Your first run

Pick the check

In the console, New run opens the picker. Full check sits at the top; under it are the five you can run on their own, each labelled with whether it needs a key.

Watch it work

A run streams. The console prints each surface as it is reached, what was measured there, and what it found, while it is happening, not at the end.

Nothing about this is a progress bar in the decorative sense. If a run stops streaming, it is stuck, and the last line tells you where.

What comes back

Findings, ranked by confidence multiplied by severity, the same order the console shows and the CLI’s actions list prints. Each one carries:

• A title in plain language. “A form cannot be submitted at all”, not “aria-required-children”.
• The surface and the selector it was found on.
• The evidence: what the browser saw, and usually the frame it saw it in.
• A fix prompt, ending in an acceptance check a machine can settle.

Then give them verdicts

This is the part that is easy to skip and should not be. A finding you confirm becomes a standing check; a finding you reject teaches the tool it was wrong. Until verdicts exist, the confidence percentages have nothing to calibrate against, and the product will say so rather than print a number it cannot support.

The four verdicts, and what each one costs.