Resolved · minor 12 Jul 2026 · 38 min · AI ranking · India region
Ranking queue backlog during a bulk import
What happened. A customer imported 61,000 candidate records while three high-volume drives were ranking, and the inference queue backed up. Rankings completed but were delayed by up to 22 minutes; no data was lost and no scores were affected.
What we changed. Per-tenant queue isolation and a burst capacity pool, so one bulk import can no longer slow another workspace. Import throughput is now rate-shaped rather than accepted all at once.
Resolved · minor 28 Jun 2026 · 1 hr 12 min · Messaging · all regions
Delayed candidate emails via one delivery provider
What happened. Our primary transactional email provider had elevated latency in APAC. Messages queued rather than failed, with the longest delay at 47 minutes. WhatsApp delivery was unaffected.
What we changed. Automatic failover to a secondary provider after 90 seconds of elevated latency, and a delivery-lag alert that pages on-call before customers notice.
Resolved · maintenance 11 Jun 2026 · 26 min · Scheduling and recording
Planned storage migration for interview recordings
What happened. Announced 5 days in advance. Recording playback was read-only during the window; new recordings continued to capture and were available once the migration completed.
What we changed. Nothing to fix — but the migration halved playback start time and moved recordings onto region-pinned storage in all four data regions.
Resolved · major 22 May 2026 · 51 min · Job board posting
Postings to one board failed silently
What happened. A partner board changed an authentication header without notice. Our retry logic treated the rejection as a soft failure, so 340 postings queued without alerting the recruiters who created them.
What we changed. Silent-failure detection: any posting not confirmed within 10 minutes now surfaces on the requisition and notifies the owner. We also added contract tests against every board's auth flow, run hourly.
