Reliability and status

Status page

status.scispace.com — per-endpoint-group status, uptime history, and incident updates. Subscribe there for notifications; we do not email status changes by default.

Availability target

99.9% monthly availability for GA endpoints, measured on successful responses to well-formed requests, excluding 4xx. The target is a published objective on Developer and Scale, and contractual on Enterprise.

What is excluded from the calculation

• 429 responses: throttling is the system working as designed (rate-limits)

• 402 responses: exhausted credits are not an outage

• Scheduled maintenance announced 72 hours ahead on the status page

Building for degradation

• Treat 503 as retryable with backoff; treat sustained 429 as a signal to lower concurrency permanently

• Fail soft on depth. If deep_review is unavailable or slow, fall back to standard rather than showing an error

• Persist resource IDs before you wait. A srch_ ID survives your process crashing; an in-memory poll loop does not

• Cache the last good answer where staleness is tolerable, within content-licensing limits

• Set client timeouts above the p95 for the depth you use (depth): 15 s standard, 45 s high quality, 240 s deep review

Incident communication

Sev-1 (API unavailable) and Sev-2 (major degradation) are posted to the status page within 15 minutes and updated at least every 30 minutes until resolved. A public postmortem follows within 5 business days for Sev-1.

Related

errors-and-retries · jobs · rate-limits · get-support