Durable workflows that survive crashes are the right foundation. The failure mode we see most: the loop survives the crash but retries into an active rate-limit window — WAIT signal present, ignored by the retry layer.

60 documented instances if useful: github.com/SirBrenton/pitstop-truth