A scheduled run did not happen

A scheduled occurrence is missing or late

Start in the customer scheduler or event handler that owns the trigger. Find the intended occurrence before submitting anything by hand. A missing local entry does not prove that Compute received no request.

Keep the template version, intended time or business-event identity, submission key, and any returned job reference while you investigate.

Check the owning scheduler first

Confirm the schedule is enabled and review its configured time zone, next intended time, pause state, allowed lateness, retry behavior, and finite catch-up policy. Daylight-saving rules and a changed template version can alter which occurrence you expected without changing an already admitted job.

For event-driven work, check the authenticated business-event identity and account scope. Do not replace a missing event with an unrelated current event merely to make the schedule advance.

Reconcile what already happened

• A job reference exists: read that job's current state. Recover the existing work rather than creating a replacement.
• The submission outcome is unknown: keep the original request association and follow the operation's idempotency and retry rules. A timeout is not proof of refusal.
• The occurrence was skipped locally: decide whether that missed outcome still matters and whether its fixed input is still available before approving catch-up work.
• The trigger was paused or access changed: restore only the reviewed trigger or permission. Pausing, revoking access, canceling admitted work, and deleting results are separate actions.

A new snapshot, changed template, changed destination, or wider task is new work. It needs its own current quote and applicable approval.

Avoid a duplicate recovery

Do not submit a manual replacement while the original occurrence may already be admitted. A local occurrence identifier is not the same as the API's idempotency authority.

If delivery failed after Compute succeeded, recover delivery of the existing result instead of rerunning inference. If a quote expired, first establish whether its request was already admitted.

Know when to stop

Success is an occurrence reconciled to one known admitted job, one explicit local skip, or one clearly unresolved submission. Stop when the original submission, fixed input, approval, or destination cannot be established safely.

Use occurrences and replay for the recovery model and pause and recover customer-scheduled work before resuming a backlog.

If the state remains unclear, use Contact Lithi with the safe job or batch reference, intended occurrence time with time zone, operation name, client version, and what you observed. Leave out payloads, keys, signed URLs, and banking information.