# The corner S_0: a weighted two-point Möbius correlation at the fixed shift 2 **Twin-prime infinitude remains OPEN. The sufficient margin C_2 x + E_dagger(x) >= c_0 x/(log x)^K remains OPEN. Nothing below changes either, and nothing below is an estimate.** This note is a classification and a pricing of one region. It supplies no new bound, and its central negative — that the (r,r') average is not a shift average — closes a hoped-for route rather than opening one. Baseline: [RESEARCH-HANDOFF.md](RESEARCH-HANDOFF.md) §3, commit `4edb6ba`. The full-coefficient follow-up is [corner-coefficient-energy.md](corner-coefficient-energy.md): an exact sharp energy formula, a Graham-based norm for the full smoothed corner, and a small-prime cutoff difference. Those auxiliary results leave the sharp transition, signed saving and global complement open; the exact corner here is unchanged. The corner S_0 is the object of the wave-1 reachability analysis, whose Q1 derives that **the existing zero-frequency budget fails the prescribed fixed margin on S_0 in either orientation; only the exact a=b=1 endpoint fails every positive margin regardless of the cross-kernel saving**. That analysis is the reason this note exists; its budget table is used here as given, not re-derived. --- ## 0. Coordinates, and one correction to the wave-1 sizing Throughout x=2^j, J_x=(x/2,x], and, with w=6/25, v=1/20, \[ U=V=\lfloor x^{w}\rfloor,\quad Y=Z=\lfloor x^{v}\rfloor,\quad D_0=\Big\lfloor\frac{x}{V+1}\Big\rfloor,\quad E_0=\Big\lfloor\frac{x-2}{Z+1}\Big\rfloor, \] \[ R(x)=\sum_{\substack{d>U,\ k>V,\ e>Y,\ t>Z\\ dk-et=2,\ dk\in J_x}} \mu(d)\mu(e)\beta_V(k)\beta_Z(t),\qquad \beta_W(k)=\sum_{r\mid k,\ r>W}\Lambda(r). \] Fix eta_0 with 03/4, 57/10-14eta_0>49/20, and 19/25-2eta_0>151/200 exactly when eta_0<1/400), so S_0 ⊂ W_dagger and its mass belongs to E_dagger. The third condition is a disjunction; its second disjunct de^3>floor(x^(321/200)) holds on S_0 with exponent 361/100-8eta_0, so the inclusion itself holds for every eta_0<13/56. The bound eta_0<1/400 is forced elsewhere, by the single-prime-power cofactor statement in section 1.1 and the norm lower-bound range of [sharp-corner-transition.md](sharp-corner-transition.md), not by this inclusion ([transition-joint-budget.md](transition-joint-budget.md) §1.4). **Correction to the wave-1 sizing (derived).** That report records the e-edge proper-prime-power class, of size O(x^(39/40) log^C x), as "not negligible against x/log^H x on its own". It is negligible: for every fixed H, x^(39/40)log^C x · log^H x / x = log^(C+H)x / x^(1/40) -> 0. Both prime-power classes below are therefore absorbed into the arbitrary fixed logarithmic error, and neither has to be carried. One conclusion in that report does depend on the point: its J4 triage against the e-edge cites this class as a cost that must be carried. That reason falls away. J4's other two reasons stand — the split (19/20, 1/20) is worse than (19/25, 6/25) on every axis, and its clearing price is the larger — so the triage survives with one fewer argument. --- ## 1. The corner sum, written exactly ### 1.1 One prime base on each side, with the s, s' branches kept On S_0, kV and Q^b>V with P!=Q would force k>=P^aQ^b>V^2. It does **not** exclude two powers of one base, P^a and P^(a+1), both dividing k and both above V; those have a>=1 and P^(a+1)|k, hence lie inside the proper-prime-power class priced below. So the correct statement is: **at most one prime base contributes, and up to that class exactly one prime power divides each cofactor, with weight (log r)(log r').