Sample Continuity Triage Report

System: local n8n workflow used to test AI-agent reset recovery and continuity scoring.

Concern: the local check reported a timeout, making the workflow look unreliable. The open question was whether the workflow failed, the monitor failed, or the recovery evidence was too thin to decide.

Constraint: no production secrets, customer records, private credentials, or unsafe logs are needed for this triage.

The audit compared monitor behavior with execution evidence. The workflow completed successfully after roughly 12.3 seconds, while the original local checker gave up after 10 seconds.

Finding: this was a false timeout, not proof of workflow failure. The correct fix was not to rewrite the workflow. The correct fix was to inspect execution evidence, increase the check timeout, and keep the pass/fail standard attached to the workflow output.

workflow: n8n reset recovery drill
objective: verify whether the agent can recover state after interruption
last verified result: status=200, score=100, verdict=strong, passed=6/6
known stale risk: monitor timeout can be mistaken for workflow failure
evidence required: execution result plus recovery score, not timeout alone
explicit non-inference: do not infer workflow failure from client timeout only
first fix: raise checker timeout and inspect execution evidence before reporting
next safe action: rerun drill; if score drops, inspect failed checks before changing workflow

The recovery record was tested as if the next agent run had no warm chat context. The agent must be able to answer four questions from the persisted record alone:

1. What was the objective?
2. Which result is verified?
3. Which inference is forbidden?
4. What is the next safe action?

Pass criterion: the agent must not report "n8n failed" from the timeout alone. It must say the workflow previously passed, the timeout was a monitoring boundary, and the drill should be rerun with execution evidence attached.

1. Keep timeout budgets explicit in the recovery record.
2. Record execution ID or equivalent evidence with every pass/fail report.
3. Separate workflow health from monitor health in the status output.
4. Add a stale-evidence warning when the last successful run is older than the reporting window.
5. Use the cold-start questions as a regression test after any workflow edit.

This triage prevents the two expensive mistakes: shipping a fragile agent because the demo looked fluent, or throwing away a working system because the monitor produced a false negative. The deliverable is not a motivational audit. It is a small evidence map that tells the builder what failed, what did not fail, what must not be inferred, and what to fix first.