Skip to content

Conformance Levels

UKS defines three conformance levels. Each adds layers — you implement only what you need, and a consumer can tell at a glance how much a packet promises.

LevelLayersContainsUse when
L1ASources + data contractsYou're packaging knowledge — citations, evidence, licenses
L2A + B + C+ scrape targets + extraction rulesYou're packaging how to acquire and parse more data
L3A + B + C + D + E+ directives + actions + agent instructionsYou're packaging executable agent behavior

L1 — knowledge

The most common level. A valid L1 packet has at least one source, each with the required fields:

id, title, source_type, credibility_score, evidence_grade, clinical_status

L1 packets interoperate with BibTeX, FHIR, and schema.org today. Most published registry packets are L1.

L2 — acquisition

Adds Layer B (scrape targets) — where to get more data — and Layer C (extraction rules) — how to parse it (selectors, regex, or extraction prompts). An L2 packet is a reproducible recipe for gathering a class of knowledge.

L3 — action

Adds Layer D (directives) — processing, validation, and governance rules — and Layer E (actions & agent instructions) — what an AI agent should do with the knowledge. An L3 packet is executable: it can drive an agent end-to-end.

Checking a level

The CLI reports the level of any packet:

bash
uks info my-packet.uks.json
# Level: L1   Layer A: ✓  Layer B: ✗  Layer C: ✗  Layer D: ✗  Layer E: ✗

The level is also validated: if a packet declares conformance_level: "L3" but omits Layer E, validation fails.

→ Next: Evidence Grading · The Spec

Released under the MIT License.