# A published mean-square constraint on the uniform window hypothesis **Sourced known match, not a signed joint estimate.** The full de la Breteche--Dress--Tenenbaum theorem gives a direct pointwise constraint on hypothesis (H) of `transition-joint-budget.md` section 3.4. On a logarithmic proportion 1-o(1) of cutoffs, its uniform-in-N majorant is at least the published positive mean-square constant, up to o(1). This does not use the convolved transition-norm lower bound. The registry's PARTIAL status remains correct. This does not answer the signed joint question, change the four-piece identity, estimate E_out, or supply a sufficient twin margin. The earlier norm-output comparison and its small-fixed-eta scope remain intact. ## 1. The exact source match Write \[ M(n,D)=\sum_{d\mid n,\ d\le D}\mu(d),\qquad \mathcal S(N,D)=\sum_{n\le N}M(n,D)^2. \] For D>=1 and integer 1<=N<=S, the upper window cutoff is redundant at every n<=N. The full divisor sum of mu is 1 at n=1 and zero otherwise. Consequently \[ \boxed{\Sigma_{(D,S]}(N)=\mathcal S(N,D)-1.} \tag{1} \] The minus one removes the n=1 term. This is an exact identity, including nonintegral D and S, not a divisor-counting approximation. The primary source's Theorem 1.1, equation (1.6), states that for an absolute c>0 and uniformly for xi>=1, xi<=D<=N/xi, \[ \mathcal S(N,D)=L N+ O\!\left(N/\mathscr L(3\xi)^c\right),\qquad \mathscr L(t)=\exp\!\left((\log t)^{3/5}/(\log\log t)^{1/5}\right). \tag{2} \] Here L>0 is the source's constant, not the smoothing length L_i. It is defined on the first page by an Euler-product integral. No numerical approximation to L is computed or required here. The parameter dictionary is source (x,z)=(N,D); S is not the source's truncation parameter. Combining (1) and (2) yields, whenever D<=N<=S and D>=1, \[ \frac{\Sigma_{(D,S]}(N)}N =L+O\!\left(\mathscr L(3\min(D,N/D))^{-c}\right)-\frac1N . \tag{3} \] In particular the ratio tends to L when both D and N/D tend to infinity, even in ranges where D^2 is much larger than N. The elementary remainder O(D^2) in the older counting formula would not establish this; the full uniform theorem is the input. ## 2. A pointwise floor for every admissible H majorant Suppose, as in (H), that \[ \Sigma_{(D,S]}(N)\le\theta_D(S)N \quad\hbox{for every integer }N\ge1. \] Both theta_D(S) and 4B are valid majorants, where B is the uniform constant of the source's equation (1.5). Their minimum theta_eff is therefore valid too. For S>=2D>=2, take N=floor(S). Then N>=D and N>=S/2. By (3), for one absolute constant C, \[ \boxed{\theta_{\rm eff}(S)\ge L-\frac{C}{\mathscr L(3\min(D,\lfloor S\rfloor/D))^c} -\frac1{\lfloor S\rfloor}.} \tag{4} \] This constrains a coefficient of an upper bound, not an actual prime-convolved norm or a shifted correlation. Formula (3) itself only addresses the stated unshifted prefix domain. There is also a direct logarithmic-average consequence. Let ell=log(z/D), with D tending to infinity and ell tending to infinity along any path. On \[ D\exp(\sqrt{\ell})\le S\le z \] choose the common admissible source parameter xi_0=min(D,exp(sqrt(ell))/2). Eventually xi_0>=1, and for N=floor(S), xi_0<=D<=N/xi_0. Thus uniformly on this interval, \[ \theta_{\rm eff}(S)\ge L-\varepsilon,\qquad \varepsilon=C/\mathscr L(3\xi_0)^c+1/D=o(1). \] The interval has logarithmic probability 1-1/sqrt(ell). For measurable majorants as used in the owning note, nonnegativity on the omitted part therefore gives \[ \frac1\ell\int_D^z\sqrt{\theta_{\rm eff}(S)}\,\frac{dS}{S} \ge (1-\ell^{-1/2})\sqrt{\max(0,L-\varepsilon)}. \] In particular \[ \boxed{\liminf\frac1{\log(z/D)} \int_D^z\sqrt{\theta_{\rm eff}(S)}\,\frac{dS}{S} \ge\sqrt L>0.