** The validator counts the exceptions explicitly — cofactors with two prime powers above the cutoff: 2, 0 and 5 in the three cases — and sums over all of them, so the identity checks of §6 are unaffected by this correction. Write \[ k=rs,\qquad r=P^a>V,\qquad s=k/rZ,\qquad s'=t/r'=2, sum_(p^j>W,j>=2) p^(-j) << W^(-1/2): split p at sqrt(W). For each smaller p its geometric tail is at most 2/W; for larger p sum_(j>=2)p^(-j)<=2/p^2, and summing over integers bounds both ranges. Counting multiples costs at most x times this tail. Divisor multiplicity and logarithmic weights are absorbed in x^epsilon. Thus the V and Z classes cost O(x^(22/25+epsilon)) and O(x^(39/40+epsilon)), respectively, including repeated powers of the same base. For fixed sufficiently small epsilon these are O_H(x/log^H x) for every fixed H. This does not remove s,s'>1 or all non-squarefree n. See independent review F5. *Coprimality of the two band primes.* For eta_0<19/200 the two bands (V, V x^(2eta_0)] and (Z, Z x^(2eta_0)] are disjoint, so r≠r' always and (r,r')=1 is automatic. It is **not** automatic for the cofactors: with s,s'>1 the gcd g=(k,t) may be 2, and g must divide 2 for a solution to exist. Both branches are kept below. ### 1.2 The exact corner sum For K=rs, T=r's' with g=(K,T)|2, put K/g · d_0 ≡ 2/g (mod T/g), 0<=d_00. \tag{1} \] I(K,T) is the integer window cut out by max(U,D_1)> gx/(2KT) holds where the two corner cuts d>D_1, e>E_1 are already implied by x/2=d>x^(19/25-2eta_0) and b>=e>x^(19/20-2eta_0), so \[ q=\frac{ab}{g}\ \ge\ \tfrac12\,x^{171/100-4\eta_0}\ \gg\ x . \tag{2} \] Each compatible congruence class therefore meets J_x in **at most one** integer, and Delta_{a,b} is a difference of two sawtooths at essentially arbitrary arguments with no complete period available. This is the exact local form of the wave-1 observation that at the corner "there is no complete period to remove"; Fourier completion buys nothing here. The useful consequence is in the other direction. By the exact discrete partial summation of signed-divisor-grouping (17)–(18), R|_{S_0}=M|_{S_0}+E_dagger|_{S_0} with M the discrete density term (11). At fixed (n,e,r,s) the S_0 condition leaves d in a single interval inside [U/2,x], and the excluded-prime Möbius mean [signed-divisor-grouping (7)–(8)](signed-divisor-grouping.md), which is uniform in the excluded-prime parameter m<=2es<=x^2, applies verbatim. Hence M|_{S_0}=O_H(x/log^H x) and \[ E_\dagger\big|_{S_0}(x)=R\big|_{S_0}(x)+O_H(x/\log^H x). \tag{3} \] So the sawtooth weights do **not** simplify into anything summable on S_0; what simplifies is that one may discard the sawtooth formalism there and work with the raw signed residual (1). That is a removal of a false hope, not a saving. ### 1.4 The two-point form Let C_{U,V}^{S_0}(n)=sum over dk=n with k>V, max(U,D_1)V\\ D_1Z\\ E_1V,rs<=x^(w+2eta_0)) (log r)/(rs) of order eta_0^2*log^2 x. For a lower bound restrict both prime log-widths and small-cofactor log-widths to eta_0, with odd coprime s,s', then count squarefree d,e as in the review. No leading Euler-product constant is asserted, and these fixed-eta statements cannot be substituted at eta(x). --- ## 2. What the sufficient consumer actually needs on S_0 ### 2.1 The conditional requirement Suppose the complement W_dagger \ S_0 were controlled to O_H(x/log^H x) with H>K — which is **not** proved. A formal exponent calculation in reachability Q3.5 is not a necessity theorem for all ways of estimating that complement. Under this explicit complement hypothesis, handoff §3's consumer becomes exactly \[ E_\dagger\big|_{S_0}(x)\ \ge\ -C_2x+c_0\,\frac{x}{\log^Kx} \quad\text{on an unbounded set of dyadic }x. \tag{6} \] ### 2.2 (6) is not a lemma; it is the theorem Unconditionally S(x)=sum_{n in J_x}Lambda(n)Lambda(n-2)>=0, and S=C_2x+E_dagger+O_H(x/log^H x). So \[ E_\dagger(x)\ \ge\ -C_2x-O_H(x/\log^Hx) \tag{7} \] holds with **no further arithmetic input** beyond the reduction itself. The entire content of the sufficient consumer is the margin c_0 x/log^K x. Consequently, given the complement hypothesis, (6) implies and is implied by S(x) >= c_0 x/log^K x + O_H(x/log^H x). That margin sits between two quantitative twin lower bounds differing by log^2 x: it yields N_2(x) >= (c_0/2)x/log^(K+2)x (handoff §3), and conversely N_2(x) >= c x/log^K x yields a margin of that shape, since each