Read a status page without overstating availability

Identify the observation boundary

A status page reports observations within a scope. Look for the time window, timezone, capability tested, sampling method, and latest update. Treat missing definitions as unknown rather than assuming coverage of every request, model, or region.

A healthy short window is not a long-term service guarantee. A limited twenty-four-hour sample is especially insufficient for a long-term availability claim. Public monitoring also cannot represent your network, concurrency, input length, or client configuration.

Connect symptoms with evidence

When a request fails, record the time and timezone, client version, public model ID, response status, and a request identifier that is safe to share. Determine whether the symptom affects one client, one task, or a wider set of requests, then compare the same period on the status page.

Separate connection time, first output, and total completion time. A stream can begin quickly and fail later. Keep request type and sample size with measurements so short text checks do not become substitutes for long-task evidence.

• Count timeouts, cancellations, rate limits, and application validation failures separately.
• Reproduce with small, non-sensitive samples and a bounded call budget.
• Send support a redacted summary without keys, authentication headers, or full user input.

Plan recovery before an incident

Provide clear failure messages and a manual retry path. Automatic retries need error classification, attempt limits, and a total time budget. Before repeating operations with billing or external side effects, establish whether execution already happened.

Use public signals alongside application monitoring. After recovery, reconcile in-flight tasks and usage records and retain an incident timeline. The NexoToken status entry is a source of public signals; this publication provides no independent monitoring conclusion or availability promise.