Recruiting briefs with receipts — every claim cites a resume line — step 8 of 8

Checkpoint: the brief gate. Three candidates, three AI-drafted briefs. Each candidate dict has name, decided_by, resume_lines (the line ids that exist on their resume), and brief (claim dicts like the last step).

Write count_violations(candidate) — a claim is a violation if its receipt is None while labeled "sourced", OR its receipt isn't in resume_lines. Labeled inferences are fine.

Then brief_gate(candidate) returning one string, rules in order:

1. violations > 0 → "HOLD — {n} unverified claims" ("unverified" covers both violation kinds — a missing receipt and a fake one; step four already told you which is worse)
2. decided_by is "auto-screen" → "HOLD — decision owner is a tool, not a person"
3. otherwise → "ATTACH — decision owner {decided_by}"

(A clean brief with a tool as decision owner still HOLDs — the receipts discipline never made the decision human's; rule 2 does.)

Print {name}: {result} per candidate. Expected output:

Maya Torres: ATTACH — decision owner D. Reyes
Jordan Lee: HOLD — 2 unverified claims
Sam Okafor: HOLD — decision owner is a tool, not a person

(Match it byte for byte — the dash after HOLD/ATTACH is an em-dash, —, so copy it from the block above rather than reaching for the hyphen key.)