More candidates than the policy cap, or a candidate older than the policy age, is an advisory doctor finding that names them; both numbers are declared once in the policy

What it means

The review queue is measured. doctor prints WARN knowledge and exits 0 when more records under candidates/ carry status: candidate than knowledge.candidates_max_files, when a candidate has waited longer than knowledge.candidate_max_age_minutes, or when a candidate is present on disk and not tracked by version control. Each finding names the paths. Both numbers are declared once in .ai/repo/policy.yaml and read by every surface, and a policy that declares neither is reported at the rule's own class, naming the missing key.

How it works

mj_validate_knowledge_accumulation in lib/knowledge.sh reads both keys with mj_pol_req, counts the files whose status is candidate (a superseded record stays in the directory and counts toward neither), and measures each candidate's review age from the oldest knowledge.derived line whose paths names the file, falling back to the commit that added the file and only then to the record's date; the finding says which source it used. Untracked files are read from git status --porcelain over the directory, and the remedy git add .ai/repo/knowledge/candidates && majordomus derive follows fix:. The rule majordomus.candidates-reviewed is advisory and dispatched from doctor only, because under watch an advisory finding is drift and exits 11, and a full review queue must not turn watch red in a hook.

How to see it

majordomus doctor | grep -E 'knowledge'                 # OK knowledge ... or WARN knowledge candidates ... over cap
majordomus knowledge candidates                         # the queue the cap is measured over
git status --porcelain .ai/repo/knowledge/candidates    # the untracked ones a WARN names
git add .ai/repo/knowledge/candidates && majordomus derive
majordomus watch; echo "exit $?"                        # unaffected by this rule

What it does not cover

It empties nothing. Whether a candidate deserves promotion is a person's judgement, and a machine that emptied the queue would be making it. It does not measure curated/, which is authored content, and it does not count superseded records, which are the rejections a person already made.

Why it exists

The deriver writes a candidate at every episode boundary that produced evidence, and nothing removes one except a person. Left alone, the directory grows, and project.accumulation-is-measured says that anything which grows on its own is measured and reported before it is a problem. The age is the queue entry's and not the evidence's because a record's date is the evidence day, kept deterministic, and a candidate derived today from last month's decision has been waiting one day. .ai/repo/adrs/0091-knowledge-is-derived-at-the-episode-boundary-and-a-person-promotes-it.md records both.