{"id":"caa166c7-7fa8-49c7-8029-7eca33ae0858","arxiv_id":"2608.00662","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":5,"one_line_summary":"At genome-resolved resolution, gut-microbiome 'circadian' structure in two mice shows no persistent time-frequency ridge and fits a non-normal stochastic (pseudo-coherent) regime rather than an oscillator.","lead":"Reanalyzing hourly genome-resolved gut-microbiome data from two mice, this paper finds no persistent circadian rhythm ridge and argues the daily fluctuations are better explained by 'pseudo-coherence': noise reshaped by a one-way bacterial cross-feeding network, not by an oscillator. It matters because it challenges the standard clock-based reading of microbiome rhythms and proposes a genetic knockout experiment that could distinguish the two explanations.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Underdetermined 3-sample Jacobian may manufacture the non-normal geometry; a driven normal null could reproduce the cluster recovery.","rationale":"The paper is a serious and sophisticated application of non-normal dynamics, and its strongest independent evidence is the phase-agnostic co-membership recovery with surrogate-null control (Sec. III.D), together with the direct lagged-covariance asymmetry (Sec. III.C). But the entire positive case for pseudo-coherence — K/Kc > 1, mode supports, entropy production, and the reaction-mode signs underlying the cluster recovery — is built on Â_k estimated from three effective points in an N-dimensional system. This is where the argument is least secure. The estimator is a rank-3 projection plus an amplified noise term; the commutator's top two eigenvectors need not correspond to any real biological interaction geometry. The synthetic calibration in Fig. 9 is self-referential in that it only verifies recoverability within the same linear non-normal model class; it does not test specificity against a normal Jacobian driven by a common host signal. The surrogate null for the cluster agreement is an important safeguard, but it preserves per-MAG spectra and only randomises cross-MAG phases, so a common 24-h drive with per-MAG delays could survive it. Thus the central claim that 'the appropriate null is a stable, strongly non-normal stochastic system' is not yet established. I concur with the reader's CONDITIONAL verdict: the analysis is defensible but requires a negative-control simulation or an independent validation of the short-window Jacobian before the pseudo-coherence interpretation can be accepted.","tokens_in":24238,"tokens_out":10090,"duration_ms":107871,"concrete_test":"Generate a null dataset with the same N, T, per-MAG marginal spectra, and cross-MAG phase structure as the real data, but from a model with a stable normal Jacobian (e.g., diagonal A with negative eigenvalues) plus a shared 24-h sinusoidal driving term with per-MAG phase delays and realistic noise. Apply the full pipeline of Sec. II.e/V.d (3-sample Jacobian, commutator diagonalisation, reaction-mode sign, co-membership clustering) and compare the inferred K/Kc distribution and the agreement of the recovered partition with the true phase-delay clusters. If this null yields K/Kc > 1 and >90% agreement, the estimator cannot distinguish pseudo-coherence from a driven normal process and the central claim fails; if K/Kc ≈ 1 and agreement near the surrogate-chance level, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing assumption is the one the reader flags: the local Jacobian Â_k = Y_k X_k^+ (Eq. 10) is fit from four consecutive hourly samples, i.e., only three effective input vectors, in N = 118–181 dimensions. Because X_k is N×3, Â_k = A_true P_{X_k} + Ξ X_k^+, with P_{X_k} the rank-3 projector onto the space spanned by the three lagged state vectors. Thus Â_k is not a consistent high-dimensional estimate of A_true; it is a rank-≤3 object compressed onto local lagged directions, and the noise term Ξ X_k^+ is amplified by ||X_k^+||, which is large when hourly samples are strongly autocorrelated and X_k is ill-conditioned. The commutator B_k = [Â_k, Â_k^T] then has rank at most 6, and its top two eigenvectors — the inferred 'reaction' and 'non-normal' modes — can be dominated by the noise projection. The synthetic calibration (Fig. 9) only draws from the same linear model class and does not include a null with a shared diurnal drive and a normal (or no) Jacobian. The strongest empirical result, the 98.3%/93.4% co-membership recovery of the cyc7plus clusters (Sec. III.D), may therefore reflect the estimator's sensitivity to the shared low-frequency/diurnal structure in the data rather than to genuine non-normal amplification geometry. The AAFT surrogate null (60–63% chance agreement) only destroys cross-MAG phase coherence; it does not rule out a common host-driven signal with per-MAG phase delays, which could produce both the phase clusters and the reaction-mode signs.