For the people who built it
Your client's automation broke. You wrote it.
Send the workflow export and one failing execution log. Within 24 hours you get the exact line that killed it, why it happened, the fix in your own UI, and a corrected file you import yourself. No account access. No credentials. No call.
[2026-07-28T02:14:07.221Z] execution 41822 node "Google Sheets" START[2026-07-28T02:14:07.884Z] oauth2 → refresh access_token[2026-07-28T02:14:08.102Z] ← 400 {"error":"invalid_grant"} ◄ cause of death[2026-07-28T02:14:08.103Z] node "Google Sheets" ERROR after 882ms[2026-07-28T02:14:08.104Z] execution 41822 FAILEDThe refresh token was not expired. It was revoked, and the cause sits outside n8n: your Google Cloud OAuth client is still in Testing mode, where refresh tokens are invalidated every seven days.
Re-authorising the node fixes it until next Tuesday. Publishing the OAuth client fixes it permanently.
2026-07-19 09:41:02 scenario 118443 module 4 HTTP → GET /v2/contacts?page=72026-07-19 09:41:02 ← 429 Too Many Requests retry-after: 302026-07-19 09:41:02 module 4 error handler: none ◄ cause of death2026-07-19 09:41:02 scenario 118443 STOPPED2026-07-19 09:41:02 bundles processed 312 / 1041One rate-limit response stops the whole scenario, because module 4 has no error handler and no Break route. The API told you to wait 30 seconds; nothing was listening.
The more expensive part is what the log does not say: the 312 bundles already written are not rolled back, so re-running from the top duplicates them.
2026-08-01T11:03:19Z webhook evt_3Pk92f payment_intent.succeeded2026-08-01T11:03:19Z Airtable → Create record rec_A OK2026-08-01T11:03:41Z webhook evt_3Pk92f payment_intent.succeeded ◄ same event id2026-08-01T11:03:41Z Airtable → Create record rec_B OKSame event id, delivered twice, 22 seconds apart. This is not a Stripe bug and not a Zapier bug — webhook delivery is at-least-once by design, and every provider does it.
The Zap needs to recognise an event id it has already seen. Nothing upstream needs fixing, which is why re-authenticating and rebuilding the Zap did not help.
Three real failure classes. Yours is probably one of them.
The problem is not that it's hard. It's that you can't ask.
When you built the automation, the broken automation is your problem. The community forum is free and full of people who would help — and you cannot post your client's workflow on it under your own name.
That is the whole service: somewhere to send it where the answer comes back in writing and nobody finds out.
Every report is confidential by default and white-label by default. If you hand it to your client with your own logo on it, that is between you and them — we are not in the conversation, and we never contact anyone you work for.
What you send
- The workflow exportn8n or Make JSON. Zapier has no export — screenshots of every step.
- One failing execution logThe whole run, not the red error line. A successful run too, if you have one — it often halves the diagnosis.
- Six questionsPlatform and version, what it should do, what changed, deterministic or intermittent, what you already tried.
Redact names, emails and tokens first. We tell you exactly how.
What comes back
- The verdictOne sentence. What is wrong.
- The proofThe exact line in your own log, quoted, and why it is the one that matters.
- The fixSteps in your UI, on your version — not general advice.
- A hardened fileCorrected workflow, yours to import. We never touch your account.
- A testThe specific thing to run, and what you should see.
- What we couldn't determineNamed honestly, with what to capture next. If that is the whole report, it is refunded.
Three things we will not do
Five failure classes, written up in full
Each one is a complete diagnosis: the symptom, the log, the question that separates the likely causes, and the fix. Free, no signup. Yours is often one of these.
One symptom, one report, twenty-four hours.
$99per autopsy
$149same day
One order covers one symptom, plus a round of follow-up questions within seven days. A new and unrelated symptom is a new order. If a second problem turns up while we are diagnosing the first, we tell you for free — buying the fix is your call, and there is no pressure attached to it.
If the report does not help you, say so and it is refunded. We will not argue about it: a disputed order costs more to defend than it earns.
Send an autopsy request →Or write to hello@workflowautopsy.com. Reply comes back with the intake form and where to send the files. No account needed.