Detector accuracy update: what got more honest
Font detection, WebRTC timeouts, cookie-store reads and benchmark labels all got stricter. Here is what changed — and how to re-run your before/after comparison.
What changed
Font probing now measures rendering instead of trusting a browser API that answered yes to everything, so reported font counts finally reflect reality.
A timed-out WebRTC probe no longer scores as a clean result, cookie storage reads use the correct browser interface, and the benchmark section dropped point scores it could never measure.
Why your score may move
Earlier runs may have been flattering: a permanent font failure or an unscored timeout could push the number in either direction. A changed score after this update reflects better measurement, not a changed browser.
Unknown is now the default wherever the browser withholds an answer — excluded from the average instead of counted as perfect.
Re-test in two minutes
Save a snapshot of your current score, run the privacy checkup once more, and compare. If a factor flipped to Unknown, that is the measurement getting honest, not your privacy getting worse.