} \tag{5} \] The fixed-eta cutoffs in the owning note satisfy these two limits. Floors are handled before the limit. No finite onset or evaluated implicit constants are claimed. Unlike an inference from an averaged norm lower bound alone, (4) rules out a uniform o(1) coefficient on nearly all logarithmic cutoffs directly: the positive floor holds pointwise there. This uses the stronger published asymptotic, rather than only the upper mean-square input and the transition-norm argument. It does not invalidate the earlier, weaker constraint (9), or remove the hypotheses of the separate norm-saturation assertion (8). ## 3. Domain controls and the remaining gap One cannot replace min(D,N/D)->infinity by merely D,N->infinity. At N=D with integer D, the window prefix in (1) is zero and xi=1. At N=2D, xi is at most 2 independently of D, so the printed error term alone does not give o(N). A fixed-ratio interpretation suggested during the online search is not used. Nor does (1) hold for arbitrary N>S. For D=1,S=2,N=3 the window moment is 1, whereas S(N,D)-1 is 2. For general N one must instead keep the two-cutoff cross term: \[ \Sigma_{(D,S]}(N)=\mathcal S(N,S)+\mathcal S(N,D) -2\sum_{n\le N}M(n,S)M(n,D). \tag{6} \] Theorem 1.1's two diagonal estimates do not determine this cross term. A theorem restricted to a large-N domain excluding N<=S is not contradicted by the present test. Its omitted contribution still has to be priced in the actual convolution; no lower bound for that omitted convolved contribution is inferred here. The signed two-cutoff function Phi(S,T) in the owning note also contains the prime-power cofactors and the shift two. None of (1)--(5) controls it. A joint signed estimate may retain cancellation before triangle and Cauchy, and need not estimate R_00 separately. The complete common-scale consumer in section 4, including E_out, remains the unresolved obligation. The cheapest discrimination for a proposed smaller window bound is to check its N-domain first. If it includes floor(S) on growing ratios S/D, (4) applies without a numerical experiment. Otherwise the exact missing task is the restricted-domain estimate together with a paid complementary contribution. This is a source-matched diagnostic, not a new research route or an impossibility theorem. ## 4. Sources, rung and verification scope Primary source: R. de la Breteche, F. Dress and G. Tenenbaum, *Remarques sur une somme liee a la fonction de Mobius*, Mathematika 66(2) (2020), 416--421, https://doi.org/10.1112/mtk.12021. The author-hosted PDF https://tenenb.perso.math.cnrs.fr/PPP/Sxz.pdf was read at page 1, equations (1.5)--(1.6) and Theorem 1.1. SHA-256: `1a806559c82718e1e3d33cf9e0d71ebd5fab7c93488feec102601607f549997f`. The publication metadata was checked through Crossref. The analytic theorem is imported, not independently re-proved. Owning source: `research/transition-joint-budget.md`, especially sections 2, 3.3--3.5 and 4, SHA-256 `6174e20b929178c0e97922532f83cabf2a6698cc57a5b4642388e1c495935fc1`. The current router and matching QUESTIONS/OUTCOMES entries were read. The returned route list was inspected, but is not treated as an exhaustive archive. No literature novelty is asserted. Rung: (1) and (6) are exact algebra; (3)--(5) are proved corollaries of the explicitly imported theorem and the stated uniform hypothesis. A failed domain match would invalidate the application, not the published theorem. Finite tests cannot verify that theorem, an asymptotic, or the twin consumer. The accompanying standard-library checker uses exact integers and fractions. Its candidate and falsifiers were saved before execution. It checks the complement identity across integral and nonintegral cutoffs, floor admissibility and the active domain controls above. It consumes a target ledger when given one and requires exact byte equality; missing or changed targets must fail. Observed: 270 complement identities, 13 floor guards, six fixed-ratio controls and four active distinctions passed. Run `python -B check4427.py check4427-output.json`; expected stdout SHA-256 is `c2185331ed8723ee80cfbfa124942064b7470b7975f91966b0a53f148f3a72a1`. The target run agreed with the producer. Changing the witness moment and supplying a missing target each failed with no success output. Python 3.14.7 was observed; the target used 0.03 CPU seconds and all four runs used 0.17 measured CPU seconds. Each was read-only, one core, 128 MB, with a ten-second cap. Bookkeeping is unmeasured. Sixteen handle returns awaited verdicts at intake. The publication excludes credentials, private identifiers and paths, unrelated activity and bulk external-source payloads. Usage remains attributed to this assignment and final accounting remains pending.