# W1 is the Hardy–Littlewood singular series in disguise: the identity is exact, and the comb now has a canonical name *(2026-08-21. Staging note; nothing here is integrated into a live document. Producer, formally embedded: `research/attack-w1hl-01.js` (0.2 s, no sieving; per-fold W1/θ table parsed from `attack-mp-derive-01.js`'s committed embed, code-sha256 248c8c88…, custody-pinned before use). Calibration marked on every claim: PROVEN, VERIFIED by exact computation, MEASURED, PREDICTED, ABSENT-WITH- CONVENTION. Salvage note: the producer is the orphaned predecessor's script, audited here — its embed verified bit-honest by `embed.js --check` (code, body and out-sha256 all match), its ν-case table independently confirmed by the red team's own 182/182 brute enumeration (`redteam-0820-night-empirical.md` §T1.a), and every figure its READINGS quote appears in its OUTPUT block.)* ## 0. The answer `mp-derivation.md` §1 derived the endpoint comb W1(v) = ∏_{q|v}(q−2)/(q−4) · ∏_{q|v±2}(q−3)/(q−4) over folded primes 5 ≤ q < p from the forbidden set {0, −2, −v, −v−2} mod q, and flagged NOT-REACHED on its prior-art position. The flag closes: > **W1 is exactly the Hardy–Littlewood singular-series ratio > 𝔖(0,2,v,v+2)/𝔖(0,2)², truncated to the folded primes and normalised by a > v-independent constant.** Per prime q ≥ 5 the HL local-factor ratio > r_q(v) = (1 − ν_q/q)/(1 − 2/q)² divided by its generic value > q(q−4)/(q−2)² equals W1's factor as an identity of rationals; over any > cutoff P and any v with 6 | v, > ∏_{5≤q q = 2, 3 included the full-series form is > **𝔖(0,2,v,v+2)/𝔖(0,2)² = 6 · C_∞ · W1_∞(v)** on 6 | v (𝔖₄ = 0 off it, > and every θ_p has 12 | θ, so the fold field never leaves the domain). > **[PROVEN algebraically + VERIFIED as exact BigInt-rational equality]** So the fold-factor field's comb half, M_p ∝ W1(θ_p), is the HL **twin-twin correlation comb**: the conjectural enhancement of two twin pairs at distance v over independent twins, evaluated at v = θ_p. The quadruple (0, 2, v, v+2) is two twin pairs at distance v, and the forbidden set of `mp-derivation.md` §1 is that quadruple's local condition, prime by prime. ## 1. What the producer verifies, exactly All in `attack-w1hl-01.js`, elementary arithmetic, BigInt rationals, nothing re-sieved: - **ν-case table** [VERIFIED]: ν_q(0,2,v,v+2) = 2 iff q | v, 3 iff q | v±2, 4 otherwise, brute residue count vs the case formula on 1147 (q, v) pairs, q ∈ [5, 97], classes mutually exclusive; q = 3 gives ν = 2 iff 3 | v and ν = 3 (𝔖₄ = 0, inadmissible) otherwise; q = 2 gives ν = 1 on even v. - **Per-prime identity** [VERIFIED, exact]: r_q/gen_q ∈ {(q−2)/(q−4), (q−3)/(q−4), 1} for all 167 primes q ∈ [5, 1009]; the q = 2, 3 factors contribute the v-independent constant 2 · 3 = 6 on 6 | v. - **Field custody** [VERIFIED]: the HL-ratio path (independent brute-ν route) equals the recorded divisibility-comb W1 as exact rationals at 12 folds spanning the field (101 … 709; e.g. W1(θ_211) = 128/21, W1(θ_631) = 2380/429) and reproduces ALL 102 embedded per-fold W1 values of `attack-mp-derive-01.js` to the printed 2 decimals (worst |dev| = 0.0050, printing precision, at p = 251). - **Truncation pricing**: W1's cutoff at q < p differs from W1_∞ only in primes q ≥ p dividing θ(θ² − 4) — q = p itself always (p | θ ∓ 2η, factor (p−3)/(p−4)) plus any prime factor > p of p − η ± 1: a 1 + O(1/p) per-fold correction, absorbed by the law's smooth normalisation. ## 2. The Gallagher block: what imports, priced - **Per-prime mean-one** [VERIFIED, exact]: the average of r_q(v) over v mod q is exactly 1 at every prime (q = 2, 3 included, inadmissible classes contributing 0). This is the local skeleton of Gallagher's theorem — singular series average to 1 (Soundararajan, arXiv:math/0605696, p. 9 eq. (3), page image read 2026-08-21; Gallagher 1976 behind it). **Import priced honestly: Gallagher's actual theorem averages 𝔖 over ALL k-element subsets of [1, h] as h → ∞. Our object is a one-parameter fixed-shape family, which his statement does not cover; the per-prime computation in the producer does the whole job for our case and the theorem supplies the convention's name, not the result.