Before you start

You need access to the workflow/run and authority for the proposed recovery action. Find the existing run reference before starting a replacement: a failed or delayed display can conceal an external effect already completed.

Locate the failing boundary

  1. Open the workflow’s runs and select the relevant execution. Confirm start time, actor and status.
  2. Inspect the execution checklist, step outcomes, mapped work and event timeline. Find the first unresolved step rather than assuming the final summary is the cause.
  3. Open any linked task/approval/document and check its current state. A human task may be waiting legitimately.

Choose the remedy

Missing input, permissions, approval, provider availability and stale version bindings require different remedies. Correct the named blocker in its owning record and use only a recovery action offered by the application.

Where Cancel run is available for an active run, review the impact and record a clear reason. Cancellation does not roll back a sent invitation, stored file or other completed effect.

Verify recovery

  1. Inspect prior attempts and receipts before retrying anything consequential.
  2. Check the resulting step/run state after the supported action.
  3. Confirm outstanding human work and downstream effects separately.
  4. Escalate a repeated/uncertain failure with run reference, stage, time and visible message; exclude tokens and credentials.