Scheduled Job Specification

You are an engineer who has debugged a cron job that ran twice, and one that
never ran at all for six weeks.

## Inputs
What the job does: {{JOB}}
How often it should run: {{SCHEDULE}}
How long it takes: {{DURATION}}
What it touches: {{SYSTEMS}}

## Task
Specify it properly.

## Decide and state
1. Schedule in cron syntax, with the timezone named explicitly. Ambiguous
   timezone is the classic cause of the twice-a-year double or missed run.
2. Overlap: what happens if run N is still going when run N+1 is due, given
   {{DURATION}}. Lock, skip, or queue.
3. Idempotency: whether running it twice for the same period is safe, and if
   not, what makes it safe.
4. Catch-up: if it misses a run, should it process the gap or skip forward?
5. Boundaries: what "yesterday" means at 00:00, across timezones, and on a
   daylight-saving change.
6. Observability: a heartbeat, so a job that never runs is detectable. A
   silent absence is the failure mode nobody alerts on.
7. Run time budget, and what to do when exceeded.

## Output format
### Schedule
Cron expression, timezone, and what it means in words.

### Decisions
| Question | Decision | Consequence if wrong |

### Monitoring
What alerts, when, and the message contents.

### Manual operation
How to run it by hand for one specific period, safely.

## Self-check
Walk through a daylight-saving transition and a run that takes twice as long
as expected. If either produces undefined behaviour, the spec is incomplete.