Results & Reliability
Outcomes, conversations, email/notifications, and the idempotency guarantees
A schedule is only useful if you can see what it did and trust that it ran correctly. This page covers result delivery and the reliability guarantees.
Seeing what happened
Every run records a visible outcome:
- Status — completed or failed — with a snapshot of the output.
- Conversation — each run is threaded into a conversation you can open to read the full prompt/response exchange, including every run over time.
- History — a list of past runs with their status and timing.
Getting the result delivered
You don't have to check the console:
- Notifications — get notified in-app when a task completes.
- Email — have the result emailed to you (great for a daily report or digest).
Turn these on for any task whose output you actually need to read.
Reliability guarantees
The fire path is idempotent and downtime-safe:
- No infinite refire — the next run time advances via a compare-and-swap before execution, so a failure to publish or run can't leave a task firing every tick. A missed run is recorded as failed rather than replayed forever.
- No double-fire under high availability — if two schedulers see the same due task, only one wins the claim; the other does nothing.
- Downtime-safe — if the platform was down when a task was due, the missed slot is handled cleanly on recovery (recorded, advanced) rather than fired repeatedly to "catch up".
- Run-now is isolated — a manual Run now doesn't disturb the schedule's next run time.
Troubleshooting
| Symptom | Likely cause / fix |
|---|---|
| Ran at the wrong hour | Timezone not set (or set wrong) — see Scheduling. |
| Task shows failed | Open the run's conversation/output snapshot for the error; check the agent/workflow it targets. |
| Didn't get the result | Enable notification/email delivery on the task. |
| Want to test without waiting | Use Run now. |
Related
- Scheduling — defining when a task runs.
- Scheduled Tasks API — programmatic access.