# Audit: `research/RESEARCH-EXECUTION.md` §A2 — the reach of (4.9) in the fixed-endpoint obligation **Caveat first.** One sentence in one paragraph changes, and only by insertion. No estimate, no identity, no obligation, no status and no number in this document is touched: §A2's lower-bound target `B>=-(C2-c0)x+o(x)`, the stronger D-margin `B+2C2*M>=-4x/25+o(x)`, board row 57, the front-matter verdict, the D target `7/200` and every other row are unchanged and remain OPEN. The change carries a restriction the record already states in four other places, and it arrives by way of an adversarial re-check of return #151 (filed as job #644's explore return), which corrected the same dropped qualifier inside `fixed-endpoint-discrepancy.md` and did not reach this document. ## Issue **§A2, lines 116–118 of the served file.** The paragraph is the assignment board's description of the remaining fixed-endpoint obligation. Its own first sentence names the object: *"Independent alternative: B is (2.9) of fixed-endpoint-discrepancy … A lower bound B>=-(C2-c0)x+o(x), c0>0 on unbounded dyadic scales, would suffice. The stronger D-margin requires B+2C2*M>=-4x/25+o(x); these are not equivalent."* Board row 57 gives the same object: *"Exact B=(Type II)+(band), signed bound OPEN."* The next sentence then offers the fixed-endpoint absolute statement without that restriction: > The absolute one-class statement (4.9) is one stronger sufficient input, not a necessary > obstruction. (4.9), in the note it comes from, is the unsigned all-modulus statement at level `2x^(1/2+eps')(log x)^(3L)` with `tau(q)^3` weights. It pays the band's BV error `E_BV^band = 3 log x sum_(q<=Q_1) c(q) D(q)`, whose main term is `O_A(x log^-A x)` by (3a.9) and Abel summation, so it makes the band piece `P_band` negligible. It does not reach the below-level Type II piece `T_II^low` of (2.7), whose moduli read as a progression sum reach `x^(1-eps'/3)`, beyond (4.9)'s level, and whose weight is signed. Since `B` is `T_II^low + P_band` exactly (2.8), with (4.9) the D-margin still needs `2C2*M + T_II^low >= -4x/25 + o(x)`. So the sentence as served presents (4.9) as an input to an obligation it does not pay. In this document that is not a wording nicety: §A2 is the section that tells a worker what to do next, and the mispriced route is a direction of work. A worker who takes §A2 at its word targets (4.9) as the way to close A2, and the signed below-level Type II piece is still open afterwards. **Revised** (insertion only; the note's own accepted phrasing, in this document's own notation, `C2` not `C_2`): > The absolute one-class statement (4.9) is one stronger sufficient input **for the band piece > P_band only, not a necessary obstruction: it leaves the below-level Type II piece T_II^low of > (2.7), so with (4.9) the D-margin still needs the signed 2C2*M+T_II^low>=-4x/25+o(x)**. ## Why this document and not the record as a whole The restriction is already carried in four independent places, so §A2 is the outlier and not a competing reading: - `research/fixed-endpoint-discrepancy.md` (served, `21dce4f3…`): front-matter verdict, §4.4 and §8, as corrected by return #151 and accepted in review 60 — "a stronger sufficient input for the band piece P_band only … it leaves the below-level Type II piece"; - `research/OUTCOMES.md` line 2573: "one stronger sufficient **band** input"; - `research/research-round-validation.md` §12a: "a stronger sufficient all-modulus input **for the band**"; - `research/SEARCH-CONVENTIONS.md` line 39: "one stronger sufficient input **for the band** in the modulus arrangement". A sentence-level scan of nine served documents (`scan-49.mjs`, sha256 `35eada22…`; output `scan-49.out`, sha256 `22275a93…`) finds seven sufficiency sentences that name (4.9) and restrict it to the band, and one that does not: this line. ## Falsifier A derivation of `2C2*M + T_II^low >= -4x/25 + o(x)`, or of `T_II^low = o(x)`, from (4.9) together with the accepted inputs. That is also return #151's own falsifier for the four places it repaired: if it were satisfiable, the restriction would be unnecessary everywhere, including in the note, and this revision would be a change for the worse. The cheaper falsifier is a served document carrying the same unqualified sentence outside the nine the scan names. ## Not changed - The author's phrase "not a necessary obstruction", which the proof note's parallel sentence renders "not a necessary condition". Aligning the two words is defensible and is not part of this audit: it is a wording preference in the same clause, not the reach claim, and a reviewer who wants it can ask for one further line. - `research/history/CHANGELOG.md` line 24 carries "(4.9) is only a stronger sufficient input", unqualified, in the entry recording the 2026-09-09 review. A dated changelog entry states what was written on that date; rewriting it would destroy the record of what changed when. Flagged here rather than changed, and flagged as an `also_fix` candidate if the project's convention is that the changelog is a live summary instead. - Everything else in the document was checked against its owning record and stands: board rows 57–62 cite review sections 2,7 / 12 / 9 / 3,8 / 10 / 4,11, and each is the section that owns that lane in `research-round-validation.md`; D's target moment saving is `7/200` per that record's §10 and §13; E's aggregate constant 4 and the two failing tests for every fixed `u>4` match §11. ## Revision `research/RESEARCH-EXECUTION.md`, served sha256 `43e7c897cd6588793f09bafc07139eca19a6336db63a3fe32645cc6d3ea334d9` (270 lines), revised sha256 `49fc27ded161bca49b59a2cbf3b04ade962cb3d36b6ae0294e9bb5f95d48fca2` (272 lines). Unified diff in `patch`: one hunk, 10 changed lines. Checked as a **pure insertion** by `make-exec-rev.mjs`: all 43 tokens of the served block survive in order, and the 26 leftover tokens are exactly the inserted clause. No clause is dropped, no number is written, the wrapping of the two following lines changes and their words do not. ## Sources - `research/RESEARCH-EXECUTION.md`, served, sha256 `43e7c897…`, §2 (board, line 57), §3 §A2 (lines 111–120). - `research/fixed-endpoint-discrepancy.md`, served, sha256 `21dce4f3…`: §2.4 (2.7), §2.5 (2.8)–(2.9), §4.3 (definition and pricing of (4.9)), §4.4, §8; the admitted correction is return #151 (audit, accepted at `verified`, `duplicate_of` #97), whose reasoning is return #96 (job #233, recorded and unreviewed). - `research/research-round-validation.md`, served, sha256 `72067492…`: §10, §11, §12, §12a, §13; section list used for the board-row check. - `research/OUTCOMES.md` (sha256 `78c5ea9f…`) line 2573; `research/SEARCH-CONVENTIONS.md` (sha256 `61601140…`) line 39; `research/history/CHANGELOG.md` (sha256 `fd50259a…`) line 24. - Prior art: none. This is a documentary correction inside the project's own record, and it claims nothing mathematical. ## Limits - The audit's own claim is a reading of one served paragraph against the record's four other statements of the same restriction. It is verified as a documentary fact (the sentences and their hashes are in the outputs), not as a theorem. - The scan that found the line covers nine named documents; it is not a sweep of the served corpus, and the document list is in the script. - One bug of mine, kept in the record because it is the reason the scan is trustworthy: the first detector tested a three-line window for the word "band" and reported this sentence as band-qualified, because the previous sentence ends "the remaining Type II/band coefficients". `scan-49-window.mjs` reproduces that wrong verdict next to the corrected one.