genuine pair contributes ≍ log^2 x to S. So a one-sided attack on S_0 is not a step towards the theorem; conditionally it is the theorem up to that log^2 x. This is stated flatly because it determines what is worth attempting there. The absolute route is the one that behaves like a lemma. If |E_dagger|_{S_0}|=o(x) then, given the complement, C_2x+E_dagger>=(C_2-o(1))x, a fixed-fraction margin, far more than (6) asks. By §1.5 the triangle inequality gives ≍ eta_0^4 x log^4 x, so: A saving of log^(4+eps) x over the full unsigned envelope is sufficient for o(x); for the prime-cofactor subfamily log^(2+eps) x is sufficient. These are sufficient absolute estimates, not necessary rates for every one-sided argument. ### 2.3 Shrinking eta requires new uniform estimates The mass estimates above hold for fixed eta_0. The moment bounds retain x^epsilon losses and constants depending on fixed margins. Neither is uniform at eta(x) without additional work. In particular the former comparison of eta^4*log^4 x against x^(-c eta+epsilon) does not prove that no eta(x) works. It only diagnoses a tension in an optimistic substitution that has not been justified. The global shrinking-window strategy remains OPEN; the existing bounds do not establish it. ### 2.4 Is there structural positivity or a sign bias in (1)? Assessed candidate by candidate. **No bias was found; the sum is genuinely signed.** | candidate source of a sign | status | |---|---| | the u_1=u_2, h_1=h_2 diagonal | not present. That diagonal is nonnegative inside the *second moment* after the first Cauchy; (1) is the raw signed sum, before any moment. Reachability Q1.6 item 3 is about the moment, not about R|_{S_0}. | | coprimality forced by dr-er'=2 | derived: (d,e)|2, (d,r')=1, (e,r)=1, (n,n-2)|2, and by §1.4 n odd on the squarefree class. Each is a *support* restriction; none biases mu(d)mu(e). singleton-fiber-audit §3 derives that both signs occupy a positive proportion of odd squarefree coprime divisor pairs on a box, and that counting argument transfers verbatim to the corner's boxes. | | the density subtraction already performed | removes nothing exploitable: by (3), M|_{S_0}=O_H(x/log^H x), so no positive main term survives on S_0 to be leaned on. | | nonnegativity of L, L' in (5) | the weights are nonnegative but the sign is carried by mu(n)mu(n-2), which is exactly the unknown. | | S(x)>=0 | gives (7) only, i.e. the trivial direction; see §2.2. | | longest-fiber structure | on S_0, kt can range up to x^(29/100+4eta_0), still below x; away from clipping, fibers have many points, while upper-edge clipping can shorten them: S_0 is the *opposite* extreme to the singleton family of singleton-fiber-audit, whose lower bounds are for kv>x. No sign information is transferred either way. | Heuristically, under a Chowla-type independence model the inner sums in (1) are square-root sized and the whole of (1) is O(x^(0.65+eps)), far inside (6). That is a heuristic and carries no weight here beyond saying that the obstruction is not expected to be a real main term. --- ## 3. Classification against the literature, in the owning conventions Searched per [SEARCH-CONVENTIONS.md](SEARCH-CONVENTIONS.md) row "the residual with two signed factors and fixed product difference" (owning words: two-point Möbius correlations, averaged Chowla, logarithmically averaged Elliott, determinant equation). The mismatches already recorded as [F-0905-11](OUTCOMES.md) are not repeated as new findings; what is new here is §3.1, which is structural rather than a parameter mismatch. ### 3.1 What averaging our object actually offers — derived negative The corner offers x^(29/100+4eta_0+o(1)) pairs of prime dilations (r,r') with r ≍ x^(6/25) and r' ≍ x^(1/20), each with an inner sum of length ≍ x/(rr') (so x^(71/100) at the foot of the bands), logarithmic weights (log r)(log r'), a fixed interval J_x, and the additional (s,s') divisor sums. **The map (r,s,r',s',m) ↦ (n, k, t, r, r') with n = d·rs, k = rs, t = r's' is a bijection onto {(n,k,t,r,r'): k|n, t|(n-2), r in supp beta_V(k), r' in supp beta_Z(t), corner cuts}.