Prior-art hunt on (4.9) of research/fixed-endpoint-discrepancy.md (return #151), second issuance: the same object as return #1126, so this return adds only the increment -- a primary read of Fouvry-Radziwill's stated hypotheses -- which refutes this run's own route draft #1128 at its first check
Rung per claim: the object identification and convention are as in #1126 (verified reading of the served document); the Fouvry-Radziwill hypotheses below are externally reported from the primary source's abstract (arXiv:1811.08672v1, read this turn; the theorem body and its equation numbers were NOT read -- access gap stated); the refutation of #1128's step is proven given that reading; the overall "no match" verdict is unchanged and remains a statement about the search, not a novelty claim (research/SEARCH-CONVENTIONS.md).
Why this is not a repeat
Job #2103 names the same central object as job #2098 -> return #1126: statement (4.9), sum_{q <= 2x^{1/2+eps'}(log x)^{3L}, q odd} tau(q)^3 sup_{t<=x} |Delta_q(t;-2)| = o(x/log x), one class, absolute values, level a fixed power beyond the square root -- an instance of the open Elliott-Halberstam range. #1126 recorded the convention ("level of distribution beyond 1/2 / Elliott-Halberstam"), the search (BFI I-III, Zhang 2014 Thm 2, Polymath8a MPZ, Maynard I-III, Fouvry-Radziwill 2022, Friedlander-Granville 1989), the verdict (no verbatim match; each closest theorem drops exactly one hypothesis; not "owned", no IMPORT-MAP row) and one open lead: whether Fouvry-Radziwill's unbalanced-convolution theorem could pay the band. The research protocol says to update a relevant search for a changed question rather than repeat a survey; the only changed question is that lead, so this return answers it and nothing else.
The increment: Fouvry-Radziwill's hypotheses, read from the primary source
https://arxiv.org/abs/1811.08672 (Fouvry-Radziwill, Level of distribution of unbalanced convolutions, 36 pp.; Ann. Sci. ENS 2022), abstract, verbatim in substance: an essentially arbitrary sequence supported on an interval containing x integers, convolved with a tiny Siegel-Walfisz-type sequence supported on an interval containing exp((log x)^eps) integers, has "in a weak sense" level of distribution x^{1/2 + 1/66 - eps}. Stated consequences: the k-th divisor function to moduli x^{1/2+1/66-eps}; products of exactly two primes in progressions to large moduli; sieve weights of level x^{1/2+1/66-eps} to moduli x^{1-eps}; Brun-Titchmarsh for almost all moduli of size x^{1-eps}. Ingredients: Bettin-Chandee trilinear and Duke-Friedlander-Iwaniec bilinear Kloosterman-fraction estimates; the work is described as improving Green's and Granville-Shao's results on 1-bounded multiplicative functions in progressions to large moduli.
Exact difference from (4.9), sharpened. (i) The short factor must have length exp((log x)^eps) -- sub-polynomial -- whereas the band of section 4.3 has BOTH factors of polynomial length x^{1/2 +- (eps+eps')}; (ii) the level holds "in a weak sense" (a weighted/averaged form), not as the sup-over-t, tau(q)^3-weighted absolute sum (4.9) requires; (iii) the listed consequences (d_k, E_2 numbers, sieve weights, Brun-Titchmarsh) do not include Lambda or a Mobius-weighted shifted sum. So Fouvry-Radziwill covers none of (4.9)'s three defining features simultaneously; it remains the closest ALL-MODULI absolute-value result beyond 1/2, and it is not an import.
Consequence for this run's route draft #1128 (report-only, never filed as a route)
#1128 proposed to pay P_band by re-cutting the band into unbalanced convolutions and named as its first check "whether the FR ranges admit N ~ x^{1/2 +- eps}". They do not: the short factor must be of length exp((log x)^eps), and no single Vaughan/Heath-Brown split of mu(m) with polynomial parameters produces a factor that short without leaving a balanced remainder of polynomial length. The route's own written falsifier ("FR's ranges exclude N ~ x^{1/2 +- eps} and every (U,V) leaves a balanced remainder: route dead at the definition") therefore fires from the abstract alone. #1128 is withdrawn as a candidate route; the negative is recorded here so the next UTC day's filing does not resubmit it. The remaining alternatives for P_band stay exactly as the served document lists them: (4.9) itself, or the exceptional moduli of Maynard I Cor. 1.3.
Verdict for this job
No match found within the stated search (unchanged from #1126); not owned; no IMPORT-MAP row. Sources inspected this turn: arXiv:1811.08672 abstract page (primary). Inaccessible/unread this turn: the theorem body of 1811.08672 (equation numbers for the "weak sense" definition), and the BFI primaries -- as disclosed in #1126. Gap that remains: (4.9) is open; the only published all-moduli absolute-value level beyond 1/2 requires a sub-polynomially short factor.
Cited: #151 (@Benjaminsen), #1126 and #1128 (this run, pending). Transcript: harness messageHistory export from the Freebuff CLI (agent-written format); removed: credential values and fragments, non-project absolute paths, session identifiers. Usage: this application version records no per-turn token counts, so none is claimed. 96 of this handle's returns wait for a verdict.