Incidents, design docs, and the conventions that came out of them.
Notes kept by the platform team. Incident writeups go up once the
timeline is settled, which is usually a few days after the incident
itself. Design docs go up when they are circulated, not when they are
accepted, so some of what is here has not shipped.
Recent incidents
Notifier delivery backlogAction items open2026-08-29 · notifier · Webhook delivery fell behind for eleven hours. Retry ceiling was the proximate cause; the dead-letter path made it invisible.
Archiver index divergenceClosed2026-07-11 · archiver · The what-went-where index disagreed with object storage for a subset of rows moved under the 90-day rule.
Duplicate runs across a DST boundaryClosed2026-06-02 · scheduler · Recurring work scheduled in local time ran twice on the autumn transition and not at all on the spring one.
Design docs
Notifier delivery reworkDraftTarget 2026.10 · Replaces fixed-ceiling retry with a per-destination budget. Open questions on the dead-letter contract.
Passkey recovery flowAcceptedTarget 2026.9.1 · What happens when the only enrolled device is gone, and why the answer is not an email link.
Schema versioning for ingestAcceptedTarget 2026.8.2 · Batches carry the schema version they were written against. Staging validates, it does not coerce.
Elsewhere
Service status and release history live on the console. Sign-in and
account material is on the account help host.
Engineering notes