, symbolic states.
  • Decision Rationale → Feature importance, saliency maps, reasoning paths.
  • Uncertainty Estimates → Confidence scores, variance, entropy.
  • Bias Indicators → Subgroup error rates, fairness metrics.

8. Extension & Composability Taxonomy​

Products must declare how they extend or compose with others.

  • Extension Types

    • Schema evolution (backward-compatible or breaking).
    • Plug-ins / adapters.
    • Custom metadata slots.

  • Composition Patterns

    • Pipelining → Sequential workflows.
    • Ensembling → Multiple models combined.
    • Chaining → Multi-step reasoning.
    • Agentic orchestration → Multi-agent collaboration.

9. Provenance & Lineage Taxonomy​

Traceability ensures accountability.

  • Provenance

    • Source models.
    • Training data origins.
    • Contributors.

  • Lineage

    • Dependencies (frameworks, libraries, upstream products).
    • Version history.
    • Downstream consumers (where declared).

10. Product Categorization Anchors​

Across BPS and AIPS, every product can be classified along three universal axes:

  1. What it is → Capability type, identity, lifecycle stage.
  2. How it is governed → Risk, policies, prohibited uses, observability.
  3. How it is consumed → Interfaces, discoverability, economics, composability.

Principle: A taxonomy is not merely classification — it is the backbone of discoverability, governance, and interoperability.