A profile sets how often a worker should checkpoint, and a stale task is reported

What it means

Each profile has a checkpoint_interval. check --checkpoint records a checkpoint; check and watch compare the last checkpoint with the interval and report a task that has gone quiet for longer than its profile allows.

How it works

checkpoint_at on the task record is updated by check --checkpoint, by handover, and by finish. lib/check.sh computes the age against the profile interval and emits a warning; lib/watch.sh emits staleness drift. The staleness report is deterministic; the worker's checkpointing is not.

How to see it

majordomus check --checkpoint
majordomus watch     # DRIFT staleness … when the interval has passed

What it does not cover

Advisory on the worker's side: whether it checkpoints is up to it. The report of a stale task is guaranteed; a stale checkpoint is a warning in check and never blocks.

Why it exists

Sessions that never end were a named failure; a checkpoint interval makes "this task has not been touched in an hour" a fact the tool can state.