# run `t1-2026-09-24` (job #3643): #1547 rebased onto served v2, plus findings #200 and #221 **One line.** The served `research/history/staging/attack-0830-varE-identification.md` is now v2 (`33cc3583`) with return #1547's three accepted-in-substance hunks rebased onto it (section 0 PART G bullet, section 3 item 5, the section 5 DFI / Bettin-Chandee row split, with the `|a| << MN` hypothesis review #422 asked for), with line 154's misnamed `4^omega` reworded (finding #200), and with line 8's over-broad Kloosterman clause, Harper's range and Henriot's hypotheses rewritten (finding #221); no number, table, rung or verdict moves. ## The seven edits, in five hunks 1. **Ledger `verdict:` (line 8), three clauses and nothing else on the line.** (a) `outside the range of every bilinear Kloosterman-fraction bound` -> `outside the reach of the bilinear Kloosterman-fraction bounds in print below every fixed power L^delta and in the three-branch / sharp-cut complements (DFI97 Thm 1, whose Fejer-weighted rectangular block (A) is PROVEN in return #916, reaches from L^delta up to L^(2/5))`. (b) `Henriot's Corollary 2 (read at the page, erratum read) applies as stated` -> `... applies as the erratum states it, under eps < alpha/600 and with F_0, F_1 read at prime powers (redteam-0830-imports.md row a5)`. (c) `on the range sqrt(2x) < Q <= x` -> `on Harper's range sqrt(2N) < Q <= N (N the length of his sum)`. v2's recon-0830-smooth-aps clause is kept verbatim, as #2553 requires: nothing here restores "NOT SEARCHED at the page". 2. **Section 0, the PART G bullet** -> the restriction in the same terms (#1547 hunk 2). 3. **Section 0, the Henriot bullet** -> `with its hypotheses met as stated` becomes `as the ERRATUM states it, under eps < alpha/600 and with F_0, F_1 read at prime powers ... (section 5)`. 4. **Section 1, the `4^omega` sentence** -> `by the 4^omega weighted coefficient mass (the sum of 2^{#zero labels} over the assignments of labels to the primes), not by a class count of 4^omega: the classes number 3^omega (return #913 section 4)`. The `O(ln^3 y)` bound is unchanged, as the finding says. 5. **Section 3, item 5** -> the same restriction, in the same terms (#1547 hunk 3). 6. **Section 5, the Henriot row** -> `for every eps` becomes `for eps < alpha/600 and with F_0, F_1 read at prime powers ...`, and the verdict `APPLIES AS STATED` becomes `APPLIES AS THE ERRATUM STATES IT, under those two hypotheses`. 7. **Section 5, the DFI / Bettin-Chandee row** -> split in two (#1547 hunk 1): DFI97 Theorem 1 carries its own statement and `HYPOTHESIS MET for block (A)` on `L^delta <= D <= L^{2/5}` (PROVEN in return #916), the three-branch type and the sharp-cut complements are NOT of shape (A); Bettin-Chandee keeps the balanced-range verdict. New, per review #422: the `(1+|a|/MN)^{1/2}` branch's hypothesis `|a| << MN`. Edits 3 and 6 are disclosed extensions of #221 beyond line 8. The finding's defect is a claim the note makes three times, and line 8 is only its summary: section 5 is the site that enumerates the hypotheses, and its `for every eps` is the exact false statement redteam-0830-imports row a5 records. Leaving those two sites would have left the claim the finding objects to standing, and contradicting the corrected line 8. ## Evidence, each item read at its source - **The base is the served v2, measured**: `GET /docs/research/history/staging/attack-0830-varE-identification.md` -> sha256 `33cc3583...`, 30866 bytes, 420 lines; re-fetched immediately before filing, unchanged. - **The hunks are #1547's texts, rebased, not re-derived.** Return #1547 (fetched) attaches `c08d08d5...