Written answers to the first two reviews, September 29, 2026

Reconciliation of the two reviews of "Approximate Antiunitary Symmetry as a Matching Problem"

Applied 2026-09-29 (about 13:20 EDT) by Claude (Fable 5.1) on the owner's instruction ("go ahead and handle the edits"), against baseline 9820d1e (the manuscript as Astra wrote it). The two reviews were produced independently and neither reviewer read the other; this file is the first place they meet. Astra's authorship of the manuscript is unchanged; every edit below is a review-driven change of exposition or hypothesis, not new mathematics, and the two reviews' own new results stay in their directories (reviews/openai/extensions.md, reviews/claude/04-new-results.md, reviews/claude/extension-note.pdf, and the Gil note).

Where the two reviews agree

Finding	OpenAI (reviews/openai/REPORT.md, claim-audit.md)	Claude (reviews/claude/REPORT.md, 06-claim-audit.md)
Main theorem, hull lemma, every corollary	correct, no substantive repair	correct, no false statement
max outside math mode, eq. (benefits)	fix the missing backslash	typo (#23)
Rigidity proposition needs only a unique optimal matching	"a unique optimal matching alone implies the complete phase-decorated form"	c_ij > 0 redundant (#13, Prop F)
Hard-constraint corollary needs no sign condition on c	(implicit in its audit)	#21
Gil, Entropy 28(8):877 (2026) is directly relevant and was not cited	the main weakness is positioning	#32
Single-observable case is classical	(not stated)	Wigner + Kramers + Mirsky (Prop D), #32
The computation is a polynomial-size matching problem; the odd-set LP is exponential	presentation change	(agrees implicitly)

No finding of either review contradicts the other.

Edits made to main.tex

1. Typo (both). Eq. (benefits): -max_{M} became -\max_{M}.
2. Polynomial-size wording (OpenAI). Section 4.1: the matrix problem is stated as a maximum-weight matching problem on a polynomial-size graph, solvable by Edmonds' algorithm; the odd-set description is described as exponential and used only for proofs and certificates.
3. Rigidity hypothesis (both). Proposition 3.2: the hypothesis c_ij > 0 was dropped from the exact-structure statement; uniqueness of the optimal matching suffices. The proof was completed accordingly: on a matched pair |K_ij| = 1 forces S_ij = 0 and unit entries through |U_ij|^2 + |U_ji|^2 = 2|S_ij|^2 + 2 <= 2; two unmatched vertices have c_ij > 4 tau by uniqueness, hence S_ij = 0. The second stability estimate keeps its own hypothesis c_min > 0, stated inline. The tie example n = 3, c = 0, tau = 1 (rotations by ninety degrees about any axis are minimizers) is named as the reason uniqueness is needed.
4. Hard constraint (both). Corollary 3.5 now states that no sign condition on the off-diagonal costs is needed, with the one-line reason (S = 0 under U^T = -U, the objective is linear).
5. Framing (Claude, consistent with OpenAI's "positioning"). Section 1: a paragraph states that the single-observable case follows from Wigner's normal form, Kramers' theorem and Mirsky's inequality and reduces to the sorted-spectrum recurrence, that the substantive content is the general cost array (tuples of two or more observables) carried by the hull lemma, and that the lemma bears on Gil's coherence maximization, an application both reviews found independently and which is developed separately.
6. Related work (both). Section 5.3: the related-work paragraph records that two independent reviews also failed to find the identity or the hull lemma in print, found Gil (2026), and found older work on congruence orbits with prescribed singular values containing the single-observable ingredients.
7. Caveat (Claude #31). Section 5.3: for a single observable the value depends on consecutive spectral gaps alone, so it certifies spectral pairing, not the symmetry of a prescribed antiunitary.
8. Bibliography. Added Kramers (1930), Mirsky (1960), Gil (2026).

Not changed: the abstract's "certified alternative to continuous symmetry searches on finite spectral data" (Claude #33 accepts it with the caveat now stated in Section 5.3); the numerical section (both reviews reproduced its counts; OpenAI's rerun gave 82 optimizer success flags against the recorded 80 with the same 96 accurate outcomes, a dependency-version effect noted in its report, so the recorded numbers stand as the record of the authoring run); the provenance and attribution paragraphs (Astra's).

Left open, deliberately

• Priority: both reviews searched and found nothing; neither claims an exhaustive search.
• Whether the four-level fixed-defect problem's exact value can be found (OpenAI, "strict relaxation gap").
• The Gil application is Paper 3 (written at reviews/claude/gil-note/, moved to paper/gil-note/ on 2026-10-01), under its own reviews.

