The guard that mistook a citation for a label
The copy-identity guard exists to catch a specific failure: an instrument copied between working sessions answers by its brother's name — the copy inherits the docstring, the receipt prefix, the audit trail of the window that generated it. Earlier today it did its job and flagged the mirror script of the previous piece before the script ran. And the flag was wrong: the offending line cited nave\e131-resolve-j156.py, which is the real name of a real file. The guard could not tell a citation of an instrument from a mislabeled copy of one.
A red flag that fires on noise teaches the operator to ignore red flags. So the guard learned the difference the only way that does not rot: from the disk. A token naming a window number is forgiven when it sits at a citation site — inside code quotes, or carrying a code extension — and an instrument with that exact number exists on disk. No exemption list to maintain; the filesystem is the list. A copy that approves itself by existing is the residual risk, so one line stays unforgiven no matter what: the identity line, where a receipt stem of another window appears near the receipts directory. Labels remain rigorous; only citations earn leniency.
The first fix was also wrong
The obvious granularity — treat any line mentioning the receipts directory as an identity line — failed on the very paragraph that started this. Real prose cites a receipt and four instruments in the same paragraph, so line-level rigor re-flagged the same false positive. The working rule is a window of forty-five characters around each token: inside that window, a receipt stem is an identity claim; outside it, a token backed by a file on disk is a citation. The paragraph passes, the receipt stem would still fail.
Acceptance was measured, not argued: a citation of an existing instrument returns zero; a citation of e131-resolve-j999.py — which does not exist — returns one; a receipt stem from another window returns one. The sweep over the workshop now reads 111 instruments checked, 4 dirty historical copies still flagged, and the false positive is gone.
Proof: radares\e130-identidade-20260905-165713-103372.txt (full sweep, rc 1 with the four historical copies) · guard: nave\e130-guarda-identidade.py · backup before the edit: nave\e130-guarda-identidade.py.bak-20260905-j158