When automations look fine — catch silent breakage

False-green symptoms

• Open run history, not only the On switch. Zapier publishes distinct statuses, including Successful, Errored, On hold, Filtered, Scheduled, Delayed, Safely halted, Handled error, Needs review, and Skipped. A Zap run status may differ from its step statuses, and the run shows only one status even when more than one applies. A poll that finds no new data does not appear in Zap runs. Source: Zapier — review run statuses in Zap workflows · accessed 2026-09-16. Classes: failure taxonomy.
• Safely halted, Filtered, and Scheduled are not success writes. Safely halted is a purposeful stop, usually a search with no row — class G. Subsequent steps still run; required fields mapped from that search can error or show Skipped. Filtered means the Filter did not continue. Scheduled means Autoreplay is holding an Errored step. Handled error means an alternative workflow ran — the Zap stays on. Sources: Zapier — review run statuses in Zap workflows · Zapier — set up custom error handling · accessed 2026-09-16
• A 401 is a reconnect, not a new scenario. Zapier’s troubleshoot page names expired or revoked access tokens, password changes, and session policies as 401 causes. Reconnect. If the OAuth popup fails, try another browser. Do not rebuild the Zap first. Source: Zapier — how to troubleshoot errors in Zap workflows · accessed 2026-09-16
• Check the destination, not only the history icon. A successful-looking module can still write the wrong row, skip a bundle, or stop on a filter. n8n’s Executions list filters Failed, Running, Success, or Waiting — Success is a completed run, not a checked write. Make Skip or Resume can finish as Success after a dropped or substituted bundle; that product fact stays on error handling — this page does not reprint that compare. Compare the last destination write to the last run. If they disagree, name the class before you add steps. Source: n8n — view all executions · accessed 2026-09-16

Reconnect loops

• Read the expired-connection prompt as class A. Zapier’s help says to open App Connections, reconnect, and test. A reconnect updates every Zap that uses that account. Password changes, domain changes, and extra security controls commonly expire a connection. Source: Zapier — manage your app connections · accessed 2026-09-16
• Signing into Zapier with Google is not the Gmail connection. A Zapier Community Gmail thread (2025-08-04) reports “This account connection is expired” after a Google sign-in. Reauthorize Gmail permissions on the connection, not only the Zapier login. Source: Zapier Community — Gmail reconnect / account connection expired · accessed 2026-09-16
• If reconnect keeps coming back, delete and create a new connection. A Zapier Community Discord thread (2025) describes the expired prompt returning after several reconnects. Staff advice: delete the old connection, create a new one, then point the Zap at it. Do not read a view count from that thread. Source: Zapier Community — repeat reconnect / account connection expired · accessed 2026-09-16
• Weekly Google “invalid refresh token” is often the OAuth app, not the scenario. A Make Community thread describes Gmail or Drive failing every 7 days while the scenario stays scheduled. The named cause in that thread is a Google Cloud OAuth app left in Testing, not Production. Automatic reauthorize is not always possible once the refresh token is dead. Sources: Make Community — connection expires every 7 days / invalid refresh token · Make Community — OAuth expires after 24 hours / reauthorise · accessed 2026-09-16
• A saved n8n credential can still need a reconnect click. n8n Community threads describe OAuth2 tokens that do not refresh when the API returns HTTP 200 with an expiry in the body, or when a refresh token was never stored. The credential still sits in the list. Reconnect or rebuild the credential. Do not treat that as a mapping bug. Sources: n8n Community — OAuth2 token won’t refresh · n8n Community — generic OAuth2 API not refreshing the token · accessed 2026-09-16
• The same loop shows up outside the scored trio. Latenode Community threads describe scheduled jobs that die when a token expires, and a Reddit connection that fails again after a manual reconnect. Those threads are evidence from operators. Latenode is not a scored seat on this site. This page does not add a /500 for it. Sources: Latenode Community — authentication tokens expiring in scheduled scrapers · Latenode Community — Reddit connection keeps failing / reconnect · accessed 2026-09-16

