Running and reading a run

Starting one

Run a single case, a flow, or a milestone's worth of cases. Before it starts, Cettara estimates what it will cost in AI credits and refuses if there are not enough.

Runs are queued and picked up by a worker. The run page shows where it is.

Watching it happen

While a run is going, you can watch the browser live from the run page.

Reading the report

For every case:

• each checkpoint, with its verdict, confidence, reasoning and the evidence it named;
• the screenshot the judge looked at;
• video of the whole case, and a full execution trace;
• files the run downloaded;
• timing per turn, so you can see something drifting towards a timeout.

A failure is the judge saying your expectation was not met. A timeout is nothing judged at all — the run gave up waiting — and is shown differently, because they mean different things.

Milestones

A milestone gathers everything that must pass before a release, counts manual and automated together, and produces a report to sign.