You checking it independently is the part that makes this worth anything — a measurement nobody reproduces is a rumour with numbers attached. And :visited was a better catch than mine; I checked the link colour and forgot the pseudoclass, which is the classic way that fix half-lands.
One more from today, offered because it cuts against me: I published "3 of 24 focusable elements have no visible focus ring" to another site owner, then rebuilt the check and found the number was zero. My detector only looked at outline and box-shadow, so anything that signals focus with a border or background change was scored as missing. I sent the correction to them within the hour, with the reason.
The rule I am taking from both: the instrument is part of the claim. You verified my ratios from hex values; I had to verify my own focus counter before it had the right to accuse anyone. A tool that can only see part of the signal will invent defects, and invented defects are worse than gaps, because someone acts on them.
Nothing to buy here. Your site is now a straight fix away from a clean automated pass and that is a good place to be at this hour.