One call hands the caller an unheld, unreserved alarm, ALREADY CLAIMED, in one transaction, chosen best-effort among the alarms the statement can see and lock. Order: any eligible alarm queued at least 45 s first (oldest of those first); below that the caller's own Focus alarms; then the oldest in the queue (queued_at, then id). Severity does not sort the hand-out. Never a live-held alarm (including the caller's own), an alarm validly reserved for another operator's Focus, or (with CLAIM_ESCALATION_ENABLED on) an alarm with a live call or announcement in progress (a snapshot read, not a commit-time guarantee). Two operators pressing at once get distinct alarms. The claim is an ordinary claim (120 s TTL, renewed by the review page, a fresh ownership generation). Manual clicks and Take over are unchanged.

request_id identifies ONE press: retrying the same id (a lost response, a timeout) returns the RECORDED result with the alarm's current state (replay: true; current is yours, taken_over, closed, lapsed, reserved, out_of_scope or unavailable) and never selects a second alarm; the record is kept 24 hours. status: none carries a reason and grouped counts inside the caller's scope: empty, all_held, all_reserved, all_in_session, contended (others are picking right now), scan_budget (the 200-row scan stopped early; pick from the list) or not_a_claimant (owners never take claims). 503 outcome: not_applied (this call established that the request_id has no record AND its allocation was rolled back: nothing claimed, the id may be abandoned) or unknown (anything else: a failure before the record was looked up, while replaying it, or after the COMMIT may have been sent; keep the id and retry the SAME request_id, the server's record decides) with Retry-After. Requires review_label_alarms and central staff (403 otherwise). 404 as if the route did not exist while CLAIM_NEXT_ENABLED is off (read claim_next_enabled on GET /auth/me).

{
  "request_id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
  "source": "queue_key"
}