# Independent review of the reduction chain, 2026-09-06 **Twin-prime infinitude and the sufficient endpoint margin remain OPEN. Nothing here is a new estimate.** This note consolidates the review verdicts of 2026-09-06 so that the reduction chain's review state is recorded in one place. Each review was carried out by a separate agent with fresh context, read-only against the repository, and re-derived the decisive inequalities rather than trusting the validators. The full reports are retained in the session record; their content that matters for reuse is below. Roughly twenty derivation notes were written in about thirty-six hours and integrated the same day. The repository's own record is that four of five same-day integrations on an earlier campaign needed correction within twelve hours. That is why review took two of five slots in this wave before any new estimate was attempted. ## 1. Scope and calibration of "checked" CHECKED means the reviewer followed the written argument line by line, reconstructed the decisive inequality independently, and confirmed each imported theorem's statement against its primary source. It does not mean the note has been refereed by an outside expert. ASSUMED-UNVERIFIED means the reviewer could not or did not verify the step and says so. DEFECTIVE would name a file, equation and line; none was recorded in this wave. The chain under review is, for every fixed H>0, \[ S(x)=\sum_{x/2 (six classical contributions), E_> to E_* ([residual-coverage.md](residual-coverage.md)), and E_* to E_dagger ([grouped-divisor-moment.md](grouped-divisor-moment.md) (19)-(20)). ## 2. Link one: S(x) to E_>, six contributions The checklist is the table in [endpoint-target-audit.md §1](endpoint-target-audit.md). All six rows were adjudicated CHECKED. | contribution | decisive point confirmed | |---|---| | M - C_2 x | b(p^j)=2C_2(1+1/(p-2)) at odd prime powers, zero on evens; quantitative PNT with the classical error gives C_2 x + O_H(x/log^H x) | | I_1 - I_2 | the odd-d main terms match exactly through d/phi(d); D=UV<=x^(12/25) sits below the BV level x^(1/2) with a fixed power margin; the mesh interpolation in prime-detection-spec is required because the imported BV has no supremum over y<=x, and the note supplies it | | -Bcal(b) | the Euler factor identity 1 - p^(-s)/((p-2)(1-p^(-s))) was re-derived; sum of h(e)/sqrt(e) converges uniformly in k and R | | P_1 - P_2 | every congruence kd = 2 (mod l) reduces to a primitive class modulo q in {l, l/2} with multiplicity at most 2, which is what the imported maximum over coprime classes supplies; level x^(1/10) against T^(1/2)>=x^(3/25), margin 1/50 | | R_{<=L} | the excluded-prime Mobius mean has t>=U/2 and m<=x^2, both polynomially bounded, and the lemma is proved uniform in m; the Abel step costs L V Z log^3 x <= x^(99/100) log^3 x | | M_> | integrality of k=n/d forces d<=D_0 and e<=E_0, so W_> contains every original term; the same Mobius mean applies | The four assembly identities S-M = I_1-I_2+B, B = Bcal(a)-Bcal(b), Bcal(a) = P_1-P_2+R, R = R_{<=L}+M_>+E_> are exact. Independent finite checks at x=512 with test cutoffs reproduced all four, and reproduced R_{>L} = M_> + E_> exactly, which exercises the CRT kernel, the sawtooth convention psi(integer) = -1/2 and the incompatible-gives-zero rule together. Finite checks are checks, not proofs. Two hazards were hunted specifically and not found: a fixed-H bound used with H depending on x or another growing parameter, and a residue class depending on the other variable used outside the imported maximum over coprime classes. ## 3. Link three: the grouped moment (1)-(20) All twelve pressure points in [grouped-divisor-moment.md](grouped-divisor-moment.md) were adjudicated CHECKED, with three ASSUMED-UNVERIFIED annotations on upstream inputs. The re-derived items: the equal-frequency count 8 N A H_floor(2 min(N,A)) and its use in (5), including collisions between distinct divisors; the endpoint variation (6) in both cases v<=1 and v>=1; the completion (7) with complete periods removed, in both cases c>M and c to E_*, and the block shape The transfer through [residual-coverage.md](residual-coverage.md) §§1-5 was adjudicated NO DEFECT FOUND IN SCOPE. Its largest dependency, flagged by both earlier reviews, was the claim that the aggregated endpoint remainder has the block shape sum_m A_left(g m) Y(m) of grouped-divisor-moment (1): a separate left coefficient in m, a separate right coefficient in u, and a harmonic weight depending only on h. The reviewer reconstructed the derivation across three notes and verified it exactly: the bijection (d,r) to l=dr in endpoint-fourier (4)-(5); the discrete-Abel endpoint kernel of signed-divisor-grouping (15)-(18); the CRT origin identity n_0/q = (2/g) mbar/n (mod 1) for compatible l=gm, j=gn with (m,n)=1, over 30,550 pairs in both g branches; and the block form itself after the Vaaler replacement, with sigma=-1, theta=2/g and c_h = -W(h/(T+1))/(2 pi i h), to maximum discrepancy 2.1e-14. CRT compatibility (l,j) | 2 becomes 1_{(m,u)=1} exactly because g=(l,j) forces (m,n)=1 and conversely. Also checked: the inverse-reciprocity left orientation reverses sigma and shifts both endpoints by 2, exactly