Claims & engineering profiles

Two layers of datasheet-backed evidence sit behind Catalog flags and Intelligence bodies.

	L1 — Datasheet claims	L2 — Engineering profile
What	Atomic assertions extracted from the PDF (limits, ratings, features) with page-level source	Structured engineering summary synthesized for agents (pins, electrical sections, …)
Shape	`claims[]` with `kind`, `text`, `source.page`, `claim_id`	Sectioned object (`pin_model`, `electrical_profile`, …)
Typical use	Cite a specific limit or feature in UI / RAG	Feed resolve / alternatives / BOM agents without re-parsing the PDF
Catalog	`has_claims` + `evidence_links.claims`	`has_engineering_profile` + `evidence_links.engineering_profile`
Intelligence	`GET …/claims` or `POST /v1/parts/claims`	`GET …/engineering-profile`

Mental model

Datasheet PDF
    → L1 claims     (quoteable facts + page)
    → L2 profile    (normalized sections for tools)

Catalog answers “does evidence exist?” (booleans + deep links).
Intelligence returns the bodies. Without Intelligence → 403 on claim/profile endpoints.
Missing / still-running L2 synthesis can return HTTP 200 with synthesis_status: "missing"|"pending".

Filter with kind_prefix (example: limit.). Treat kind as a stable taxonomy string, not free-form UI copy.