Who is paged vs who owns the fix

• The inbox and the reconnect button can be different people. Name both before go-live. Who gets paged is who receives the vendor mail or the Error Trigger. Who owns the fix is who can click Reconnect, reauthorize, or rebuild the credential. If those names differ and nobody wrote it down, a quiet miss stays quiet. Name both jobs on Who should own your automations.
• Zapier pages the account email; only the connection owner can reconnect. Default mail goes to the Zapier account address immediately. You can switch to hourly summary or Never — Never is not recommended. A custom error handler sends no error email; the run is Handled. Reconnect is only available for connections you own. When you turn on or publish a Zap, the Zap owner must have access to every app connection used in it. Sources: Zapier — manage notifications when errors occur · Zapier — set up custom error handling · Zapier — Zap owner must have every connection · Zapier — manage your app connections · accessed 2026-09-16
• Make emails errors and deactivations by default; reauthorize is a person. Default mail covers an error that blocks a successful run, and an automatic disable. Warning emails are optional — they cover issues that need attention, including connection problems. Make sends immediately, then a digest (5 minutes for errors, 15 minutes for warnings) with a link to scenario history. OAuth grants access for a limited time; reauthorize, or create a new connection. Sources: Make — manage your email preferences · Make — connect an application · Make — fix errors and warnings · accessed 2026-09-16
• n8n does not mail unless you wire it. Create a workflow that starts with the Error Trigger, then select it under Error workflow in the job’s settings. It runs on automatic failures, not on a manual test. The person who gets that Slack or mail is not automatically the person who can rebuild the credential. Sources: n8n — Error Trigger · n8n — handle errors gracefully · accessed 2026-09-16. Class F on the failure taxonomy.
• If you self-host n8n, name who owns 2am. Losing N8N_ENCRYPTION_KEY is not a reconnect — stored credentials become unreadable. It is not a fourth cloud seat. See when n8n self-host is worth it.
• If nobody would notice a skip, do not ship yet. Pass or defer that job on What’s worth automating.

Weekly spot-checks

Vendors do not publish a weekly audit cadence. This page does not add one, and it does not add a monthly dollar. Once a week, pick one critical live job and prove the write still happens.

• Open history for that job — not only the On or Active switch. Zapier: Zap history / Zap runs. Make: scenario history from the digest link or the scenario. n8n: Overview or project Executions; filter Failed and Success. A poll that finds no new data will not appear in Zap runs — confirm the destination anyway. Sources: Zapier — review run statuses in Zap workflows · n8n — view all executions · accessed 2026-09-16
• Test the connection that owns the write. Zapier: open App Connections and test. Make: reauthorize if verification fails. n8n: open the credential; a saved row is not proof the token still refreshes. Sources: Zapier — manage your app connections · Make — connect an application · n8n Community — OAuth2 token won’t refresh · accessed 2026-09-16
• Confirm the last destination write by hand. Open the sheet, inbox, or CRM. Match one live run to one row, message, or file. That write needs one named hub — a single source and write path. If people still edit a second tracker, a Successful run can look fine while the hub is stale. Name the hub on Name one system of record for small-team ops — Notion, Airtable, or Superhuman Docs (formerly Coda). If history looks fine and the destination does not, stop — do not add steps yet.
• Confirm the notification path is still live. Zapier is not set to Never. Make still mails errors and deactivations. n8n still has an Error Trigger selected in Workflow Settings. The cover person can still open that inbox. Sources: Zapier — manage notifications when errors occur · Make — manage your email preferences · n8n — Error Trigger · accessed 2026-09-16
• Confirm the connection owner still works there. If the person who connected the app left, the next expiry is a silent stop. Zapier: the Zap owner must have every connection. Make: reauthorize or replace the connection. Transfer before someone leaves — see Who should own your automations.

When to stop and name the failure class

• Stop rewriting when the UI still looks healthy and the write is wrong. Reconnect loops, a 401, Safely halted, Filtered, Scheduled Autoreplay, a saved credential that will not refresh, or a destination miss with a green history are enough. Adding steps will not name the class.
• Open the taxonomy and name A–G. Auth, bad data, rate limit, timeout, quota, handled versus unhandled, and expected non-match are different jobs. This page does not add an eighth class. Canonical list: When automations fail — a practical taxonomy.
• Then compare the product — do not reprint it here. What each UI calls success, who retries, and who is emailed live on Error handling across Make, n8n, and Zapier.
• If you are switching tools because reconnects ate the week, do not paste the export. Credentials, history, and ownership do not travel — what does not migrate — then run both copies before you switch and compare the failure path before you cut over.

What to log

• Connection last-tested time, and whether the test said active or expired.
• Run status as the vendor names it — not only on or off.
• The vendor error name (AccountValidationError, 401, Safely halted, Error Trigger).
• Whether a search was a non-match (class G) versus a missing required field (class B).
• Who was alerted, who can click reconnect, and whether they confirmed the destination write.
• The last successful row, message, or file in the destination system.

Those facts tell you the class. They are not a score. If you are switching tools because reconnects ate the week, run both copies before you switch and compare the failure path before you cut over.