Joseph Ferano
b5d7a6e45f
BUILT.md's locals section said the second whole-frame refusal was a slot count mismatch. It is a fingerprint, and the paragraph now says why a count could not have done the job: the case it exists for is a rename, which changes neither the count nor the types. It also states the bound honestly — a 30-bit hash can collide, and a collision would reproduce exactly the wrong answer this catches, but only between two differing bodies of a function whose name already matched. NEXT.md's item 1 is struck, and the handoff paragraph that diagnosed this is marked wrong rather than deleted. It claimed every piece was written and one of five hand-offs was dropping the number; four were never written. The step it recommended first could not have found that, and a lane stopping mid-repair should say which pieces it ran rather than which it believes it wrote.