Rebuilt after the edits: 9 pages, no warnings, no em dashes; verification scripts untouched because no numerical claim changed.

Addendum, 2026-10-01 (Claude Fable 5.1): the Looi paragraph

The outreach plan's datedness check of 2026-10-01 (docs/outreach/2026-10-02/PLAN.md, section 2) found one paper adjacent to this manuscript and posted after both reviews: S. Looi, "Stability of isometries of quantum states", arXiv 2609.37133, posted 2026-09-29 (41 pages, math.FA). It was read in full on 2026-10-01 from the arXiv PDF before anything was written. What it proves: dimension-independent Hyers-Ulam stability for surjective approximate isometries of the positive trace-class cone in the Bures metric (within 2 sqrt(2) epsilon of a Wigner symmetry, Theorem 1.2) and the trace metric (within 9 epsilon / 2, or 3 epsilon when zero is fixed, Theorem 1.3 and Corollary 1.4), with no linearity, affinity or continuity assumed on the maps; on the state space, instability in infinite dimension (Theorem 1.5) and a modulus in each finite dimension that cannot be chosen uniformly in the dimension (Corollary 1.6). Its subject is a map on states and how close it is to one conjugation by a unitary or antiunitary (the Molnar-Timmermann classification made quantitative). This manuscript's subject is a fixed tuple of observables and the least error over all antiunitaries in commutation together with the square relation, given exactly by a matching. No observable, commutation error or square relation appears in Looi's theorems; no distance between states or approximate isometry appears here; the shared ancestor is Wigner's theorem. No priority claim is made in either direction.

Edits (three, no change to any statement or proof): one paragraph at the end of Section 1 that cites and distinguishes the paper as above; one sentence at the end of Section 5.3 recording the 2026-10-01 search and its one adjacent find; the bibliography entry looi. The attribution block records the 2026-10-01 addition in one clause. LITERATURE.md carries a dated line. Rebuilt with tectonic; page count recorded in paper/STATUS.md (the refresh of 2026-10-01/02).

Consistency pass, 2026-10-01 (fresh Claude Opus 5.5 agent, directed by the Claude Fable 5.1 outreach thread)

A bounded pass, begun 2026-10-01 at 23:09 EDT, checking the abstract and Section 1 against the theorems, the Looi paragraph and sentence against Looi's text, quotations and source citations, the attribution block against paper/common/README.md, and the house language rules; it is not a third review of the mathematics.