** The refined all-subset asymptotics are Montgomery–Soundararajan 2004; structured sums of singular series (large sets; entries in arithmetic progressions) are Kuperberg arXiv:2210.09775 and arXiv:2301.06095 — none of them index the argument by shifted primes. - **Prime-indexing shifts the mean** [PROVEN + VERIFIED, exact]: on the fold field v = θ_p = 2(p − η), the class p ≡ 0 mod q is empty, which kills one of the two q | v ± 2 classes; the per-prime mean over admissible classes is exactly 1 + 3/((q−1)(q−4)) (both η), NOT the integer-indexed (q−2)²/(q(q−4)). Products: E_p[W1_∞] = 2.3252 (prime-indexed) vs E_v[W1_∞] = 2.5197 (integer-indexed). - **The field's mean lands on the zero-parameter prediction** [MEASURED]: measured mean W1 over the 102 embedded folds = 2.2823 (se 0.120) vs the per-fold-truncated prime-indexed prediction 2.3215 — gap −0.33 se. A consistency read, not a powered test: θ_p is deterministic, the se treats folds as independent draws, and the equidistribution of p mod q behind the class weights is Dirichlet, assumed not proven at this scale. - **What does NOT import** [stated so it is not mistaken for coverage]: nothing in the 4-tuple singular-series theory touches δ = 0.277, the depth term e^{−δz}, or the far-tail flattening of `mp-derivation.md` §5. The singular series constrains v's arithmetic only; the depth residual is orthogonal to this identity and stays underived. Second-moment singular-series results would bear on an integer-indexed field variance, not on the measured per-fold residual sd ≈ 0.18. ## 3. Prior-art position (searched per `SEARCH-CONVENTIONS.md`) The owning convention is **singular series for prime constellations / k-tuple local factors / correlations of twin primes**, searched 2026-08-21 on WebSearch + arXiv full texts, channel calibrated in-session (known positives retrieved: Gallagher's theorem, Lemke Oliver–Soundararajan 1603.03720, the Kelly–Pilling trilogy). Position, standard first: - **STANDARD — the singular series and its ν-case computation.** 𝔖(H) = ∏_ℓ (1 − ν_H(ℓ)/ℓ)(1 − 1/ℓ)^{−k} with admissibility 𝔖 ≠ 0 is Hardy– Littlewood 1923 (Partitio Numerorum III, Acta Math. 44, Conjecture X's apparatus), stated in exactly our local-factor form in Soundararajan, arXiv:math/0605696 p. 8 eq. (2) — page image read. Computing ν by cases on which primes divide the tuple's difference products is standard in print: Goldston–Ledoan, arXiv:0910.2960 §4 p. 6 (page image read) do it for the triple {0, d′, d} via Δ = d′d(d − d′), "If p ∤ Δ … ν_D(p) = 3", and their eq. (2.2) p. 3 is the pair comb 𝔖(d) = 2C₂ ∏_{p|d, p>2}(1 + 1/(p−2)) — the k = 2 analogue of W1's q | v factor. **Nothing in §0's algebra is claimable; it is a two-line instance of 1923-vintage theory.** The corpus's recorded W0 = ∏_{q|θ}(q−1)/(q−2) is that eq.-(2.2) pair comb — right divisor classes, wrong tuple. - **STANDARD — the quadruple's interpretation.** Under the HL conjecture the number of twin pairs at separation v up to x is 𝔖(0,2,v,v+2) · x/log⁴x, so 𝔖₄(v)/𝔖₂² is the twin-twin pair correlation at distance v. The ratio normalisation by 𝔖² is the standard move of the consecutive-tuples / bias literature (Lemke Oliver–Soundararajan, arXiv:1603.03720, for primes). - **NOT FOUND — the explicit W1 comb on twin-pair separations.