Built around one idea.
A check that could not run is never counted as a pass. Almost everything else about this product follows from taking that seriously.
The problem
Search tooling splits into two halves that both stop short. Reports end at a number: a score, a PDF, four hundred rows, and a bill, with the fixing left as your weekend. Content tools end at a brief. They tell you what to write, hand it back, and nothing gets published or measured.
Underneath both sits a quieter problem: the numbers are frequently soft. Checks that could not run get counted as passing, so a 94/100 can mean "we looked at a third of your site and liked what we saw."
The idea
Three outcomes, not two. Pass, fail, and unmeasured, and unmeasured is never rounded up. Coverage is reported next to every score, so a number is always a bounded claim about what was actually examined rather than an implied claim about everything.
This sounds like table stakes and is not. It is the reason a Crawld score reads lower than a competitor's on the same site, and the reason it means more.
What follows from it
- Every check declares its method (deterministic rule, graph analysis, third-party API, or model judgement) so you can weigh a finding before acting on it.
- Every check declares who may fix it. Where only a person can decide, the engine reports and stops rather than guessing convincingly.
- Verification labels follow the access. Build-verified means a build ran. A hosted CMS has no build, so it is never given that badge.
- The loop closes. Changes are re-measured, because a fix that shipped and a fix that worked are different things.
What it refuses to do
The loop stops twice for human approval and that cannot be turned off. An unsupervised agent editing a production site is a different product with a different risk profile, and not this one. If the value you want is that nobody has to look at it, this will feel like friction permanently.
It also does not promise rankings. Nobody can. The rubric measures things within your control; where you place is not among them.
Where it currently is
The measurement half is real and running: the free scorecard crawls, scores against the full rubric, and fails honestly when the engine is unavailable. The delivery half (pull requests, CMS drafts, the access ladder) is designed and not built. The transparency page lists which is which, and every integration page carries its own status.
Publishing that distinction is not modesty. A product whose entire claim is honest measurement cannot be vague about its own state.