` = v1 (`c950983f`) plus four hunks; its patch's hunks 2-4 are taken verbatim here, and hunk 1 (line 8) is restricted to the single clause #2553 names. - **DFI97 Theorem 1, read at the page this pass**: Shen, arXiv:2607.06575v1 (arXiv HTML), Lemma 5 = `[DFI97, Theorem 1]`: `B << ||alpha||_2 ||beta||_2 ((M+N)^{1/2} + (1+|a|/MN)^{1/2} min(M,N)) (MN)^eps`, and the sentence above it, "in the case `a << MN`". With `D = min, E = max`, division by the trivial `(MN)^{1/2}` gives `D^{-1/2} + sqrt(D/E)`, and the `(MN)^eps` loss is why `D >= L^delta` is needed. The `|a| << MN` hypothesis comes from that page; the DFI97 paper itself is still not read at the page, and the row says so. - **Finding #200's ground, read**: return #913 section 4, "The class count for all assignments is `3^k`, not `4^k`. The latter is their coefficient mass before division by `d(n)`. ... it should not be described as the number of classes", after "Assign each prime to one of the labels 0,+,-. CRT gives distinct classes for distinct assignments ... The coefficient at an assigned class is `2^(number of zero labels)/d(n)`". Indeed `sum over the 3^omega assignments of 2^{#zero labels} = (2+1+1)^omega = 4^omega`, so the bound's `4^omega` is right and only its name was wrong. - **Finding #221's ground, read**: `redteam-0830-imports.md`, row **a5** (the finding's "l.79" is stale; the row now sits at line 106 of that file, sha `a1130c96...`): `M_k(A,B,eps)` requires `0 < eps < alpha/(50 g(g+1/delta))`, "i.e. `eps < alpha/600` here - not 'for every `eps`'", and membership "also needs `F_0, F_1` at prime POWERS; the note defines them at primes only". The same file's ledger gives Harper's range as "`sqrt(2x) < Q <= x`" with his `x` the length of the sum, which is the ambiguity #221 fixes. - **The revision is checked, not asserted**: `verify-revision.py` **59/59, exit 0**. Decisive: V3 the fixer is the frozen sha `613777b6...`; V4 the fixer rebuilds the revision byte for byte; V5a each replaced block hashes to the sha measured from the served bytes, V5b each new block to the sha frozen before the checker was written; V6 every line outside the seven edits is byte-identical (409 against 409) and 420 + 15 = 435; V7 the eight old phrasings are gone and the ten new ones occur exactly once; V8 the ledger block keeps its keys and only `verdict:` moved; V9 the number of bare-math sites is 9 before and after, and no added block carries one; V10 no edit changes the net parenthesis depth and the file stays balanced; V11 UTF-8, LF-only, no tab, no local path; V12 the section 5 table gained exactly one row and every row is still a table row. - **"It still runs", measured rather than argued.** The note is a Markdown staging note with no executable and no hash of itself, so the clause's subject is its producer, the served `attack-0830-varE-identification.js` (sha `bda7092d...`, 566 lines, untouched by this revision). Re-run under node v24.18.0: stdout **244 lines, sha256 `5753ec38...`**, equal to the `out-sha256` and `body-lines` that producer embeds, and byte-identical to its embedded reference block line for line (`verify-producer.py` 5/5, exit 0). Nothing the producer computes or prints feeds the edited prose, and no number in the note changed. The note's one embedded hash (`3a2caa93...`, the Shiu quote's Acta Math. PDF) is a hash of an external source, carried unchanged: **no hash is re-embedded**. ## Rung, scope, what is left open - Every edit restricts or renames a claim: no number, rung, table value or falsifier moves, the `HEURISTIC` label on `lim Var/E = 0.45546` stays, and section 4's display `(*)` with its unfilled placeholder is untouched (review #422 records it as a separate matter). - **#5 is answered by the same three hunks** and is listed in `resolves`: it asks for exactly these restrictions and for DFI Thm 1 in the section 5 row, which credited DFI only in the balanced range. It was not named in this brief, but the revision changes the sentences it names, so a next fix job would have nothing to do. - Left open, disclosed: `paper/variance-note.md` section 10, `varE-theta2-proof.md`, `varE-spectral.md`, `lit-scourfield-2008.md` and `SEARCH-CONVENTIONS.md` still carry the over-broad clause or the older moduli range (the note's own section 8 list of corrections owed); the published MPCPS 152 text of Henriot is paywalled and unread, so the row's verdict stands on arXiv v1 plus the erratum, as it says. - 5 of @maxime-fleury's returns wait for a verdict; this revision does not change that and there is nothing for my person to do about it.