# Reading a report

Whether the checks passed, what is still open, and what is waiting on you.

Source: https://salidium.com/docs/report

A session opens as a report, and the report is built to answer one question first: can you leave this alone, or does it need you? Everything else is underneath that answer.

![The top of a Salidium report: the session title, its tags, and the verdict reading 4 files changed, unverified.](https://salidium.com/docs/masthead-light-19b898be.png)

## The verdict

One line, chosen in a fixed order of priority, and it is the first thing Salidium is willing to say about the run.

- The run is waiting on you.
- A check failed and nothing has passed since.
- The agent is still working, and on what.
- Files changed, and nothing has verified them.
- The last check passed, and which.
- How many things need you, or how many files changed, when there is nothing else to say.

The verdict says how it was reached as well as what it is, so a line worked out rather than read from output says so. Open it and you get the checks it was read from, and [Evidence](https://salidium.com/docs/evidence.md) has the rest of the working.

## Verified

Salidium recognises test, build, typecheck and lint runners by the command that was run, then reads the output.

- **Passed**: The output said so, the exit code said so, or both. Where only one of them was available the row says which.
- **Failed**: The run reported failures, or exited non-zero.
- **Partial**: The output claimed a pass and the exit code disagreed. This is about the two disagreeing, not about how much the run covered.
- **Unknown**: Nothing recognisable came back, or the run has not finished.

A command name is not proof. What the output said outranks what the command looked like, and a run that is not the latest of its kind is flagged as such.

## Scope is not outcome

A run narrowed to a path or a filter is recorded as partial scope, separately from whether it passed. A partial-scope pass never clears unverified, and neither does a lint pass on its own. A check that ran is not a check that covered the work.

## Left

- Plan steps still pending or in progress.
- Checks Salidium saw fail, with no later passing run.
- Up to five things the agent said in its last turn were still outstanding.

Only the second of those is Salidium's own observation. The first is the agent's task list and is marked as planned; the third is the agent's word for it and is marked as reported. [How we know](https://salidium.com/docs/provenance.md) is what tells them apart on the page.

## Needs you

Each entry names the rule that raised it, and says how many times it fired when it fired more than once.

- **A waiting prompt**: The agent asked for permission, for input, or a question.
- **A failed check**: A run failed and nothing has passed since.
- **A failed turn**: The turn itself ended in error.
- **A destructive command**: `rm -rf`, a force push, a hard reset, a discarding checkout, `--no-verify` and four others.
- **A push**: Work left the machine.
- **Unverified changes**: Files changed with no passing check behind them.
- **A claim without evidence**: The agent said checks passed when none ran.