# Evidence for job #3509: regenerate `research/QUESTIONS.md` (28 open findings) | what | sha256 | bytes | |---|---|---| | served text (base of this revision) | `d47cc818ba743a989de21c32f99fa948c530f9320d25426c57a0438e50424b29` | 602076 | | regenerated by the served generator (path-order fallback) | `d29064ff67de7ee14152` | 615029 | | **revised text (this return)** | `20d8389314c466b6c76feb99c31732e5a86b5a8a71a969167c88f614d7a6aa61` | 613606 | ## What was run, verbatim 1. `GET /docs/research/QUESTIONS.md` -> `d47cc818ba743a989de21c32f99fa948c530f9320d25426c57a0438e50424b29`, 602076 B. 2. `GET /docs/research/{qc.js,gen-questions-index.js,qc/questions.js,qc/corpus.js}`. `qc/questions.js` is the served `1d2d785e...` finding #137 names. 3. Mirrored the exact file set `indexedFiles()` reads - `research/history/staging/*.md` (442) plus `research/.md` top level (141), plus root `TODO.md` - **583 files, 0 fetch failures**, into `work/mirror/`, and copied the served scripts in beside them. 4. `cd work/mirror && node research/gen-questions-index.js` -> `554 questions from 581 indexed notes, 0 unindexed`. The served file carries 554 rows too, and the regeneration adds and removes no id: the whole difference is 44 replace blocks - the section-0 note (4 -> 8 lines) and 43 row lines covering **28 ids**. ## Why the regeneration is only partly decidable here, and which part `research/qc/questions.js` chooses a shared id's displayed verdict by **last-commit date** and, in its own words, "A corpus without git yields an empty map, and then summarise falls back to path order, exactly what it did before this rule existed". The docs tree has **no `.git`**, and `MIRROR.md` says why: *"This is a filtered public edition ... prepared from private commit `d0cef20` on 2026-09-16 ... History and the unmodified working material stay in the private repo."* Finding #137 anticipates exactly this: *"Regenerate in a full-history git checkout and say so in the commit: without git (a /docs mirror, shallow clone or squashed export) the verdict column falls back to path order. ... registry-drift reports the 17 changed shared-id rows."* Measured, and matching #137's number exactly: of the 28 changed ids, **11 are singletons** (one note carries the id, so no ordering rule can matter - the generator's row is the authoritative one) and **17 are shared, and all 17 have records that disagree on the verdict**, so each of their rows depends on the date rule this environment cannot supply. So this revision applies the generator's output for the ordering-independent part and nothing else: the 11 singleton ids in every layout they appear in (23 row blocks) plus the section-0 note. The 17 shared ids (26 row blocks) are left exactly as served, and are the named open obligation. ## The 11 applied rows (each equals `node research/gen-questions-index.js` output verbatim) - `Q-centered-discrepancy-measurement` - `Q-head-remainder-0830` - `Q-kstar-prereg` - `Q-oeis-seam-submission` - `Q-recon-0904-sifting-limit-floor` - `Q-roughpair-error` - `Q-structured-dispersion-estimate` - `Q-var41` - `Q-varE-identification-0830` - `Q-xchan-at29-prereg` - `Q-xchannel-offset` ## The 17 left open (generator disagrees, but the choice needs git) - `Q-ab-coupling` (2 records) - `Q-c2prime-drift` (2 records) - `Q-f-census` (2 records) - `Q-foldL-window5` (2 records) - `Q-greedy-oracle` (2 records) - `Q-import-hypergraph` (2 records) - `Q-kk-substitution` (2 records) - `Q-null-limsup` (2 records) - `Q-obstruction-audit` (2 records) - `Q-quadpoint-transplant` (3 records) - `Q-record-deficit` (2 records) - `Q-records-placement` (3 records) - `Q-redteam-0821-wave2` (2 records) - `Q-rho-maximal-law` (2 records) - `Q-rho2-bound` (2 records) - `Q-shadow-amplitude` (2 records) - `Q-varE-limit` (8 records) The named rows the findings ask about that are **immutable here**: Q-xchan-at29-prereg (#122, #195), Q-recon-0904-sifting-limit-floor (#2584), Q-xchannel-offset (#2588), Q-oeis-seam-submission (#2589), Q-roughpair-error (#2590, #167, #169), Q-structured-dispersion-estimate (#790, #777, #786), Q-kstar-prereg and Q-var41 (#127), Q-head-remainder-0830 (#153), Q-centered-discrepancy-measurement. **#209's Q-varE-limit is shared (8 records, 8 distinct verdicts)** and is therefore among the 17 left open rather than applied. Nothing here is hand-edited: every changed line is byte-identical to the generator's own output for a row whose value cannot depend on the ordering rule, and `edit_bp.py` asserts that no shared id moved.