as written in three notes, verified mod 1 over 39,654 cases; the Fejer form of the Vaaler majorant with the bound D_T <= min(1/2, 1/(8(T+1)^2 ||v||^2)), tight at integers so that psi(integer) = -1/2 is load-bearing; the divisor-count bound on the full positive majorant with no pairs at k=0 and k=2; the Perron separation with absolute mass exactly D_0 E_0 V Z = x^2 times log^2 x and error exponent -111/20; the untwisted density on every initial d-segment the cuts produce; the budget table of residual-coverage §3 in exact rationals; the assembly E_> - E_* through the cut {de <= L} contained in A; and that incompatible congruences contribute exactly zero (366 pairs enumerated). Two steps are correct but unwritten in the owning note: the Perron separation must be read box by box, and the two-line reduction of E_> - E_* to pieces covered by residual-coverage (15). One sub-item stays ASSUMED-UNVERIFIED: the right-orientation rewrite of the prime-power and first-branch moment is asserted from the sign-agnostic ingredients rather than written out. ## 5. Imported theorems and one provenance defect | import | source confirmed | hypotheses met | |---|---|---| | quantitative PNT | Tao, 254A Notes 2, Corollary 39 | yes | | Bombieri-Vinogradov for Lambda | Tao, 254A Notes 3, Theorem 17 | yes at Q=x^(12/25); no supremum over y<=x, hence the mesh step | | Vaughan identity | Tao, 254A Notes 3, Lemma 18 | yes | | Mobius mean, q=1 | Tao, 254A Notes 2, Exercise 66 | yes; ineffective | | Mobius Bombieri-Vinogradov | derived in mobius-bv-derivation.md from Koukoulopoulos GSM 203 (Cor. 13.4, Thms 26.2, 26.6, eq. (26.3)); EPFL sheet kept as teaching reference | yes as used; derivation reviewed once, no defect | | Ramanujan and Weil bounds | Pascadi, GAFA, Lemmas 3.2-3.3, confirmed verbatim at the publisher page on a later pass | yes; composite moduli, nonprimitive numerators and n=0 admitted | The Mobius Bombieri-Vinogradov statement in [shifted-prime-decomposition.md (4)](shifted-prime-decomposition.md) had no published citation. A scout searched five channels on 2026-09-06 in the owning convention of [SEARCH-CONVENTIONS.md §1](SEARCH-CONVENTIONS.md) (Mobius function in arithmetic progressions, Bombieri-Vinogradov) and located no published theorem statement of the Mobius case (Granville- Shao assert it without a locator; Koukoulopoulos GSM 203 proves only the prime case; Iwaniec-Kowalski §17.2 and Opera de Cribro §9 were not reachable). [mobius-bv-derivation.md](mobius-bv-derivation.md) now derives (4), with the maximum over y<=T inside, from Koukoulopoulos Corollary 13.4, Theorems 26.2 and 26.6 and equation (26.3) with Vaughan's identity for mu, at level Q<=T^(1/2)/(log T)^(A+6). The provenance sentence in the owning note points there. The derivation was reviewed in the same wave with no defect: the four Koukoulopoulos statements were confirmed verbatim by a second reader, the identity proved, and the bookkeeping found to over-estimate safely. Effectivity: the Siegel-Walfisz inputs are ineffective, so the whole reduction is ineffective and no onset exists. The audit and the handoff already say so. ## 6. Definition drift Load-bearing objects are consistent across the seven notes: beta_W, the cutoffs U, V, Y, Z, D_0, E_0, the alpha_r/f_r expansion, the sawtooth convention and W_dagger. Symbol reuse that a reader must watch: the fourth variable is v in two notes and t in the handoff; H denotes both the fixed logarithmic exponent and the harmonic band; A and B each carry four meanings; c is a Vaughan coefficient, a margin constant, a completion modulus and a Perron abscissa; R is the residual, a set of odd primes and a pair numerator. None of these produced an error. The overload of H is the one worth renaming, since the chain's precision claim is stated in H. Stale rates: three note bodies stated S = C_2 x + (remainder) + o(x) where the derived rate is O_H(x/log^H x). They were updated on 2026-09-06 to the derived form with a pointer to the rate bookkeeping. ## 7. What no reviewer checked The interior of prime-power-dispersion §§4-5 was re-derived in a later pass of the same wave with no defect (zero-numerator count, prime-power gcd identity with the full factor retained, harmonic averages, the assembly (12)-(13), the region (14), the supremum 5757/6700 and the consumption in residual-coverage §3); the singleton-fiber counting argument of §3 was re-derived and shown to transfer to the corner boxes with the error bookkeeping redone. Still unwritten: the right-orientation rewrite of the prime-power and first-branch moments, which sets the 19/25 product edge through J_R. sparse-dispersion and dispersion-range beyond their hypothesis lists. The excluded-prime Mobius lemma of signed-divisor-grouping §2 was checked once, in the link-one review, not twice. Pascadi Lemmas 3.2-3.3 and Baier-Zhao Lemma 2.2 as published statements (numbering and wording). The older budgets cross-referenced in grouped-divisor-moment §6. polylog-fold-transfer §§1-3 and §5. Anything bearing on whether the OPEN margin holds. A local review of one link does not certify the whole reduction, and the whole reduction being correct does not make the signed margin any closer to proved.