Skip to content

How it works · Step 4 of 4

Confirm: did the fix hold?

After the fix runs, the Worker re-checks the target the fix was meant to repair. If the check fails, the reverse steps its author wrote run in reverse order.

The step

Confirm re-checks the fix's own target.

AI drafts the check when the ActionBook is written. Your engineer corrects it. Your team decides what recovered means.

A run that only says “executed” has not told you whether the service is back.

Illustration: the Worker re-checks its own target on the server. The check passes, and the run closes on the record.

The fix held

The run closes, and the record says so.

Illustration: the re-check on the server fails, so the event stays open. The reverse steps the fix's author wrote run in reverse order. The record names what could not be undone.

The fix didn’t hold

The reverse steps its author wrote run in reverse order. The event stays open, and the record names what could not be undone.

What it checks today

Confirm runs the check written into the fix.

Here is the line.

Illustration: Confirm checks the fix's own target on the server. A before-and-after measurement dashboard, greyed out, is not part of Confirm today.

The run ends with a record. The record shows the event, the facts the Worker gathered and how the run ended. If a fix ran, it also shows the fix and its re-check.

How the boundary and the record work →