If your server-side events fire correctly for a day or two and then start failing with HTTP 400, the payload shape is almost never the problem. Something that changes on a timer is.
| Cause | What you see | Fix |
|---|---|---|
| Access token expired or permissions changed | 400 with an OAuth error / code 190 | Use a system-user token with no expiry, stored in a credential, not inline in a module |
| Test event code left in the live path | Events land in Test Events, or the call rejects it | Strip test_event_code from production, keep it in a separate branch |
| event_time in the wrong unit or window | "invalid event_time" | Unix seconds, not milliseconds; not in the future; not older than 7 days |
| Missing action_source | "action_source required" | Every event needs it: website, physical_store or system_generated |
| Unhashed or malformed user_data | "user_data invalid" | Lowercase, trim, SHA-256 the email and phone; send fbp/fbc when present |
A token that expires, a batch that grows past the request limit, or a retry loop that starts double-counting are all quiet until they are not. The scheduler keeps running, the workflow keeps reporting success, and the only signal is a 400 body nobody reads.
event_id sent on every event, so retries never double-count a purchaseBroken right now? Send the failing response body and I will tell you which of the five it is. Fixed fee for the repair, and if events are not accepted you do not pay.
perun.intelligence@gmail.com — n8n and Make automation builds.