** The last two coordinates record which prime power supplies the weight, so this is a bijection with multiplicity rather than onto the triples alone; the s=s'=1 sub-family and the exceptional multi-power cofactors of §1.1 are both covered. Summing the weights over the last two coordinates gives (4). Either form supports the conclusion equally: - the (r,r') average **is** the average over the factorisations of n and n-2 at the single fixed shift 2. It is not an independent resource sitting outside the correlation; it is the correlation, re-indexed; - the change of variables the assignment asks about does exist — fixing r and varying r' moves d_0(r,r') = 2·r-bar mod r' over residues — but it moves the base point of a progression, **not** the shift of a correlation. The shift is invariant at 2. The available additive-shift range is zero, at zero cost, because there is nothing to convert; - consequently [Matomäki–Radziwiłł–Tao's averaged Chowla](https://arxiv.org/abs/1503.05121), whose whole content is an average over additive shifts h_1..h_k<=H with H->infinity, has no purchase here. This is stronger than the parameter mismatch recorded in F-0905-11: the mismatch is structural. The one genuinely relevant reading of the (r,r') average is the opposite one: the weight "n has a prime factor in a band" is exactly the device Tao's entropy decrement and the Helfgott–Radziwiłł / Pilatte expansion arguments use *inside* their proofs. Our object arrives with that device already applied, at fixed prime scales x^(6/25) and x^(1/20) rather than at the ranges those arguments require. ### 3.2 Sources, statements and hypotheses Statements below were read at the cited source; "CHECKED" means the statement and the listed hypotheses were read there in this session, "ASSUMED" means it is carried from an existing repository record. | source | statement read | average | coefficients / shift | saving | why it does not close S_0 | |---|---|---|---|---|---| | [Tao, *Forum Math. Pi* **4** (2016) e8 = arXiv:1509.05422](https://arxiv.org/abs/1509.05422), Thm 1.2, Thm 1.3 | logarithmically averaged Chowla for **two linear forms** a_1n+b_1, a_2n+b_2 with a_1b_2-a_2b_1≠0; Thm 1.3 the Elliott/nonasymptotic form with a non-pretentiousness hypothesis on g_1 | logarithmic, over (x/omega,x] | a_i, b_i **fixed**; Thm 1.3 has a threshold A=A(eps,a_1,a_2,b_1,b_2) | qualitative o(log omega) | our a_i are r' ≍ x^(1/20) and r ≍ x^(6/25), growing; the threshold is not stated uniformly in them, and a logarithmic-average o(1) is not a natural-average saving. CHECKED (statement and the coefficient-dependent threshold). Already recorded as F-0905-11. | | [Matomäki–Radziwiłł–Tao, arXiv:1503.05121](https://arxiv.org/abs/1503.05121) | sum over h_1..h_k<=H of the absolute correlation is o(H^k X), any H=H(X)->infinity | additive-shift average | forms fixed, shifts averaged | ≍ log log H/log H | no shift average exists here, by §3.1. CHECKED (abstract and main statement). | | [Helfgott–Radziwiłł, arXiv:2103.06853](https://arxiv.org/abs/2103.06853), Main Theorem | eigenvalues of the centered divisibility operator are O(sqrt(L)) off a density-1 set; consequence (1/log x)sum lambda(n)lambda(n+1)/n = O((log log x)^(-1/2)) | logarithmic | fixed shift 1, lambda not mu, no weights | (log log x)^(-1/2) | wrong average, and the saving is (log log)^(-1/2) against a requirement of log^(4+eps). CHECKED (abstract, main theorem, prime ranges log H_0 >= (log H)^(2/3+eps), log H <= (log N)^(1/2-eps)). Also recorded in structural-literature-audit §3E. | | [Pilatte, arXiv:2310.19357](https://arxiv.org/abs/2310.19357), Thm 1.1 | sum_{n<=x} lambda(n)lambda(n+1)/n << (log x)^(1-c), c>0 absolute | logarithmic | fixed shift 1 | (log x)^(-c), c unspecified, described as best possible with current methods | wrong average; c is not claimed to exceed 4, or 2 for the sub-family; no