# Ledger extraction: the project's data is out of the QC engine **Headline.** All three suppression ledgers now live in `research/qc/ledgers.js`; the fast gate and the selftest are byte-identical to their pre-change output, and all 25 entries were verified verbatim against `git HEAD` by key, order and reason text. ## What moved | Ledger | From | Owning check | Entries | |---|---|---|---| | `INTENDED_MISSING` | `qc/corpus.js` | `refs` | 3 | | `ADJUDICATED` | `qc/checks.js` | `transfers` | 13 | | `ABSENCE_VERIFIED` | `qc/checks.js` | `absence` | 9 | Each moved with its explanatory comment. The `transfers` docstring, the `INTENDED_MISSING` "why this is in code and not in prose" paragraph and the absence-claim TRAP paragraph all travelled with their data. What stayed in `checks.js` is only the *mechanism* by which each suppression lapses, which is engine rather than data. `checks.js` and `corpus.js` reach the ledgers through `L.*`; `corpus.js` no longer exports `INTENDED_MISSING`. ## The gate, before and after Captured before any edit and re-run after. Diffed with timings normalised; **no other difference**. ``` BEFORE AFTER refs 0 finding(s) refs 0 finding(s) quotes 0 finding(s) quotes 0 finding(s) crosslinks 0 finding(s) crosslinks 0 finding(s) scripts 0 finding(s) scripts 0 finding(s) transfers 0 finding(s) transfers 0 finding(s) calibration 0 finding(s) calibration 0 finding(s) absence 0 finding(s) absence 0 finding(s) sourcing 0 finding(s) sourcing 0 finding(s) TOTAL 0 TOTAL 0 ``` The two description strings that carry the ledger counts are unchanged, which is the part that would have moved had an entry been lost: - `transfers` — `(13 adjudicated KEEP and suppressed; they return if either side is edited)` - `absence` — `(22 absence phrases, 9 verified and dated, 0 unverified)` Selftest, before and after, `diff` clean: ``` All 15 known positives fire and all 7 controls stay silent. A zero from qc.js means something. ``` ## Calibration: the ledgers are load-bearing Per the standing rule that a zero proves nothing until the instrument is checked against a known positive, `ledgers.js` was moved aside and the gate re-run: ``` refs 0 → 4 transfers 0 → 13 absence 0 → 9 TOTAL 0 → 26 ``` `refs` returns 4 rather than 3 because one exempt path is cited from two sites. So the wiring is live, the suppressions are real, and losing the file **returns** findings rather than hiding them, which is the safe direction. ## Judgement calls, not mechanical moves **1. `INTENDED_MISSING` moved, and it was the one real decision.** The case for leaving it in `corpus.js` is that it is keyed on repository paths and `corpus.js` owns paths. That is not the right cut. Everything else in `corpus.js` is true whether or not a check ever runs — `ROOT`, the walk, the generated set, the shorthand grammar. `INTENDED_MISSING` has no meaning outside the `refs` check: its own docstring calls it an allowlist earned entry by entry, which is an adjudication, not a layout fact. `ABSENCE_VERIFIED` is also keyed on a repo path and nobody would call that config. The practical test is porting: `corpus.js` is **rewritten** for a new project, `ledgers.js` is **emptied**, and leaving one ledger behind would make a porter edit `corpus.js` for two unrelated reasons and plausibly carry this project's exemptions into a corpus nobody had read. **2. `ledgers.js` is loaded tolerantly, and this was forced.** `selftest.js` builds its fixture by copying exactly `['checks.js', 'corpus.js']` into a temp root, so a plain `require('./ledgers')` throws there. `selftest.js` was outside this task's ownership, so `checks.js` falls back to three empty maps when the file is genuinely absent. That is correct in both places: no fixture passage carries a real fingerprint or a real `file|phrase` key, so an empty ledger suppresses nothing there and changes no case — confirmed by the identical selftest output. A `ledgers.js` that exists but throws is still a hard error. **The tidier fix is one line in `selftest.js`**: add `'ledgers.js'` to that copy list and drop the `fs.existsSync` guard. Worth doing when that file is free. **3. The README's porting claim was weakened deliberately.** The obvious edit was "nothing project-specific is left in the engine". That is false, so it says instead that the project's *data* is out and names the three config literals still in `checks.js`: the top-level directory names `research|paper|web|attestation` in `PATH_RE` and in the relative-path guard, the `ENTRY` set used by `crosslinks`, and `INTENTIONAL` in `transfers`. `MARKER` and `SETTLED` are argued as a deliberate exception rather than an oversight: a project that does not adopt that calibration vocabulary has not adopted the check. ## COVERAGE - **Not run: `node research/qc.js --full`.** The brief reserves it. So `audit-numbers.js` has not been run against this change. The risk is close to nil — nothing here touches a number, a computation or a `.js` under `research/` outside `qc/` — but it is unrun, not clean. - **The three config literals in point 3 were found by grep, not by an exhaustive read of `checks.js`.** There may be a fourth. The claim in the README is scoped to what was measured. - **Where I am most likely wrong: the `fs.existsSync` guard.** It is a silent fallback, and silent fallbacks are how blindfolds get installed. It is defensible here only because the failure direction is un-suppression, and it should be deleted as soon as `selftest.js` can be edited. - **Entry content was not re-adjudicated**, by instruction. The nine `ABSENCE_VERIFIED` entries carry a standing re-verification duty that this move does not discharge; the two dated 2026-08-18 are the freshest, the seven dated 2026-08-17 are the ones a new wave would stale first.