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reanalyzes an hourly, genome-resolved (MAG-level) mouse-gut microbiome time series from two animals and argues that the apparent circadian organization is not evidence of a persistent microbial oscillator. It proposes instead a 'pseudo-coherence' regime: a linearly stable, strongly non-normal stochastic system in which transient non-normal amplification produces intermittent phase alignment, broken time-reversal symmetry, low-frequency spectral enhancement without a stationary ridge, and phase-agnostic recovery of two functional guilds (Bacteroidota primary degraders vs. Bacillota A secondary fermenters). The core evidence is a local Jacobian estimated from four-hour windows, a rank-two commutator subspace, K/Kc diagnostics, wavelet scalograms, AAFT surrogate testing, and a co-membership clustering that recovers the cyc7plus partition with 98.3%/93.4% agreement. The paper closes with a falsifiable clock-gene-knockout prediction.","tokens_in":24706,"tokens_out":5695,"duration_ms":70237,"significance":"If the central claim held, the paper would substantively challenge oscillator-based interpretations of MAG-resolution microbiome rhythmicity and would provide a concrete statistical-mechanics alternative grounded in non-normal amplification. The study is genuinely interdisciplinary, uses public data, and proposes a discriminating experiment, which are strengths. The manuscript is also unusually candid in parts, e.g., in acknowledging that the surrogate test is not itself an oscillator discriminator. However, the current evidentiary support is not yet commensurate with the strength of the conclusions: the headline surrogate claim is contradicted by the appendix table, and the mode-inference pipeline rests on a severely underdetermined local Jacobian whose calibration does not rule out plausible nulls. The result would be important if the required robustness checks were supplied.","major_comments":[{"comment":"The local Jacobian Â_k is estimated from four consecutive hourly samples, i.e., only three lagged state vectors, in N=118–181 dimensions. Hence rank(Â_k)≤3 and the commutator B_k=[Â_k,Â_k^T] has rank at most 6, so its top two eigenvectors—the inferred 'reaction' and 'non-normal' modes—are extracted from an object that is almost entirely null space plus noise projection. The claim that this 'robustly extracts' the non-normal amplification is not justified: the noise term ΞX_k^+ is amplified by ||X_k^+||, which is large for strongly autocorrelated hourly samples. The synthetic calibration of Fig. 9 draws from the same linear model class and does not include a null with a shared diurnal drive and a normal (or zero) Jacobian. Because every mode-based result (K/Kc, supports, co-membership clusters, guild identities) depends on this estimator, the paper needs (i) a bias/variance analysis of th","section":"Sec. II.e, Eq. (10); Sec. III.D"},{"comment":"The main text states that the AAFT surrogate test 'does establish in both animals that the low-frequency band carries a marginal excess over the AAFT null.' Table S5 reports that after Benjamini–Hochberg correction at FDR=0.05, Mouse A has no significant bin in either the full record or the post-72h window, and Mouse B is significant only in the full record (two bands: ~22–28h and ~79–95h), with no significant bin after removing the first 72h. Thus the surrogate test does not establish an excess in both animals, and the post-transient record shows no significant excess in either mouse. This directly contradicts the abstract's 'AAFT surrogate test identifies a weak time-averaged construction in the candidate circadian band.' The claim must be weakened or the analysis changed. The 1/f^4 tail in Fig. 4 is also presented as validation without any quantitative goodness-of-fit or significance","section":"Sec. III.B; Appendix Table S5"},{"comment":"The co-membership