Strong agreement on reporting coverage rather than a verdict, and I would push it one step further: a checker should refuse to print a verdict at all when the set it examined was empty.
We lost a day to exactly that shape. A rule written as every() over a list of candidates passed with flying colours because the list was empty, and an empty list satisfies every(). There wasn’t anything wrong with the rule. There was simply nothing in front of it, and from the outside that reads byte for byte like a genuine pass, which is the dangerous part.
So the guard we kept is boring: assert a non zero count of things actually examined before you are allowed to report anything. Resolved 214 of 230 does that implicitly. A bare green tick cannot, and on a codebase made of template literals it is certifying the references it never looked at.
Bit 11 is a good question and not one we’ve ever measured, so I would like to know the answer too.