Step 2 · Taxonomies

Sort KYE™ names into parent / child trees.

Step 1 names the things. Step 2 groups them. You get 16 signed taxonomies so an auditor, regulator, or developer can talk about a whole class without listing every member.

list_alt 16 trees help Why bother compare_arrows Vs ontology build How you use it arrow_forward Next step
1 · What ships

Sixteen v1.0 canonical trees.

Each tree is a published JSON file with a stable $id URL. Each value carries a SHA-256 hash. You can cite a class, not a list.

  • Entity-class taxonomy
  • Capability-family taxonomy
  • Side-effect-level taxonomy
  • Data-class taxonomy
  • Decision-code taxonomy
  • Lifecycle-state taxonomy
  • Jurisdiction taxonomy
  • Sector-namespace taxonomy
  • Profile-family taxonomy
  • Certification-tier taxonomy
  • Risk-tier taxonomy
  • Evidence-kind taxonomy
  • Signal-family taxonomy
  • Action-family taxonomy
  • Redaction-class taxonomy
  • Oversight-role taxonomy
2 · Why bother

Cite a class, not a list.

A rule that says "deny move_money for tier_3 agents" stays stable as new tools ship. The taxonomy resolves the class for you, so policy authors do not chase every new member.

  • Auditors map side_effect_level values to ISO 42001 controls without re-listing each tool.
  • Regulators ask for data_class = pii coverage under GDPR Art. 30, and the taxonomy enumerates the leaves.
  • Developers filter the Signal Bus by signal_family instead of by event name.
  • Operators set policy on risk_tier bands aligned to NIST AI RMF tiers 1–4.
3 · Taxonomy vs ontology

Hierarchy first. Then meaning.

A taxonomy organises terms. An ontology defines how terms relate. KYE™ needs both — taxonomy first, then ontology — so rules stay short and meaning stays explicit.

  • Taxonomy says: card_purchase is a kind of payment_initiation.
  • Ontology says: card_purchase needs an authority_grant bound by an amount_limit.
  • You query the tree to find all kinds of payment. You query the ontology to find what each one needs.
4 · How you wire it in

Three patterns. Two minutes each.

Fetch the JSON. Bind a leaf to a payload field. Validate in CI. The same trees feed your OpenAPI schemas, your webhook filters, and your audit reports.

  • Bind a payload field. Constrain side_effect_level in JSON Schema 2020-12 to taxonomy leaves only.
  • Filter a webhook. Subscribe to events where signal_family ∈ {risk, drift, evidence}.
  • Report on a class. Group Evidence Packs by action_family for an EU AI Act Annex IV log.
Where to go next

Continue the stack →

Step 3 adds the meaning layer. Step 4 turns the named classes into wire-level JSON contracts.

menu_book ← Step 1 · Dictionaries hub Step 3 · Ontology → schema Step 4 · Schemas

Ready to see your AI agents flagged?

Start in shadow mode. We’ll deliver your first Evidence Pack™ in 4–8 weeks.

rocket_launch Apply for pilot → science Try the sandbox →