Orchestrator drained an answered-decision wake signal (#74 merge_now) without executing the approved action — the merge was silently dropped

Operator answered decision #74 (merge #278 live-roster migration) with `merge_now` at 2026-07-05T15:48. A `decision_wake` signal (#177, target orchestrator daemon 53) was enqueued. By ~16:19 recall_signals was empty (the signal had been drained), the decision showed answered/acked — but the orchestrator had NOT performed the approved merge/migration: `flower/278` was still unmerged, the migration unrun, the live daemon_agents key unchanged. The orchestrator's session goal had moved on to dispatching #275/#276 (it had been absorbed in #110 tailscale-serve just before). So an operator-approved action was consumed (wake drained / decision acked) without being executed — it fell through the crack and I had to run the merge+migration manually. Likely cause: the decision-wake → recall_decisions → decision_ack path can ack/drain the wake without a durable follow-through on the action, especially when the orchestrator is heads-down on another task and its queue backs up (4 signals were pending at once). Suggest: don't let a decision be acked until its action (or a follow-up dispatch/todo) is created, or re-enqueue if the action isn't recorded. Context: brief #278, decision #74, signal #177, orchestrator daemon 53.

{
    "brief": 278,
    "signal": 177,
    "decision": 74,
    "observed": "wake drained + decision acked but merge/migration not executed",
    "operator_answer": "merge_now",
    "orchestrator_daemon": 53
}

created

3h ago

triaged

—

resolved

—

resolved by

—