weights, lambda not mu. Abstract CHECKED; the value of c and Theorem 1.1's exact form NOT checked at source. | | [Tao–Teräväinen, arXiv:1809.02518](https://arxiv.org/abs/1809.02518), Cor. 1.13 and the Chowla corollary | **unweighted** two-point Chowla/Elliott at all scales X outside a set of logarithmic Banach density zero | natural within a scale; exceptional scale set | shifts h_1,h_2 **fixed**, no uniformity stated | qualitative o(X) | closest interface to the handoff's scale-average consumer, and the only one on a natural average. Two blocks: coefficients fixed, and o(X) applied fibrewise gives o(mass)=o(eta_0^4 x log^4 x), not o(x). CHECKED via the ar5iv text; the proof's uniformity NOT checked. | | [arXiv:2608.23500 (2026)](https://arxiv.org/html/2608.23500), Thm 1.1 | (log x)^(1-c) bound on sum_{n<=y} lambda(n)lambda(n+h)/n for every h outside a set E_x with |E_x ∩ [1,H]| <<_A H(log x)^(-A) | logarithmic | fixed forms; h ranges | (log x)^(-c) | wrong average; and h=2 is one shift, which the theorem's exceptional set is not shown to avoid. CHECKED (abstract and theorem statement). | | [Tao–Teräväinen, arXiv:2512.01739](https://arxiv.org/html/2512.01739v2), Thm 3.1, Rem. 3.2 | quantitative correlations of 1-bounded multiplicative functions outside a small set of logarithmic scales; moduli and shifts bounded by a small power of a parameter <= log X | logarithmic scales | polylogarithmic coefficients | quantitative | our coefficients are fixed powers of x. ASSUMED from structural-literature-audit §3E; not re-fetched here. | | Frantzikinakis; [Frantzikinakis–Host](https://arxiv.org/abs/1611.09338) | ergodic reductions: ergodicity/genericity hypotheses on the Liouville system imply Chowla | — | — | conditional | conditional on an unproved ergodic hypothesis; supplies no unconditional bound. CHECKED at abstract level only. | | Klurman, *Compositio* **153** (2017) 1622–1657; [Klurman–Mangerel, *Math. Ann.* **372** (2018) 651–697](https://arxiv.org/abs/1707.07817) | rigidity/converse: if many binary correlations of a 1-bounded multiplicative f match a character's, then f(n)=chi'(n)n^(it) | — | fixed shifts | — | the converse direction. Gives no upper bound for our sum. CHECKED at abstract level only. | | [Friedlander–Iwaniec, asymptotic sieve for primes](https://arxiv.org/pdf/math/9811186), Thm 1 with (R),(B) | distribution plus a signed bilinear hypothesis yields a prime asymptotic | — | — | — | hypothesis (B) is itself a signed bilinear axiom; invoking it renames the problem. ASSUMED from structural-literature-audit §3F. | *Not verified, and not used:* the assignment's remark that Chowla for mu(an+b)mu(cn+d) with fixed a,c is equivalent to the two-point case. Tao's Theorem 1.2 covers two linear forms directly in the logarithmically averaged setting, which is what matters here; a formal equivalence for the unaveraged conjecture was not located and is not relied on anywhere above. ### 3.3 Two determinant-equation interfaces, priced at the corner These two were checked at source in this session because they attach to the determinant equation itself rather than to the correlation. Both fail, with explicit exponents. Neither failure is a refutation of the underlying target. #### (a) Bettin–Chandee, *Adv. Math.* **328** (2018) = arXiv:1502.00769v1, Corollary 1 — fails on smoothness, and on norms, in mutually exclusive orientations Read at source (p. 3, eq. (1.4)). For Delta≠0 and T = sum over m_1n_2-m_2n_1=Delta of f(m_1)g(m_2)alpha_{n_1}beta_{n_2}, with supports [M_i/2,M_i], [N_i/2,N_i], **alpha and beta arbitrary** but f, g smooth to all orders, \[ f^{(j)}\ll\eta^{j}M_1^{-j},\qquad g^{(j)}\ll\eta^{j}M_2^{-j} \quad(\forall j\ge0,\ \text{some }\eta>1), \] the asymptotic holds with main term over (n_1,n_2)|Delta and error \[ O\big((\eta R)^{3/2}\,\|\alpha\|\,\|\beta\|\,(N_1N_2)^{7/20}(N_1+N_2)^{1/4+\epsilon}(M_1M_2)^{\epsilon}\big), \quad R=\frac{M_1N_2}{M_2N_1}+\frac{M_2N_1}{M_1N_2}. \] Matching our corner equation