The Parse API in the EU region returned no errors after 21:18 UTC. This incident is resolved.

The Parse API is fully recovered. This incident is resolved.

This incident is resolved. Parse latency has been normal since 8:35 PM PDT on September 27.

This is resolved. Parse requests with embed_pdf_metadata set to true are operating normally.

The change was rolled back at 16:33 PDT, restoring normal Deep Split operation. The degradation lasted approximately 11 minutes and was limited to Deep Split requests.

This is resolved. The GPU cluster issue has been fixed and Extract latency, including requests that use citations or confidence scores, has returned to normal. We confirmed full recovery. Thank you for your patience.

This is resolved. The GPU cluster issue has been fixed and parse latency has returned to normal. We confirmed recovery across the fleet. Thank you for your patience.

We corrected a misconfiguration in our GPU inference fleet. Parse latency returned to normal levels.

This is resolved. The cause was a Redis connection issue, which we have fixed. Extract and Split latency is back to normal. Thank you for your patience.

We identified all related bottlenecks and have deployed a hardened version of our API. We are seeing recovery on our end.

This incident is now resolved. The upstream Google Cloud Vision OCR degradation recovered at approximately 09:48 PT / 16:48 UTC. Our fallback OCR paths have fully disengaged, error rates have returned to baseline, and Parse/Extract requests are operating normally. We will continue to monitor. Thank you for your patience.