#	check (1-6)	item checked	source and page	command or method	finding	verdict	what was done
1	1	Abstract, sentence 1: a commuting tuple of Hermitian matrices; least combined error in commutation and T^2=-I	main.tex Cor. 3.1 (hypothesis), eq. (1)	read side by side	hypothesis and quantity match	NO FINDING	none
2	1	Abstract formula: the objective over U(n) equals the minimum over all matchings of 2 sum dist(lambda_i, lambda_j)^2 + 4 tau (n - 2 #M), with Euclidean dist	Thm. 2.1 eq. (3); Cor. 3.1 eq. (8) with c_ij = dist(lambda_i, lambda_j)^2	term-by-term comparison	factor 2, 4 tau, n - 2 #M, empty matching included, minimum attained; tau >= 0 is implicit in the abstract's "combined error" and stated at (1)	NO FINDING	none
3	1	Abstract optimizer: signed swaps on matched pairs, scalar phases on unmatched states	Thm. 2.1 (block [[0,1],[-1,0]] per pair, entry 1 per unmatched vertex, "an optimizer"); Prop. 3.2 (unit scalars)	read	existential claim; entry 1 is a phase, and Prop. 3.2 allows every phase	NO FINDING	none
4	1	Abstract: squared entries of a skew-symmetric contraction give a point of Edmonds' polytope	Lemma 2.2	read	matches (K^T=-K, operator norm at most 1)	NO FINDING	none
5	1	Abstract: odd-set inequalities supply the parity obstruction missing from a degree-only relaxation	Remark 2.3 (n=3, c=0, tau=1)	read	matches	NO FINDING	none
6	1	Abstract: "generic optimizer rigidity"	Prop. 3.2 after the 2026-09-29 reconciliation: the only hypothesis is a unique optimal matching (c_ij > 0 dropped)	read	a tie between two matchings is one of finitely many hyperplanes in (c, tau), so uniqueness is generic; the abstract adds no positivity or simple-spectrum hypothesis	NO FINDING	none
7	1	Abstract: "stability estimates"	Prop. 3.2 eq. (9): first estimate under uniqueness and n >= 2, second also under c_min > 0; Prop. 3.5 eq. (14)	read	plural matches; the abstract states no hypothesis that contradicts either	NO FINDING	none
8	1	Abstract: "exact dual certificates"	Sec. 4.1: dual (16), bound B, exact rational check, three-vertex certificate (z=8, B=8, energy 4)	read; three-vertex arithmetic redone	matches	NO FINDING	none
9	1	Abstract: "linear-time recurrence for one sorted spectrum"	Prop. 4.1 eq. (17), O(n) once sorted	read	matches	NO FINDING	none
10	1	Abstract: classical odd-dimensional defect bounds and Kramers pairing as boundary cases	Cor. 3.3 eqs. (10) to (12); Cor. 3.4 eq. (13)	read	matches	NO FINDING	none
11	1	Abstract: "certified alternative to continuous symmetry searches on finite spectral data"	Sec. 4.1 (no continuous unitary search); Sec. 5.1 (BFGS exact in 96 of 120 attempts, a local optimizer is not a certificate); Sec. 5.3 caveats	read	matches; kept as accepted on 2026-09-29 (Claude #33)	NO FINDING	none
12	1	Sec. 1 para. 1: T = U C, T^2 = U Ubar, error HU - U Hbar; exact commutation with T^2=-I forces even multiplicities	Cor. 3.3 eq. (12)	direct computation; read	the value is 0 exactly when no multiplicity is odd	NO FINDING	none
13	1	Sec. 1 para. 2: definition (1) with tau >= 0, unnormalized Frobenius norms, no scalar-square assumption	Thm. 2.1 (tau >= 0); Prop. 3.5 (same tau >= 0)	read	matches	NO FINDING	none
14	1	Sec. 1 para. 3: run-18 observations (odd-dimensional Frobenius defect 2, negative square as skew symmetry, eigenbasis conjugations)	Sec. 5.2 nodes 2649, 2676, 2677; Cor. 3.3 (the unsquared Frobenius residual has minimum 2 in odd dimension)	read	matches	NO FINDING	none
15	1	Sec. 1 para. 4: contribution; single-observable case classical, hard or soft, reducing to Prop. 4.1; substance of Thm. 2.1 is the general cost array carried by Lemma 2.2	Prop. 4.1 eqs. (17) (soft) and (18) (hard, even n); Cor. 3.4	read	matches; "the value of (1)" under the hard constraint reads as the constrained problem of Cor. 3.4 and eq. (18)	NO FINDING	none
16	1	Sec. 1 para. 4: Gil maximizes coherence at prescribed populations over a restricted class of antisymmetric parts, and the hull lemma removes the restriction	Gil, Entropy 28(8):877, Sec. 3.4 (maximum over the aligned class; global optimality over arbitrary Youla orientations left open there); reviews/claude/03-literature/gil-2026-fulltext.txt lines 199 to 231	grep -n -i for aligned, Youla	matches; the aligned class is the restriction	NO FINDING	none
17	1	Sec. 1, Looi paragraph: "the data are a fixed tuple of observables, ... which Theorem 2.1 gives exactly as a matching value"	Cor. 3.1 (commuting hypothesis); Sec. 5.3 (the exact spectral formula requires a finite commuting Hermitian tuple)	read side by side	the commuting hypothesis was missing; for a noncommuting tuple only the bounds of Prop. 3.5 hold	CORRECT, applied	"commuting" inserted: "a fixed commuting tuple of observables"
