finding: check-role-word.mjs hardcodes its own regenerated baseline size (43) in two prose sites — accurate today, wrong on the next --update

[claude]

Filed by the domain:devx PM seat from PR #9944 (#9915)'s H4 sibling sweep. Named there rather than fixed, per that dispatch's ruling 4.

The defect

scripts/check-role-word.mjs states its own regenerated baseline's size in two prose sites:

• L133 — check-role-word: OK (43 baselined file(s), no new occurrences).
• L144 — ledger untouched — `43 problem(s)`, exit 1

scripts/role-word-baseline.json holds exactly 43 entries today, so both are accurate right now.

⇒ Latent, not live. They go wrong on the next --update, which is the sanctioned operation that rewrites that baseline.

Why file it while it is still correct

This is the identical state check-engine-double-contract.mjs's L206/L214 were in — and L214 had been refreshed to a correct value two hours before PR #9944 measured it stale again. Refreshing a self-describing number does not fix it; it resets the clock.

The class, stated once: a script that hardcodes a figure its own sanctioned regenerator invalidates, with nothing to catch the drift. --update is not a mistake anyone makes — it is the supported remedy the gate tells you to run.

The fix shape that worked next door

PR #9944 established that run-time derivation is not available for comments — nothing renders a comment, and making --update rewrite the script's own source would be a behaviour change (and would flip Clause-② to yes).

What worked instead: rest the claim on something the ratchet already guarantees, so the sentence cannot go stale in the direction that matters. Its examples:

• 308 generated rows → "a census that ALREADY OUTNUMBERS them and, by the opposite polarities below, can only outnumber them further" — monotone-true given grow-only/shrink-only invariants;
• one problem per census row -- hundreds of them, and growing, since this ledger is grow-only — a floor on a grow-only ledger cannot rot.

⚠️ check-role-word's baseline is shrink-only (it is debt being paid down), so the polarity is the opposite one and the phrasing must follow its invariant, not that file's. Do not copy the sentence; copy the method.

⚠️ Both sites are sample output inside comments — text a reader compares against what the gate actually prints. Check whether they are load-bearing as examples of format (in which case the number could go generic) or as measurements (in which case they need the treatment above). Those want different fixes.

Scope

• scripts/check-role-word.mjs only. Prose/comment sites.
• ⛔ No verdict, exit code or population change — that would flip Clause-② and bind the entry gate.
• ⚠️ PR fix(gate): check-role-word's green line reports what it READ, not just the ledger #9931 (finding: check-role-word's green line counts only the LEDGER, so when the debt is finally paid a total scan failure and a clean repo print the same OK #9910) merged into this file this morning, adding the per-root Scanned: line and a fix to the --update confirmation. Read that first: it changed what the success branch prints, so the sample output in these comments may already be stale in a second way the sweep did not test for.

The bound on how this was found

PR #9944's sweep enumerated scripts/ files carrying a --write/--update/--fix mode (11), resolved each one's artifact, computed its array lengths, and grepped the script's comments for those values. Four other regenerators came back clean: check-error-status-conformance (34), check-i18n-coverage (12), check-query-options-erasure-ratchet (29/17), check-slot-lookup-ratchet (25).

⚠️ That method only catches a figure still EQUAL to the live size — it finds latent sites and misses already-stale ones. In its author's words: "it would not have found this card's own 308s." So a clean result for those four means "no latent site", not "no stale site". Finding already-stale siblings needs the per-file hand classification #9915 used, and is not this card.

Refs: PR #9944 / #9915 (source, method, and the fix shape) · PR #9931 / #9910 (this file, same morning) · #9943 (the sibling class this card is not: present-tense population sizes that --write does not invalidate).

---

Generated by Claude Code