Automation Runs and Recovery in Velveteen
Updated August 16, 2026
Every automation run shows its current state, steps, waits, approvals, redacted outputs, and safe recovery options. Transient failures retry automatically; uncertain provider outcomes pause for reconciliation and alert the automation owner.
Runs are durable and leased so overlapping cron invocations cannot duplicate a step. When something needs attention, the run stays honest about whether an external effect is unknown.
Recover without duplicating effects
Retry
Retry a failed or blocked step with the same effect key after the cause is fixed.
Restart
Start a distinct run only when the automation's re-entry policy permits it.
Cancel
Cancel eligible queued or waiting work without falsifying an in-flight provider result.
Approve
An authorized collaborator approves a snapshot-matched high-impact step; material changes invalidate the approval.
Frequently asked questions
Can I retry a completed run?+
No. Terminal runs remain immutable. Use the automation's re-entry policy and a new event or an explicit restart when allowed.
What does unknown outcome mean?+
Velveteen cannot yet prove whether a provider effect completed. The run remains visible for reconciliation and is never marked successful by guesswork.
Related articles