18	2	Posting date: Sec. 1 (September 29, 2026, after this manuscript was written on September 28); Sec. 5.3 (posted after the reviews); attribution (posted 2026-09-29)	Looi p. 1 stamp arXiv:2609.37133v1 [math.FA] 29 Sep 2026; arxiv.org/abs/2609.37133 history: v1 Tue, 29 Sep 2026 09:36:14 UTC	pdftotext -enc UTF-8 -layout -f 1 -l 1; curl -sSL https://arxiv.org/abs/2609.37133	v1 is 05:36 EDT on 09-29, after the 23:20 EDT 09-28 snapshot and after both review reports (file times 09-28 23:52 and 09-29 00:40 EDT)	NO FINDING	none
19	2	Bures constant 2 sqrt(2) epsilon for a surjective cone map, zero not fixed	Looi Thm. 1.2, eq. (1.4), p. 5	pdftotext -enc UTF-8 -layout -f 5 -l 7	matches	NO FINDING	none
20	2	Trace constants 9 epsilon/2, and 3 epsilon when zero is fixed	Looi Thm. 1.3, eq. (1.6) and the sentence after it, p. 5; Cor. 1.4, eq. (1.8), p. 5 (also under delta-surjectivity)	same	matches	NO FINDING	none
21	2	Hypothesis and conclusion: a surjective map of the positive trace-class cone preserving the Bures or trace distance up to additive error epsilon lies uniformly near a conjugation A -> UAU*, U unitary or antiunitary	Looi eq. (1.3) and the stability problem, p. 4; Thms. 1.2 and 1.3, p. 5; Wigner symmetry defined on p. 3	full pdftotext -enc UTF-8 -layout read	matches	NO FINDING	none
22	2	"with no linearity or continuity assumed"	Looi p. 4 (no linearity, affinity or continuity assumption anywhere in the paper)	grep -ci affinity (1 hit, p. 4)	accurate paraphrase, not in quotation marks; dropping "affinity" only weakens the claim	NO FINDING	none
23	2	"constants independent of the dimension" for the cone	Looi abstract, p. 1 (dimension-independent stability); p. 4 (cone bounds with constants independent of dimension)	read	matches	NO FINDING	none
24	2	State space: no modulus uniform in the dimension	Looi Thm. 1.5, eqs. (1.9) and (1.10), p. 5 (infinite dimension); Cor. 1.6, p. 6 (a modulus in each finite dimension); p. 4 and the abstract (none uniform in the dimension)	read	matches	NO FINDING	none
25	2	Exact classification attributed to Molnar and Timmermann: bijective isometries of the cone and the state space, Bures and trace, are Wigner symmetries; Looi's results are its quantitative form	Looi Thm. 1.1 (Molnar-Timmermann), p. 3, citing their Theorems 1 to 4; p. 4 (quantitative versions of Thm. 1.1)	read	matches	NO FINDING	none
26	2	"no observable, commutation error or square relation enters Looi's theorems"	all 41 pages; the statements of Thms. 1.1 to 1.5 and 4.1, Cors. 1.4 and 1.6, Lemmas 3.1 to 3.5, 4.2, 4.3, 5.1 to 5.4	pdftotext -enc UTF-8 -layout, then grep -ci with page numbers: observ 2, commut 5, square 12, matching 0, time.revers 0, Kramers 0	observ: one motivating sentence of the Introduction (p. 1) and "observe" (p. 37); commut: inside proofs (pp. 19, 25, 36) and a reference title (p. 41); square: square roots and squared quantities; none in a statement	NO FINDING	none
27	2	"what the two share is Wigner's theorem behind the unitary-or-antiunitary dichotomy"	Looi pp. 1 and 3 (the conclusion is a Wigner symmetry, a unitary or antiunitary conjugation; Wigner's theorem is not named)	grep -ci wigner (5 hits)	interpretive sentence about shared background, consistent with Looi's definitions	NO FINDING	none
28	2	Sec. 5.3 sentence: a search on 2026-10-01 for work posted after the reviews found one adjacent paper and nothing stating the identity	docs/outreach/2026-10-02/PLAN.md section 2 (receipts 2026-10-01 20:38 to 22:00 EDT); LITERATURE.md addendum (Liu et al., arXiv 2610.01555, judged not adjacent)	read	matches	NO FINDING	none
29	2	Bibliography entry looi: S. Looi, Stability of isometries of quantum states, arXiv:2609.37133, 2026	Looi p. 1; title of the abs page	read; grep -o '<title>'	matches	NO FINDING	none
30	2	Attribution clause: "on 2026-10-02 added one paragraph in Section 1 and one sentence in Section 5"	file times: Looi PDF saved 2026-10-01 22:46 EDT, main.tex last written 2026-10-01 23:04 EDT before this pass (02:46 and 03:04 UTC on 10-02); clock at the start of this pass Thu Oct 1 23:09 EDT	ls -la; date; TZ=America/New_York date	the addition was made on 2026-10-01 in Eastern time; 2026-10-02 is its UTC date and the ship-target date; the house convention is Eastern	CORRECT, not applied (the same date stands in the 2026-10-02 addendum above, which this pass may not rewrite, in LITERATURE.md, and in the heading set for this section; changing the paper alone would split one record into two dates)	reported to the directing thread, which owns every copy of the date
