D336 — The doctor is called until the Observatory is empty; a new artifact is not "never used"

Decision. kb_status.conformance.doctor_suggested counts every lint finding the caller can see, not only the conformance checks, and conformance.lint_findings reports that number. artifact_unused does not report an artifact as never activated (or only catalogue-loaded) when the KB's git history shows it added, or renamed into place, inside usage_stale_days.

Why. A real KB sat at 18 info findings in the Observatory with doctor_suggested: false: none of them was a conformance check, so nothing ever called the kb-doctor session whose own definition of done is zero (D306, D332). Seventeen were artifact_unused on skills committed four days earlier against a 42-day threshold — a claim of disuse with no time to observe it, and the one kind of noise that would make a zero target unreachable.

Alternatives rejected. - Add the content checks to conformanceChecks: that set is the structural drift the standard defines (D290, D318); the trigger and the severity counts are different questions, so only the trigger widens. - Measure the grace from the usage store's first report: the store is local state an ephemeral data dir loses at every restart (D333), and a client's scan already covers its transcript window; the artifact's own age is what the check is missing.

Consequences. The doctor interval still decides when (D299): a KB is not nagged daily for one info finding. Every finding already has a way out (D313, D332), so a session can always reach zero. A KB that is not a git repository keeps the previous behaviour.