Recover the diagnostic path and hand off cleanly

Verifiedon 2026-07-302026.7.1-2

Match scope before action

doctor --fix targets named local setup repairs. channels dead-letters resubmit targets one retained inbound event after its cause is corrected. update repair converges update finalization after a core change but does not install a new core package or restart the gateway. These are not interchangeable recovery buttons.

Minimal reproducible handoff

Use this template:

Time window and impact:
Installed version / install method:
Symptom and affected layer/account:
Non-destructive checks + redacted results:
Hypothesis and confidence:
One approved change (or explicit no-change decision):
Post-check result:
Rollback condition / next owner / evidence location:

Escalate when evidence conflicts

Escalate if a gateway snapshot and a live channel probe disagree, a log source changed during follow mode, a config write would replace an unknown object, or a proposed recovery needs credentials, message payloads, or production access you do not own. A precise no-change handoff is a successful operator outcome.

Checkpoint

Which is safer when a dead letter’s cause is unknown: resubmit, delete, or preserve and escalate? Preserve and escalate. The retained failure is evidence; retrying can create new uncertainty before the cause is understood.