31	3	Quotations attributed to a source	main.tex	grep for paired backticks, paired single quotes and the double-quote character	no quotation marks in the file; nothing is quoted from a source	NO FINDING	none
32	4	Page, section, theorem or equation numbers of sources cited in the text	all eight cited works at every \cite site	read; grep -n cite	no source is cited by page, section, theorem or equation (Gil and Looi are cited without numbers; Looi's theorem numbers appear only in this ledger)	NO FINDING	none
33	4	Edmonds entry (J. Res. NBS Sec. B 69B(1-2):125-130, 1965, title) and the attributions of eq. (4), the dual (16) and Edmonds' algorithm	NIST PDF: Vol. 69B Nos. 1 and 2, January-June 1965, printed pp. 125 to 130 (6 pages); odd-set inequality (3) over sets of 2r+1 nodes, Theorem (P), and the dual with node variables y and odd-set variables z on p. 126; weighted matching algorithm in the abstract, p. 125	curl -sSL the NIST URL; pdftotext; grep -n	matches	NO FINDING	none
34	4	Higham entry: Gover and Barnett, eds., Applications of Matrix Theory, pages 1-27, Oxford University Press, 1989	reprint footnote on p. 1 of the PDF at the bibliography URL	curl -sSL; pdftotext	matches	NO FINDING	none
35	4	Gil entry: Entropy 28(8):877, 2026, doi 10.3390/e28080877, title	header of gil-2026-fulltext.txt (PMC13512070, published 2026 Aug 4)	head	matches	NO FINDING	none
36	4	Loring entry and "a recent constructive treatment"	Loring PDF p. 1 (arXiv:2508.13004v2, 20 Aug 2025; the abstract calls Wigner's method constructive and gives two algorithms)	pdftotext -f 1 -l 1	matches	NO FINDING	none
37	4	Wigner 1960, Kramers 1930, Mirsky 1960 entries	not fetched	per the brief: no quotation or in-text page citation depends on them, and none was found	bibliographic data not verified against the sources	NO FINDING	none
38	5	Both reviews named, each verdict in one clause	reviews/openai/REPORT.md (no substantive repair, two presentation changes); reviews/claude/REPORT.md Verdict (everything proved is correct, two hypotheses stronger than needed, one typo, single-observable case classical); README pattern items 3 and 4	read	matches	NO FINDING	none
39	5	"applied in full", "eight edits, no new mathematics", 2026-09-29	this file, sections above (applied 2026-09-29 about 13:20 EDT; eight edits)	read	matches	NO FINDING	none
40	5	"No human mathematician has reviewed this paper."	paper/common/README.md pattern item 4	grep	present verbatim	NO FINDING	none
41	5	Place in the set; no paper called "the first"	paper/common/README.md item 5 and the set table; sibling blocks in paper/unfolded-zeros/main.tex (after Astra's paper) and paper/gil-note/main.tex (the last of the three)	grep -n -i first (4 hits: a proof step, an estimate, a recurrence, "first version"); read	"one of three manuscripts ... the earliest of the three" agrees with both siblings	NO FINDING	none
42	5	Last clause: the others "were written without reading it while it was being written"	paper/gil-note/main.tex attribution (it found Gil's questions while reviewing this paper and reuses its hull lemma); README item 5 (a paper states that it was written without reading the others)	read	read naturally, false for the Gil note and vacuous under the other parse; README item 5 not stated	CORRECT, applied	clause now reads "came after it, and it was written without reading them"
43	6	US English	main.tex	Python scan for the British -our, -re, -ise and -yse families, doubled-l inflections and about twenty British-only words; non-ASCII scan	no hit; the file is pure ASCII	NO FINDING	none
44	6	No em dash	main.tex; rendered main.pdf	grep for U+2014 and for three hyphens in the source; pdftotext -enc UTF-8 main.pdf - with grep -c for U+2014	none in the source, 0 in the rendered text; -- only in ranges and paired names (Hyers--Ulam, Molnar--Timmermann, population--coherence)	NO FINDING	none
45	6	No claim on the Riemann hypothesis, zeta zeros or infinite-dimensional operators beyond Sec. 5.3	main.tex	grep -n -i for riemann, zeta, infinite	hits: the Sec. 5.3 disclaimer and the attribution's name of the Riesz-basis paper; the Looi paragraph reports Looi's dimension-free cone bounds as Looi's and claims nothing for this paper	NO FINDING	none
46	6	No Clay Mathematics Institute (A)/(B) statement	main.tex	grep -n -i for clay, millennium	no hit	NO FINDING	none

Build after the two edits: bash build.sh (tectonic) wrote main.pdf, and the log reads "Output written on main.xdv (10 pages, 64916 bytes)"; pdftotext main.pdf - | grep -c $'\f' gives 10; the warnings are the two of the previous build (the ../common/hypnos-paper package-name notice and the inputenc notice), with no Overfull or Underfull box.