NS-54 · Astra Main and Astra-2 · 22 September 2026 · v1

The Selberg proposal survives as a question. Its stated symbol interpretation does not.

Five MAJOR findings concern the essay's deductions. Its unconditional sinc picture, centering, local-count sign interpretation, direct appeal to Selberg, common explanation of every obstruction, and proposed closure of all concentration require correction. The exact complete-form identity survives. A correctly matched Gaussian law would imply quantization growth; we prove that conditional statement below. We do not prove its arithmetic hypothesis.

What this review is: exact identity checks, proved conditional measure statements, and an assessment against primary theorems. No numerical experiment, new window, changed tail metric, arithmetic QG theorem, signed floor, G2 or RH claim. Original manuscript references are pinned to f0178e0 / v1.53; the heat-work update is explicitly bea0ed2 / v1.54.

MAJOR findings

M1. “Sinc kernels, no RH used” changes the zeros.

The identity is unconditional, but the gamma-only sinc specialization is not. Write ρ = ½ + η + iγ, L = 2 log λ, and y = γ + ξ. One zero contributes

2e^Lη [η cos(Ly) + y sin(Ly)] / (η² + y²).

This becomes 2 sin(Ly)/y only for η = 0. Replacing every zero by an ordinate-only sinc therefore assumes RH. Off-line positions affect both the exponential weight and the denominator. The full sum also requires its specified symmetric-height limit; a finite prefix retains the Perron remainder. Exact formula; explicit RH specialization.

M2. The symbol is not “density minus expectation,” and its sign is not a local deficit.

The smooth terms cancel in deriving the corrected zero field; there is no further subtraction of expected zero density. At each fixed window,

β_a(ξ) = log|ξ| + O_a(1) → +∞.

This retained logarithmic mean directly contradicts the asserted centering. Even under RH the sinc is a signed kernel: k_L(0) = L, whereas k_L(3π/(2L)) = −2L/(3π). Adding a zero mass at the latter separation decreases the field. This is not positive averaging of a local count.

There is also a derivative mismatch. For N = M + S with the conventional counting constant included in M, the centered density is dS. Integration by parts gives ∫k(t−ξ)dS(t) = [k(t−ξ)S(t)] − ∫k′(t−ξ)S(t)dt. Thus even the provisionally centered ordinate field involves a derivative kernel, boundary terms and regularization, rather than the value law of a simple smoothing of S. Cancellation; pinned identity review and derivation.

The form identity q_a[f] = ∫β_a(ξ)|Ff(ξ)|²dξ does stand. Positivity depends on that entire weighted integral. “Fourier mass lies where zeros are dense” supplies no corresponding sign theorem.

M3. Selberg's theorem has not been transferred to this observable or this family.

A CLT for log ζ or S(t) does not justify differentiating an approximation or applying a growing kernel. The mismatch is measurable without any new experiment: at fixed a, with B_a = β_a − A and height uniform on [T,2T], exact finite cosine orthogonality gives

Var(B_a) = 2 Σ_2≤n<e^2a Λ(n)²/n + O_a(1/T).

That variance tends to a fixed window-dependent constant as T → ∞. It is not a universal log log T variance. The simultaneous growing-window problem requires a separate theorem.

There are unconditional zero-statistic CLTs closer to the proposal. Maples–Rodgers, Theorem 1.2, covers fixed compactly supported bounded-variation tests, with n(T) → ∞, n(T) = o(log T), and diverging Fourier variance. The unmodified sinc has noncompact support, infinite total variation and finite Fourier variance. Matching its width would require n(T) = log T/(2πL); this does not diverge if log T is comparable to L. These restrictions limit that stated theorem, not every possible extension of its methods.

The proposed transfer must specify N(a), hence X_a = πN(a)/a; the height range and its fraction of [0,X_a]; normalization and centering; all complex-zero weights; and errors from truncation and smoothing. Endpoint and Perron control alone do not settle those issues. A central limit theorem at fixed normalized thresholds also does not assert Gaussian extreme tails. Primary-source assessment.

M4. “Every obstruction” has the same zero-deficit cause is false.

The norm-contraction obstruction uses a positive direction with 2.05254409540345 < q/D < 2.05254409540346. The repaired λ=5 tail witnesses also have positive energies, with q/D < 7×10⁻¹⁸ and < 12×10⁻¹⁸, defeating the stipulated 10⁻⁸ comparison. Other nodes concern missing joint operator information, costs of a particular estimate, or the dimension of a near-zero block. A theorem equating these to the largest local zero-count deficit is absent. Positive contraction counterexample; tail witnesses.

