How the eight fit together

They are not five versions of the same thing. Three are deterministic measurement; two are a model reading your product. Knowing which is which is most of knowing when to run what.

The eight

Check	What it does	Key
Speed & errors	Core Web Vitals, accessibility, links, console errors, and controls that do nothing when clicked.	No
Re-checks	Turns every finding you confirmed into a standing check and re-runs the suite.	No
Report	Assembles coverage, what changed and the next moves into one shareable document.	No
Assistant readiness	Checks the product works for an AI assistant acting for a person: named controls, keyboard reach, machine-readable identity.	No
Round trips	Types a marked value into a form, saves, comes back and reads it out again.	No
Environment diff	Captures the same pages on two versions of your site and compares them, pixels and words.	No
Site review	Reads every page like a first visit: clarity, proof, trust, and whether it looks finished.	Yes
Test drive	Sends up to six simulated testers through the product, thinking aloud.	Yes

Measure versus reason

The six free ones produce the same answer twice given the same page. They are measurement: a contrast ratio is a number, a dead link is a status code, a blocked submit is an event that did or did not fire. Nothing is being judged, so nothing needs a model and nothing costs.

The run picker: Full check at the top, then the checks you can run on their own, with what each one costs. — The picker lists every check and marks the ones that need no key.

The two that cost are reading your product the way a person would, which is a judgement. That is worth paying for exactly when a judgement is what you need (is this page convincing, would somebody give up here) and worth nothing when you want to know whether the button works.

What to run when

Every save, while you work: node cli.cjs watch. Defaults to the free check, because a loop that costs money on every save is a loop you switch off.
Before a commit: Speed & errors, plus Re-checks once you have confirmed findings.
Before a release: Full check, then read the Report.
When the product changed shape: Site review, then Test drive on the flow that changed.

Full check

node cli.cjs sweep

Runs Speed & errors, Site review, Re-checks and Report in one pass and opens the console on the result. Test drive stays its own decision, because six testers is a different size of run and a different size of bill.