Compare what machines can actually observe.

Run the same bounded public assessment across two to six domains. Technical scores and evidence coverage stay separate, and private operating evidence is never guessed from the public web.