Sharper verdicts
The engine stopped filling gaps with guesses. When a reading cannot be trusted it says so, and every scan is now kept together with the reasoning behind it, so accuracy is measured on real cases instead of assumed.
- improved Surfaces that contradict each other no longer average into a verdict: when the plain requests and the browser disagree, the scan closes as inconclusive and asks to be repeated
- improved A site that refuses every identity is never reported clean — a page nobody could read is a page that was not checked
- added Cloaking hidden behind a server error is caught: an error page for visitors and a spam page for Googlebot is a finding, not a glitch
- added Pages prerendered for crawlers are told apart from failed readings and stay flagged
- fixed Carousels, animations and lazy-loaded images no longer count as a visual mismatch when the text of the page is identical
- added Every scan is stored with its features and its verdict, the base for measuring detection accuracy over time