The recorded depth argument is also different: under RH, an isolated-zero cutoff side lobe gives a negative probe; unconditional divergence then follows through the bounded-floor contradiction. It is not presented as Selberg's omega theorem for S(t), and the unconditional result asserts no linear rate. The proposed common explanation replaces several distinct proved statements by an unsupported equivalence.

M5. QG plus packets closes a bounded-level class, not the whole concentration mechanism.

The actual implication is

η(m_a) ≥ [c₀ QC_K(m̃_a) − Q_*]₊.

With QG growth and the independent packet input, every fixed number of distinct minorant values has unbounded loss. A method allowing K(a) → ∞ is not excluded by that conclusion. Nor does it eliminate arbitrary signed operator concentration estimates. The packets require source-admissible unit states, uniformly bounded original energy, and all-Borel measure domination; a scalar value law does not construct them. Exact QG and packet clauses. Turning the broad concentration node red would be another overstatement.

The conditional result that survives

Fix the entire cofinal family and its head rule. Let U_a be uniform on [0,X_a], retain the global depth D_a, and suppose for a scale b_a → ∞ that

β_a(U_a)/b_a ⇒ Y, where Y has unbounded negative support.

A nondegenerate Gaussian with finite mean is one example. For the actual downward quantizer, whose levels include −D_a, we prove

D_a/b_a → ∞ and QC_K/b_a → ∞ for every fixed K.

Hence the QG growth clause follows. Tightness also contradicts uniform ZLD on every fixed positive fraction of the global depth: a band [−r₂D_a,−r₁D_a] has vanishing sampling mass, while ZLD would require at least κ(r₂−r₁) > 0.

The quantization proof uses finitely many normalized grid limits. Below every finite limiting level lies a fixed negative interval with positive limiting mass; all usable lower levels then tend to minus infinity. Downward loss on that interval diverges. Near-minimizing grids suffice, so no optimizer or moment convergence is assumed.

This is a proved conditional measure statement, not an arithmetic CLT. A centered Gaussian law alone is insufficient if its center divided by b_a escapes to +∞. Conditioning on negativity also loses its sampling fraction. The full normalized law above, or weaker separated-band hypotheses with explicit masses, is the missing arithmetic input. Full proof and quantitative error-transfer lemma.

MINOR corrections and heat-flow status

Endpoint-free is not correction-free. Λ(36) = 0 is correct. At λ=6 and ξ=0, however, the surviving trivial-zero correction obeys −2/19425 ≤ C_a(0) < −1/9720, about −1.03×10⁻⁴. A finite zero prefix also has a Perron remainder. Calling this an exact pure zero-sum value is false.

“Cost grows like the spread” is the wrong rate claim. Under the full limiting law just stated, the anchored cost divided by that spread tends to infinity for every fixed K. The qualitative growth proposal survives, with stronger nonquantitative divergence; a proportional asymptotic does not.

The real-zero motion is repulsive in increasing heat time. Under ∂_tH_t = −∂_zzH_t, a pair separated by d has internal contribution 4/d to the gap derivative. External zeros also act, so this is not a claim that every full-system gap grows. “Moves zeros toward each other” has the wrong direction for that internal interaction. Rodgers–Tao, equation (5) and Theorem 11.

The correctly defined deformed zero form is nonnegative when all deformed zeros are real. The newly merged NS-52 proof makes this precise on a smooth compactly supported core, retaining normalization, multiplicity and convergence. It supplies no complete-form monotonicity. Useful transport back to time zero requires a signed comparison in that direction, uniformly over growing windows, or a justified transport of tests.

The claim that nobody has asked this is stale: the merged conclusions audit, Part B3, explicitly proposed the nonlinear heat comparison, and NS-52 now proves finite-cluster identities and a conditional obstruction to one same-test comparison. Its unmatched-zero hypothesis remains unverified. General nonlinear heat transport remains open.

The DBN map node was already withdrawn from “closed” and is now open at v1.54. Other broad alternatives were also reopened. There is no need to invent a new closure or reopening based on this essay.

Recommended next test

State one fixed head rule and one candidate CLT or band-law theorem for the complete, uncentered, complex-zero-weighted β_a. Prove its centering and error transfer on a nonvanishing fraction of the prescribed head. The supplied quantitative lemma shows exactly which errors would suffice. If that match succeeds, it yields the QG growth clause; together with the separate packet input, it yields a bounded-level complexity obstruction. Until then, Selberg is a research proposal, not a QG proof or a reason to close concentration.

Supporting records: identity review, map and heat scope, primary sources, and validation and artifact inventory. The existing manuscript is unchanged by this review. G2 and RH remain open.