--- name: an-entrys-quoted-figures-can-come-from-a-looser-meter-than-its-prescribed-method description: "A register entry can PRESCRIBE one pipeline and QUOTE figures taken with a looser one; the disagreement survives review because the entry's conclusion is right, so nothing ever presses its numbers" metadata: node_type: memory type: feedback originSessionId: f46cd010-f436-4ce1-821b-68e7e869fea2 modified: 2026-08-29T10:09:07.157Z --- **Measured 2026-08-29, building IR-57's `xtask pick-audit`.** The entry prescribes hashing `git show --format= | grep -E '^[+-]' | grep -vE '^(\+\+\+|---)'` and quotes **310 lane / 308 pick** for its motivating defect. My implementation of that exact method returned **302/300**. Neither instrument was broken: - `^[+-]` alone over `b2194aa9` = **310**; with the header-excluding second grep = **302**. - The commit touches 4 files; 4 × (`---` + `+++`) = **8**, exactly the gap, on both sides. So the entry's **figures were taken with the first grep only** — the meter its own prose rejects. **Why it survived every reading, and this is the transferable half:** the 8 header lines cancel, so the DELTA (2 lines lost) is identical under both meters and the entry's conclusion is untouched. A figure that supports a correct conclusion is never pressed. The disagreement is only reachable by someone who runs the prescribed method and *compares the number*, which is exactly what implementing an entry makes you do — and the temptation there is to report your own figure and move on, or worse, to assume you got it wrong and go hunting your own code. **How to apply.** When you build the instrument an entry prescribes, reproduce the entry's QUOTED FIGURES as a leg, not just its method. If they disagree: run the entry's literal pipeline in both forms before saying anything — the discriminator is usually one grep — then hand the author the reconciliation (both numbers, the mechanism of the gap, and whether the conclusion moves), never a bare correction and never a silent substitution. Report the delta's survival explicitly: "your conclusion stands, your cited numbers do not" is a different message from "your entry is wrong", and only the first one is true here. Distinct from [[derived-figure-without-its-formula-dies-with-the-context]] — there the formula is lost, here the formula is written down and the numbers simply were not produced by it. Kin: [[dont-take-a-diagnosis-as-measured]] (an entry wears a measurement's clothes too), [[identical-readings-across-opposite-outcomes-indict-the-meter]] (vary the instrument), [[a-coverage-fix-is-measured-over-the-population-it-claims]].