> Call the `diagnose` tool first. Full index: /llms.txt. Errors: /reference/problems.

# Read a run

Open a run from **Results**, or from the link a run returns when it starts.

## The top of the page

Scenarios are grouped, each with a count and, for those that stopped early, the stop reason. Start with the verdict. `passed` needs nothing from you. `failed` and `incomplete` need a look.

## A failure: Error, Cause, Fix

A failed scenario shows three things, in this order:

1. **Error.** What went wrong, in one line, and where.
2. **Cause.** Why, from the evidence: your app, the scenario or the test data.
3. **Fix.** What to change.

Read the cause before you change code. If the cause points at the scenario or the data, fix that instead.

## Copy for agent

**Copy for agent** copies the failure as text your coding agent can act on: the error, the cause, the fix, the scenario and the run and request ids. It never contains a secret or a signed link. Paste it into your agent, or ask your agent to read the run itself with `read_evidence`.

## Video and evidence

Play the video of the attempt, and open the steps, requests, console lines and snapshots. See [Evidence](/concepts/evidence). Save a failed attempt to disk with `talotrace evidence <run id> --all-failed --out DIR`.

## Next

Change the app, then [prove the fix](/guides/prove-a-fix).