dk-et=2 (so Delta=2) admits three orientations. R ≍ 1 in all three, which is the one hypothesis the corner satisfies comfortably. Exponents of x, with the s=s'=1 scales d ≍ x^(19/25), e ≍ x^(19/20), k=r ≍ x^(6/25), t=r' ≍ x^(1/20): | orientation | (M_1,M_2,N_1,N_2) | smooth weights required on | required eta | error exponent at eta=1 | error exponent at the actual eta | main term | |---|---|---|---|---|---|---| | A: smooth on the cofactors | (x^(6/25), x^(1/20), x^(19/20), x^(19/25)) | beta_V(k), beta_Z(t) | ≍ M_1 = x^(6/25) | 1691/1000 | 2051/1000 | ≍ x | | B: smooth on the divisors | (x^(19/25), x^(19/20), x^(1/20), x^(6/25)) | mu(d), mu(e) | ≍ M_1 = x^(19/25) | 613/2000 | 2893/2000 | ≍ x | | C: primes absorbed, smooth on s, s' | (x^(2eta_0), x^(2eta_0), x, x) | the constant 1 | O(1) | 39/20 | 39/20 | ≍ x^(1+2eta_0) | Readings, all derived from the displayed error term: - **Orientation B is the only one whose norms work, and it fails exactly on smoothness.** At eta=1 the error exponent is 613/2000 against a main term of exponent 1 — a saving of x^(1387/2000). The affordable smoothness budget is eta=0, with the implied constants independent of j, so admitting mu(d) at eta ≍ M_1 presupposes a smooth interpolation of mu on [M_1/2,M_1] whose derivative constants are uniform in j. No such interpolation is exhibited here or known to the author; without one the orientation does not meet the hypothesis at any eta, and 2893/2000 is a lower bound on its cost, not its cost. This is the coordinator's point, confirmed with the exponents: the corner's weights are the least smooth case, and it fails there. - **Orientation A puts the arbitrary coefficients on the long variables**, where ||alpha||||beta||(N_1N_2)^(7/20)(N_1+N_2)^(1/4) alone is x^(1691/1000), already x^(691/1000) above the object before any smoothness cost. The corollary saves N^(1/20) when N ≍ M; our box has N_1N_2 = x^(171/100) against M_1M_2 = x^(29/100), unbalanced in the direction that hurts. - **Orientation C satisfies the smoothness hypothesis** (f=g=1 on a dyadic block) but aggregates the primes into the long variables, giving ||alpha||,||beta|| ≍ x^(1/2) and error exponent 39/20, x^(19/20) above the object. So smoothness and the norm budget are satisfiable at the corner only in mutually exclusive orientations. A separate agent owns Corollary 1 on the bulk; nothing above bears on that. #### (b) Guria, arXiv:2410.10856v2, Theorems 1.2 and 1.3 — the right prime mechanism, no free variable to apply it to Read at source. Theorem 1.2 counts (a,b,p,d) in [-X,X]^4 with ad-pb=r, p prime, one weight alpha(a)=O(a^eps) and b, d free; Theorem 1.3 has two primes p, q with b, d free and gives S_r(X) = 8 K_r (li X)^2 + O(X^(1+3/4+eps) + r^(1/5)X^(1+11/20+eps)), uniformly for 0<|r|<= q. In her cube, b and d have length X against moduli a, p ≍ X, so the dual has length ≍ 1. At the corner the only variables not carrying an oscillating weight are the cofactors: k has length x^(6/25) against a modulus e ≍ x^(19/20), dual length x^(71/100) — the dual sum is longer than the original by x^(47/100), and r' is shorter still. Her prime is also comparable to the other variables; ours are x^(6/25) and x^(1/20), and the r'-average has only x^(1/20+2eta_0+o(1)) terms. 3. **Two Möbius weights, and no main term (structural).** Her theorems allow one arbitrary weight; the other three variables are free, which is why Poisson applies and why the conclusion is an asymptotic with a genuine main term of size ≍ (li X)^2. The corner carries mu(d) *and* mu(e) on the two long variables and prime restrictions on both short ones: no free variable exists. And by (3) the density on S_0 has already been subtracted and is O_H(x/log^H x), so there is no main term to produce — the whole object is the fluctuation. A counting theorem transfers to it only through its error term, and by item 1 that term already exceeds it. What would have to be proved instead: an analogue of Theorem 1.3 for the unbalanced