Missed Cron runs: define catch-up and lateness policy

A future preview is not execution history

Cron Parser enumerates matching future instants from a start. It does not know whether a deployed scheduler actually ran them.

To diagnose downtime, compare the schedule with recorded executions and use scheduled time separately from observed start time.

Every 5 minutes; service unavailable 00:01–00:16
Candidate missed instants: 00:05, 00:10, 00:15
Whether to replay them is a product policy.

Choose the recovery meaning

A daily aggregate can sometimes recompute one current result, while a sequence of billing periods may require every missing interval.

Define replay-all, latest-only or skip policies explicitly. Each has different latency and load consequences.

Bound lateness and catch-up volume

Schedulers can impose deadlines or limits on missed runs. Kubernetes startingDeadlineSeconds limits how late a job may start under its documented rules.

Check the exact scheduler behavior; resuming a suspended schedule is not universally equivalent to replaying every missed instant.

Make recovery auditable and idempotent

Store the scheduled interval with the result and deduplicate by that interval. Backfills should reuse the same identity as an on-time execution.

Recover in bounded batches and record skipped periods. A successful recovery command is not evidence that every expected interval produced an output.