** The written form ∏_{q|v}(q−2)/(q−4) · ∏_{q|v±2}(q−3)/(q−4) as the twin-twin correlation, or any divisor-modulated law for gaps between twin primes, was not found; searched in the owning convention (twin-prime pair correlations; gaps/separations between consecutive twin primes; LOS-bias analogues for twins) per `SEARCH-CONVENTIONS.md`. Three adjacent works checked at full text: **(i) Kelly–Pilling, arXiv:hep-th/0108241v1 (full HTML read 2026-08-21): separations between consecutive twins are exponential, slope m ≃ 1.321/log π₁(N), HL invoked only for the slope's decay — the paper contains NO divisor-dependent factor. The predecessor's negative ("exponential separation law but no comb") is RE-VERIFIED at source.** Same for the trilogy's abstracts (math/0103191, math/0104205, math/0106223). (ii) Sahoo, arXiv:2111.09053v3 (full HTML read): three empirical biases in twin distribution, the third on the difference D of consecutive twin pairs ("D±1 is more likely to be a prime than an odd composite") — purely empirical, 30M pairs, explicitly no formula; an observed shadow in the comb's neighbourhood, no comb. (iii) the math.GM item arXiv:1703.08039 (PDF retrieved, extraction poor): counting methodology, no separation comb found in the legible portions. A negative here carries the usual weight of a one-session literature sweep, not of an exhaustive one; zbMATH/MathSciNet were not swept for this object (see §5). - **OURS — the application.** Using the folded-prime truncation of 𝔖₄(θ)/𝔖₂² as the per-fold field predictor M_p (with the depth term), the anchored reading θ = 2(p−η) tying the comb's classes to the divisor structure of p−η and p−η∓1, and the prime-indexed Gallagher-type mean 1 + 3/((q−1)(q−4)) with its field-level −0.33 se hit — none of it found in print, all of it elementary given the frame; the value is the application to M_p, not the algebra. ## 4. Proposed consequences, for adjudication rather than applied - `mp-derivation.md` §1's "No prior-art search was run on W1 itself; the object is singular-series-shaped and nothing here is claimed as new" can now carry: *searched and settled — W1 IS the HL singular-series ratio 𝔖(0,2,θ,θ+2)/𝔖(0,2)² truncated to the folded primes (exact, `attack-w1hl-01.js`); the algebra is standard (HL 1923; Soundararajan math/0605696 eq. (2); Goldston–Ledoan 0910.2960 §4), the application to M_p is ours; Gallagher-style prime-indexed averaging predicts the comb field's mean at −0.33 se with zero parameters (`w1-singular-series.md`).* - `SEARCH-CONVENTIONS.md` §1 could gain a row: object "the endpoint comb W1 / twin-pair correlation at fixed separation", owning convention **"singular series for prime constellations; correlations of twin primes; gaps between twin primes"**, living at HL 1923 / Soundararajan math/0605696 / Goldston–Ledoan 0910.2960; and §3 a row for the Kelly–Pilling negative (now full-text, twice verified). - The vocabulary gain is free: "the endpoint comb" = "the HL twin-twin correlation at the fold's θ" is the framing-safe way to present M_p's comb half outside the corpus. ## 5. What was NOT reached - **No page image of Hardy–Littlewood 1923 itself**; Conjecture X is carried here through Soundararajan's and Goldston–Ledoan's statements (both at page images) and Goldston–Ledoan's reference [3] to Acta Math. 44, 1–70. - **zbMATH / MathSciNet were not swept** for the twin-separation comb; the §3 negative rests on WebSearch + arXiv full texts only. Wolf's early twin/cousin-gap preprints (1990s, IFTUWr) were not retrieved and are the likeliest place a written comb could still hide. - **No derivation of δ = 0.277 was attempted** — §2 prices that nothing in this import bears on it. - **The prime-indexed mean's se treats folds as independent**; no attempt to price the θ-field's actual correlation structure into that −0.33. ## 6. Reproduction ``` node research/attack-w1hl-01.js # 0.2 s, no sieving node research/qc/embed.js --check research/attack-w1hl-01.js ``` --- *This document states current understanding. Superseded claims, retired numbers and the reasons they changed are in [../CHANGELOG.md](../CHANGELOG.md), indexed by document.*