Throttle
Inspect delay signals and their estimated impact.
Open Insights → Throttle to investigate rate limits, repeated retries, and service-outage signals in your activity.
Review a delay
Section titled “Review a delay”- Select the date range and read the throttle rate and the estimated wait. Waits, events, and the trend cover only the latest days of the selected range; the screen states which days.
- Find a large bar in Wait time by period or a large event in the Throttle events table. The color of a bar shows which detector caused the wait.
- Read the event’s time, detector, details, estimated time lost, and session.
- Compare any main cause shown in the summary with the actual request and service conditions.
Interpret the detectors
Section titled “Interpret the detectors”- Rate limit identifies requests delayed by a 429 or 529 response.
- Prompt retry identifies a prompt that was retried repeatedly.
- Service outage identifies a window where requests failed together across several sessions. A banner counts them and says how many were confirmed on the Anthropic status page.
The summary groups waiting time as improvable or platform-related. These groupings suggest where to investigate; they do not prove who caused a delay. A rate-limit event is not proof that your prompt was poor.
Only confirmed throttling counts. Normally slow responses and idle time are not part of the estimate. Waiting-time values are estimates from captured signals, not a complete measurement of lost working time. An empty table means no matching captured events in that range, not that the service never slowed down. Use Telemetry for request and error detail.