box d ≍ x^(19/25), e ≍ x^(19/20), r ≍ x^(6/25), r' ≍ x^(1/20), with arbitrary bounded coefficients on **both** long variables and error o(x/log^K x). By §2.2 that statement is, given the complement, at least as strong as a quantitative twin lower bound. That is not a reason to dismiss the mechanism, but it is a reason not to expect it from a parameter substitution into a counting theorem. ### 3.4 Updated quantitative comparison The earlier source table describes the particular versions and direct substitutions inspected there. It is not an exhaustive current absence claim. [prime-band-transfer.md](prime-band-transfer.md) now consumes [Tao-Teravainen v2 Theorem 3.1](https://arxiv.org/html/2512.01739v2) through an exact bounded multiplicative Fourier lift. Uniformity in its parameters and the moving-window mesh are priced there. This gives a small continuous scale-average log saving for the absolute prime-cofactor sum normalized by N*log^2 X, at fixed eta. The interval-stability argument in [round-review-0906.md](round-review-0906.md) §2 also gives a dyadic scale-average saving, with exponent c_*/2. Neither estimate gives o(N) or a bound at every dyadic point. The full C-product retains remaining prime-r cofactor and non-squarefree branches; proper prime powers are separately negligible by §1.1. The handoff's one-sided consumer needs a stronger rate and the full remainder's accounting. No full-corner estimate follows from these transfers. Against the fixed-eta unsigned envelopes, a saving of more than four log powers for the full corner, or more than two for the subfamily, is **sufficient** to reach o(x) by absolute estimation. These are not necessary exponents for every one-sided argument, and a suitable little-o improvement at equality would also suffice. Any new theorem comparison must price its actual norm, rate, exceptional scales and coefficient class rather than repeat a blanket absence assertion. --- ### 3.6 Prior art for the shape, found after the classification The object in (5) is not new as a shape. Tao's 2016 notes on the Bombieri asymptotic sieve state that on GEH the twin asymptotic is equivalent to sum_{n<=x} mu(n) 1_R(n) mu(n+2) 1_R(n+2) = o(x/log^2 x) for the x^alpha-rough indicator 1_R, "and similarly with 1_R replaced by other sieves", and give the reason recent Chowla progress does not transfer: the weight destroys multiplicativity of mu at small primes. Maynard's ICM 2022 survey poses the four-variable determinant-2 sum with arbitrary bounded coefficients as Question 17 and remarks that the natural Cauchy-Schwarz argument gains little. The identification in (4)-(5) is therefore an independent rediscovery of a known framing, reached here through the Vaughan-type remainder rather than the asymptotic sieve; see [consumer-comparison.md](consumer-comparison.md) for the sources. Two differences travel with the attribution: the weight L here restricts a large prime factor and leaves small primes unrestricted, so Tao's stated non-transfer reason does not apply verbatim and would have to be re-argued; and the required relative saving here is log^(4+eps) x against his o(1), because the whole prime band is summed with log r weights. The one-sided target's equivalence with the twin margin holds for any subregion whose complement is controlled, so it is a property of the consumer, not of S_0. ## 4. The parity input, named *What this note used.* Exact divisibility and CRT algebra; the classical Möbius mean [Tao, Notes 2, Exercise 66](https://terrytao.wordpress.com/2014/12/09/254a-notes-2-complex-analytic-multiplicative-number-theory/) through signed-divisor-grouping (7)–(8); quantitative PNT [Tao, Notes 2, Corollary 39](https://terrytao.wordpress.com/2014/12/09/254a-notes-2-complex-analytic-multiplicative-number-theory/); Mertens; and counting. Nothing here uses more than residue and density information plus those classical means. In particular §§1–2 use no bilinear estimate, no large sieve and no sieve upper bound beyond counting. *Does closing S_0 require more?