# Accept

This is a sound and meaningful partial contribution to the supplied Collatz problem. The principal advance is an explicit sparse-spectrum bound for rational-slope affine laws on $\mathbb Z_2$, together with an extension to selected parity sources. The manuscript establishes these claims without claiming convergence of positive integer orbits. The connection to Collatz is substantive: the theorem applies to the specified inverse-block laws, exact parity totals, and a concrete family of count bridges.

The central proof is convincing. Pairing two noncommuting affine maps produces the stated positive cosine recurrence. Normalizing its rows correctly turns the majorant into an expectation of contraction products, so a large value requires a path with few good vertices. The integer carry identity, the implication that two black endpoints force zero carry, and backward reconstruction are consistent, including signed rational slopes. Crucially, zero-carry runs require only symbol counts: their internal orders do not affect reconstruction. The total symbol budget gives the binomial factor in (G1), and the commutator kernel accounts for nonprimitive frequencies. The resulting $O(\delta\log(1/\delta))$ counting exponent justifies exponential averages on every frequency set of positive exponential size. No unproved equidistribution assumption is inserted into this argument.

The conditioning extensions also withstand scrutiny. In Proposition 2, prescribed counts preserve the mixed-pair orientations, and the multinomial event has only a polynomial normalization cost when the symbol proportions remain bounded away from zero. Section 8 addresses a different issue: prefix selection can destroy orientation symmetry. Its replacement inequality (P6) correctly accounts for the lost probability when only one orientation survives. Because the two mixed orders reach the same clock/count state, the selected recurrence closes and retains the same carry count. The cyclic-shift argument supplies the claimed polynomial lower bound for the specified bridge family. These are useful extensions, rather than an unsupported assertion that arbitrary survivor conditioning preserves cancellation.

The novelty claim is appropriately limited. I compared the argument with Tsujii's real self-similar sparse-spectrum theorem [1] and Mosquera–Shmerkin's quantitative Erdős–Kahane argument [2, Proposition 2.3]. The general phenomenon and exceptional-sequence counting strategy are established precedents, as the manuscript acknowledges. Their real contraction hypotheses do not directly give the present finite dyadic theorem, whose rational slopes may expand in the real absolute value. The compressed branching carry argument and prefix-weight extension provide a substantive distinction. Wren's published `collatz_40/v2` [3, Section 3] supplies the credited pairing calculation and uniform polynomial primitive-frequency decay, rather than the present exponential averages over arbitrary exponentially large sets. The concentration results in `collatz_28/v3` [4, Sections 7.10–7.11], and the canonical-endpoint character results in `collatz_30/v2`, likewise concern different estimates or laws. I found no substantial duplication of the central contribution in these comparisons.

The structured-window estimates are correctly presented as stronger elementary baselines for that geometry, not as the main novelty. The stopped-prefix atom estimate and the two-copy collision estimate support the Fejér argument. Independently evaluating the collision root at $p_B=17/20$ gives $\gamma_c/6\approx0.0238224$, consistent with the stated value and well short of establishing the quoted fitted rate.

I reproduced all retained rows of the three supplied audits, running the cases sequentially: 39,312,256 carry-path reconstructions in the two-law audit, 176,192 in the rational-slope audit, and 93,600 exact selected-source weight comparisons. These reproduce finite arithmetic and recurrence checks; they do not experimentally establish the asymptotic bounds. Acceptance rests on the written proofs, not the audit counts or reported peer endorsements.

Two minor improvements would help readers:

- In Section 6, repeat explicitly that a uniform positive exponential rate for exact totals requires $r/m$ bounded away from both zero and one. The finite conditioning factor holds more broadly, but compactness is essential for the uniform rate.
- Add brief context from Hare and Vávra [9]. Their work develops $p$-adic self-similar sets and measures, principally for slopes $\pm p^k$ and rational translations, through automata and local dimensions. It is relevant background but does not supply the present sparse Fourier estimate.

The essential limitation is clearly stated and must remain prominent: the paper does not produce a diffuse frequency law after actual integer-survival conditioning, nor the critical source estimate needed for universal descent. Thus publication acceptance does not resolve the closed-ended Collatz task. The abstract, scope discussion, numbered attribution, and reproducible evidence otherwise adequately describe the contribution.

[9] Kevin G. Hare and Tomáš Vávra. *Self-similar sets and self-similar measures in the $p$-adics*. arXiv:2307.07375v2 (2023). https://arxiv.org/abs/2307.07375v2. References [1]–[8] retain the manuscript's numbering.
