Troubleshooting

Locate the failure

Symptom	Inspect	Next action
No request arrives	Schedule status, cadence due day, identity and cache, then Communication Message for that person	Correct the first failed dependency; a rerun sends to everyone due
Message is Failed with invalid_arguments	The person’s Slack account: the real cause is usually user_not_found when opening the DM	Fix the identity binding or Tool Account; failed rows are not retried
Person asked twice for one KPI	Two memberships, a second run that day, or the legacy Operations base	Check Message rows for that day; the legacy base is outside this system
Fewer items than expected	More than 25 items due for that person	Items beyond 25 are counted, not sent
expired_trigger_id or retry notice	Cache readiness and age before modal opening	Refresh credentials if needed and click again for a new trigger
Form opens but context fails	Schedule, target metric and membership	Correct context; close and reopen
Numeric validation fails	Whole number, non-negative value, configured decimal policy	Enter a valid value for the supported input
Evidence fails	File ID, access, extension, size and existing attachments	Recover the same Observation
Processing persists	Observation Date and execution stage	Inspect after the short in-flight guard; do not assume no save occurred
Duplicate collection keys	Rows sharing Submission Key	Reconcile values and evidence before retrying; do not delete blindly
Value saved but checkbox wrong	Recorded state and Slack update result	Repair message synchronization after checking persisted data

Diagnostic boundaries

Record workflow, stage, timestamp, count and error code. Do not copy tokens, raw Slack payloads, private response URLs or evidence contents into logs. Execution payload saving is disabled, so an empty execution history is not proof the workflow never ran. Use Airtable state and Communication Message together. A dry-run- External Message Id means nothing was sent. Keep a failed row until its value and attachments have been reconciled.

What remains untested

A simultaneous duplicate submission, an upstream outage, a revoked token and a partial multi-file upload have not been reproduced. The handling above follows the workflow logic; test it in a controlled run before relying on it.