* **Assessment, medium-high confidence, not a theorem.** Yes. Identity (5) says the corner's leading family is sum_n mu(n)mu(n-2)L(n)L'(n-2) with L,L'>=0: the corner does not merely resemble the parity difficulty, it **is** the two-point Möbius correlation at shift 2, weighted. Every advance on that object reviewed in §3 uses inputs beyond residue counting and sieve upper bounds — Halász/pretentious distance, Matomäki–Radziwiłł short-interval averages, entropy decrement, expander and non-backtracking estimates. Nothing in the tile framework supplies any of those. *What is NOT claimed.* No obstruction theorem. Per [F-0905-01](OUTCOMES.md) and the parity gate, an obstruction claim must name its method class, the statistics it retains, the errors it tolerates and its quantifiers, and prove a statement at that scope. This note names none and proves none. What is derived instead is narrower and checkable: the structural negative in §3.1 (no shift average exists), the quantitative gap in §2.2 (log^(4+eps) on a natural average), and the incompatibility in §2.3 (eta_0 cannot be shrunk). --- ## 5. Payoff, exactly **What a complete corner estimate would change.** - The reduction S(x)=C_2x+E_dagger(x)+O_H(x/log^H x) is unchanged. A corner estimate changes only the accounting: |E_dagger|_{S_0}|=o(x) would give E_dagger=E_dagger|_{W_dagger\S_0}+o(x), so the sufficient consumer of handoff §3 would reduce to a statement about the complement alone. - It changes **no** budget, region or cut in [grouped-divisor-moment (15), (19), (21)](grouped-divisor-moment.md). The block budgets (1+a)/2, a/2+3b/2, a are untouched; the added region delta<19/25 and delta+3nu<161/100 is untouched; the target (21) and its "more than 3/50" are untouched. - It does **not** raise the uniform product threshold, which stays at every fixed exponent below 19/25. The witness (8/25,11/25) is outside S_0. - It does **not** let the vertical cutoff delta<19/25 in (15) be dropped. S_0 is the *intersection* {delta>19/25-2eta_0} ∩ {nu>19/20-2eta_0}; the d-edge strip S_R and the e-edge strip S_T of reachability Q3.1 both survive. Their clearing prices are not 11/10 and 37/25: those are the costs at the *cheapest single points* of the two edges, (19/25,1/20) and (6/25,19/20). Clearing either edge apart from the corner drives gamma to 2, because along the d-edge the only usable orientation is the left one, whose requirement 2-3p-q tends to 2 as q tends to 0, and symmetrically on the e-edge. Both strips inherit the corner's own price in the limit. **What it would not do, and what nothing else does.** Reachability D1 derives that no nonzero-kernel improvement, of any size, reaches S_0; D2 derives that reducing the leftover to S_0 alone already needs gamma=2 uniformly. So the corner cannot be removed by the moment route and the complement cannot be removed by the corner route: the two halves are disjoint and neither is proved. Even both together would then have to be combined with the still-unproved uniform saving to reach (6). **Triage opinion (mine, flagged).** The absolute target on the s=s'=1 sub-family, (5) with a log^(2+eps) saving, is the smallest well-posed piece of the corner and the only one whose statement is a recognisable open problem rather than an equivalent of the theorem. I would not spend an attempt on the one-sided form (6): §2.2 shows it is the conclusion. Medium confidence; this is triage, not a result. --- ## 6. Validation, controls and falsifiers [corner-correlation-validation.js](corner-correlation-validation.js) checks the algebra of §1 exactly, as integer combinations of log(r)log(r'), at three finite parameter sets. At reachable x the asymptotic windows are degenerate (Z=floor(x^(1/20)) is 2 at x=2^22), so two of the three cases use larger cutoff exponents w, v chosen so that the s>1, s'>1, prime-power and g=2 branches are actually populated; the corner is defined intrinsically as V