PAGEADVICE / METHOD
How a review is made
Every claim below can be checked against a report this service produced. Where the review cannot know something, it says so rather than filling the gap.
What the review is looking at
A review loads the page you submit the way a visitor would, captures it, and works from what was actually on the screen: the text, the headings, the controls, the images, the colours, and the measured timings of that one load. It does not run scripts of its own against your site, does not sign in, and does not click anything. Anything that would need those is outside what a report can honestly claim.
The standards it works from
Usability findings are assessed against Nielsen's ten usability heuristics. Accessibility findings are assessed against the subset of WCAG 2.2 Level A and AA criteria that a still capture can genuinely evidence — contrast, alternative text, structure, labels, target size, and the like. Criteria about keyboard operation, focus order, and timing are deliberately excluded, because nothing here operates the page and a citation we cannot demonstrate would be worth less than no citation at all.
Why a finding cites a clause, or does not
A finding names a criterion only when its own evidence demonstrates that criterion. Most findings cite nothing, and that is the intended outcome. A wrong clause number borrows an authority the finding has not earned, and one checkable error would discredit every other citation in the report.
Why the reports are short
A widely cited study by the Baymard Institute, reported by Jakob Nielsen, put a general-purpose model to work on ecommerce screenshots and found that 19% of its suggestions were sound, 72% would have wasted implementation time, and 9% were actively harmful — averaging eleven invented findings per screenshot. Length is easy to produce and expensive to read: every false finding costs you the time to investigate it and the trust you had in the rest of the list. A report here normally carries three to eight findings. That is a choice about signal, not a limit on what could be listed.
Why the marks on the screenshot are in the right place
The capture records the position of each element it found, and a finding can only point at the page by naming one of those recorded elements. The coordinates are resolved afterwards from the capture, so a location cannot be estimated or invented. When no captured element matches, the finding is shown without a marker instead of being anchored approximately — mis-marking the wrong part of your page is a worse failure than not marking it at all.
How the score is arrived at
The overall score is bounded by the findings rather than chosen freely. Any P0 finding caps it at 69; any P1 caps it at 84; a page whose findings are all P2 starts at 75; and the overall score never sits more than 15 points above the weakest dimension. So a low score always has a specific finding behind it, and a high score cannot coexist with a serious one.
What this does not do
It does not produce an attention or eye-tracking heatmap. Those require gaze data from real people; a model asked to guess where attention falls can produce a convincing picture with nothing behind it, and this service would rather show you nothing than something plausible and unfounded. A review is also not a substitute for testing with real users, nor an accessibility, legal, or security certification. It is a fast, evidenced first pass — the last word belongs to you.