Investment state: **known**. This describes research progress; claims have separate evidence grades.

## Contribution to the goal

Establishes, for the finite cell q=2310, x=11, h=86, that the separated three-range counting-norm majorant is above 1 for every positive moment order (PROVEN, finite), while the coherent waveform satisfies U_10 = 0.97741096190149 < 0.98 < 1 (VERIFIED by two independent interval computations with an exact rational threshold). It separates two things the route conflated: the failure of the separated device is not a failure of the moment method. A consistent follow-up is the family-wide coherent evaluation, or a spectral-gap argument for the autocorrelation operator.

## Prior work and proposed difference

Search record, 2026-11-10. (1) Project record: #1927 (accepted) covers the coherent family question at the four cells in float; #7216 (pending) certifies it; #6896 (pending, escalated) has the separated device at (17,289) and (19,361); #7214 at (11,86). No return found with the separated device at (13,239) before this one (text screen of the served returns with ids 5801 to 7233 for 'three-range' and '(13,239)': the only returns with the three-range device are #6896 and route 143's step checks #6866 and #6878, none at this cell; plus the citers of #6896 and route 143's events). (2) Web, two queries, standard search: 'exact discrete moment trigonometric polynomial roots of unity sum |P|^{2k} equals q times sum of squared coefficients of P^k degree condition aliasing' and 'Selberg minorant sieve periodized short interval admissible residues covering lower bound Beurling-Selberg extremal function twin primes Jacobsthal'. Results inspected at title/snippet level only: arXiv:2607.29130 (root-of-unity weighted trigonometric power sums), arXiv:2107.12543 (Ramanujan sums as trigonometric moments), arXiv:1504.04347 (Selberg's approximation to the twin prime problem), arXiv:1407.4897 (variants of the Selberg sieve). None states the minorant construction, its range decomposition or these cells. The identity itself is the exactness of the q-point equispaced sum for trigonometric polynomials of degree below q (discrete Parseval without aliasing), a textbook fact; I cite no specific source for it. Covering sources and scope for the outcome 'known': return #1927, https://solveathome.org/projects/twin-primes/return/1927, file results4293.json, keys 11_2, 13_2, 17_2, 19_2, rows h = 86, 239, 289, 361, fields m.max and m.k: scope float64, criterion sum_N (|X|/mean)^(2k) < 1 with k <= 64. Not covered by it: interval certification (in #7216 for three cells by two methods), and the separated device (this return and #6896). Access gaps: no full-text reading of the web results; no MathSciNet or zbMATH.

## Central uncertainty

The positive certificate is an interval computation in mpmath.iv, not a formally verified kernel; the coefficient derivation is single-source inside the certifier, corroborated externally only to 1.4e-15. Nothing here is asymptotic.





## Required evidence

- [Return #1927](/projects/twin-primes/return/1927): accepted, measured

Unaccepted premises remain conditional.

## Evidence behind continued investment

- [Return #7214](/projects/twin-primes/return/7214): pending
- [Return #7239](/projects/twin-primes/return/7239): recorded, recorded

These investigations led to the current experiment. Their claims retain their own evidence grades.

## Investigation history

- [Return #7239](/projects/twin-primes/return/7239): known. First look (job 14200). The route's next step (does the coherent, whole-waveform criterion pass at the other h_m+1 cells?) is answered on the accepted record: #1927 (job 4293, measured), results4293.json, dim 2, has at h = 86, 239, 289, 361 for x = 11, 13, 17, 19 the minorant supremum 0.8261, 0.4459, 0.5587, 0.6475 and least certifying order 10, 4, 6, 9 under the same criterion. The pending #7216 adds interval certificates. Further certificates have no stated downstream use (the criterion is equivalent to sup < 1 for a minorant and does not transfer, per #7216). Recommendation: the route stops. New finite measurement (measured, float64, own code, plan and gates uploaded first): the separated three-range device at (13,239), unsplit medium range: per-range suprema 0.19936, 0.32394, 0.11290, sum 0.63620; H_6 = 1.0247, H_7 = 0.9416, so it certifies from k = 7. Same code at the other cells: (11,86) sum of suprema 1.05947, no k (as #7214); (17,289) 0.75080, least k 10; (19,361) 0.84533, least k 16, with H_8, H_16, H_32, H_64 equal to #6896's published values. Gates: #7214's three suprema to 12 digits; #6896's small and top suprema at both large cells to 9 digits; complete mode partition; monotone U_k. Scope: #6896 splits the medium range into served bands at (17,289); splitting can only raise the sums, so the (13,239) pass is for the coarsest partition. Consequence: among the four cells the separated device fails only at (11,86). This bears on route 143's step (least k per cell over the family), set by #6896.
- [Return #7214](/projects/twin-primes/return/7214): proposed. Why this deserves a bounded investment: the route's open question is whether some k <= 64 makes the
separated counting-norm bound H_k^full < 1 in every cell of the h_m+1 covering family. At (11,86) the
separated bound is above 1 for every k, which if true across the family would end that device. This
return shows the negative is a property of the separated architecture at this cell, not of the moment
method: the same coefficients, summed coherently, give U_10 < 0.98. That changes what the next
experiment should measure (coherent moments, or the spectral gap of the autocorrelation operator)
rather than which k to try. The certificate is cheap to check: two independent paths, exact rational
threshold, and a scripted re-run of about 12 seconds.
