5.1.0: Traceable Evidence

Included changes

01Complete extracted documents are supplied to AI analysis up to 500,000 characters each. Larger inputs stop explicitly instead of being silently truncated.
02New analyses retain model, prompt, request and response fingerprints, source snapshot references and reproducible request recipes.
03Capture metadata is frozen. Repeated content remains a separate dated occurrence while identical text can share a content artifact.
04A missing or incomplete source section is rejected rather than replaced by the full page. Live-check freshness retains the existing 30-hour threshold.
05Evidence packets distinguish recorded, hash-verified analysis lineage from missing historical provenance.

01Full input coverage is not proof of complete semantic analysis or legal correctness. Existing evidence, review and consecutive-observation gates remain active.
02Historical analyses are not retroactively certified. Raw AI responses remain in private provenance; canonical dual-write follows its existing activation gate.
03No new subscriber notifications, mobile release or automatic provider action is introduced.