Open your first investigation

An investigation takes one change from the briefing and shows what it rests on, down to the individual run.

1. Start from the change

The headline repeats the change with its numbers, and the line below names the baseline and how many runs were measured. If those numbers are not convincing, stop here: the investigation will not make a weak signal stronger.

2. Check the steps worth checking

Each kind of change comes with a few steps, for example comparing the slowest runs with typical ones. They say when a step helps; they do not claim a cause that the data does not show.

3. Look at the runs

The evidence lists the linked runs first, then other runs from the same window. If a deployment was recorded just before the change, it is shown alongside. Open a run to see its steps, durations, errors and, when your agent records them, its decisions and reasons.

How runs are ranked is described under Evidence.

You are set up

Your agent sends runs, the briefing compares them, and every change can be traced back to the runs behind it. To line changes up with releases, add lanternfish.record_deployment() to your deploy step.