Why it refuses to guess
A tool that quietly assumes what your product is will hand you confident findings about
something else. Ranking a marketing page for “clarity in five seconds” is right; doing it to an
internal admin panel is nonsense. So init asks, and every module reasons through
the answers.
What the profile holds
| Field | What it changes |
|---|---|
| Name and URL | The target every run drives, and the name that appears in reports. |
| Description | What the product claims to do. Site review checks the pages against it. |
| Audience | Who Test drive draws testers to resemble, and whose expectations copy is judged against. |
| Surfaces | The pages every run covers. This is the list, not a crawl limit. |
| Competitors | What “different enough to remember” is measured against. |
| Code location | Lets a finding name the file, so a fix prompt points at something real. |
| Test account | Optional. Without it, runs see only what a logged-out visitor sees. |
Surfaces are the unit of coverage
A surface is a page the profile declares. Runs drive surfaces; coverage is measured against them; the memory graph is built out of them. If a page is not a surface, no run will meet it.
You do not have to write the list by hand:
node cli.cjs mapTwo browser passes (one as a visitor, one signed in if there is a recipe) recording which pages exist, which stand behind the sign-in and which are dead. Free, no model calls, and it writes the map every later run reads.
--scenarios filters the surfaces the profile already declares. It cannot add
one. If a run only covered your landing page, the profile only had your landing page in it.
Knowing what is not covered
node cli.cjs coverageHow much of the product is actually tested: the pages your front door links to, measured against the surfaces runs really drive. Free, no key. It is the honest counterweight to a green report: a clean run over three pages is not a clean product.