Provider health & errors
Every connection has a health state derived from collection attempts and the age of its last success.
| State | Meaning | Display behavior |
|---|---|---|
| Healthy | Latest refresh succeeded. | Current data is shown. |
| Stale | Latest refresh failed, but a recent last-known-good snapshot exists. | The saved value remains visible and is marked stale. |
| Error | The provider cannot be queried and no sufficiently recent usable value exists. | Error guidance is shown; no zero is invented. |
| Never connected | No refresh has ever succeeded. | Setup/retry guidance is shown. |
Staleness follows the connection's polling interval: approximately two missed expected refreshes, rather than one universal duration. A failed refresh does not create a zero analytics observation or erase successful history.
Error details and retry
Settings → Connected providers shows a warning below affected connections, including a safe action, HTTP status and collection stage when available, last attempt, and last successful sync. Use Retry to poll that connection again. Authentication failures normally require replacing or reauthorizing the credential; network and rate-limit failures may succeed after the cause clears.
Provider errors are sanitized before storage/API display. Credentials, request headers, query secrets, response bodies, and credential-like text are redacted.
Common cases
- No usage: confirm API is enabled, the credential has the required scope, then retry. Some providers expose balances or quotas rather than token usage.
- Usage but no cost: usage access does not imply historical billing access. Configure billing and inspect pricing coverage/Hermes mappings.
- Stale data: check provider reachability, rate limits, and polling status; the displayed reading is last-known-good.
- Partial pricing: map the provider/model correctly and inspect unknown models or missing token classes.
- Hermes not attributed: sync the source, widen the selected range, and fix unmapped aliases under Settings → Data sources.
See Troubleshooting for provider-specific checks.