Skip to content

Schema Layers A–E

A UKS packet is organized into five layers. You use only the layers your conformance level requires — L1 packets use just Layer A; L3 packets use all five.

Layer A — Data Contracts & Sources

The foundation, required at every level. Layer A answers what the data is.

  • sources[] — each source carries id, title, source_type, credibility_score, evidence_grade, clinical_status, plus optional doi, authors, license_label + rights_url, summary, and key findings.
  • data_contracts[] — the shape and semantics of fields the packet describes.

Additive, optional siblings extend a source without breaking older consumers: claims[] (claim-level grade/status), temporal_validity, and quality_dimensions.

Layer B — Scrape Targets

Where to acquire more data. URLs, APIs, authentication, and pagination — everything needed to fetch the next batch of source material reproducibly.

Layer C — Extraction Rules

How to parse and normalize what Layer B fetches — CSS/JSON selectors, regex, or LLM extraction prompts that map raw content into the data contracts.

Layer D — Directives

Under what rules the data is processed — validation, routing, enrichment, and governance logic. Directives are a closed, executable vocabulary (directive_type), because an unrecognized rule is one no runtime can apply.

Layer E — Actions & Agent Instructions

What an AI agent should do with the knowledge — typed actions, triggers, and step definitions. Layer E is what makes a packet executable rather than merely descriptive.

The knowledge graph

Across layers, entities and typed relationships form a knowledge graph. In the Registry, these relationships become first-class edges that power graph-aware retrieval, consensus detection, and authority scoring — see Trust & safety.

Normative source

This is an overview. The binding definitions live in the normative specification:

→ Next: Conformance Levels · Governance

Released under the MIT License.