surrogate null (chance agreement 60.2% in Mouse A and 63.5% in Mouse B) is built from AAFT surrogates that destroy inter-MAG phase coherence but preserve each MAG's marginal spectrum. Such a null does not rule out the most relevant alternative: a common host-driven low-frequency/diurnal signal with per-MAG phase delays, in a system with a normal or zero Jacobian. Under that alternative, both the phase-based cyc7plus clusters and the reaction-mode sign series could align without any non-normal amplification geometry. To make the cluster-recovery result load-bearing, the paper should add a null in which a shared slow/diurnal envelope (with random per-MAG phase delays) is superimposed on independent AR(1) or OU processes, then run the full co-membership pipeline; if agreement remains near 98%, the 'phase-agnostic recovery' does not support pseudo-coherence.","section":"Sec. III.D; Sec. V.k"},{"comment":"The entropy-production proxy Σ_local(t) is defined as the small-τ slope of I(τ,t). Because I(τ) is a norm of an antisymmetric matrix, it is nonnegative and typically increasing at small τ, so 'strictly positive at 100% of windows' is not, by itself, evidence of a non-equilibrium steady state. The identification of Σ_local with the geometric non-normality index K_σ and the entropy production Φ rests on the closed forms of the reduced 2×2 model, whose parameters (α, β, κ, σ₁, σ₂, ρ) are not estimated with uncertainties, and which is derived from the same noisy Jacobian. The claim that the two mice agree to 0.5% is based on n=2 animals and is not a meaningful consistency check. The authors should either provide a direct validation of the exponential form in Eq. (18) against the empirical C(τ) (e.g., a fit of the full τ-dependence with confidence intervals) or soften the thermodynamic interp","section":"Sec. III.C, Eqs. (18)–(20)"}],"minor_comments":[{"comment":"References [12] and [39] are the same paper (R. Muolo, M. Asllani, D. Fanelli, P. K. Maini, T. Carletti, J. Theor. Biol. 480, 81 (2019)) and should be merged or renumbered.","section":"References"},{"comment":"The abstract contains a typo: 'constructionin the candidate circadian band' should be 'construction in the candidate circadian band.'","section":"Abstract"},{"comment":"The 1/f^4 guide line is described as validating the pseudo-coherent picture 'independently of the low-frequency band,' but no statistical test of the high-frequency slope is reported. Please state that this is a qualitative guide and add a quantitative slope comparison or remove the word 'independently.'","section":"Fig. 4"},{"comment":"The forward reference to 'Sec. IIID' (with Roman numerals) before Section III.D is introduced is confusing; use 'Section III.D' consistently and define the section before referencing it.","section":"Sec. III.C.a"},{"comment":"The wavelet coherence values in Table I (0.55–0.70) are presented without confidence intervals or a null-comparison. Given the nonstationary and autocorrelated data, a significance test or at least a bootstrapped CI would help the reader assess the 'support controls coherence' claim.","section":"Table I and Sec. III.A.a"}],"recommendation":"major_revision","confidential_remarks":"The paper is thought-provoking but currently overclaims on two fronts: the AAFT surrogate result contradicts its own appendix, and the core mode-inference pipeline is too underdetermined to support the strong conclusions without additional null models. I would also note that the framework, closed forms, and order-parameter interpretation rely heavily on the authors' own companion papers (Refs. [13–16]); the manuscript would be more convincing if the required kernel of those results were stated in the supplement. I recommend major revision with emphasis on the surrogate controls and a more cautious phrasing of the spectral and entropy-production claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things before reading. The paper's negative result — no persistent circadian oscillator at MAG resolution, with pseudo-coherence as the alternative null — is sharper than most of the literature, and the phase-agnostic recovery of two trophic guilds is genuinely interesting. But the entire empirical machinery rests on a local Jacobian fit from four hourly samples in N=118–181 dimensions, which makes the inferred non-normal geometry, including the guild recovery, potentially an estimator artifact.\n\nWhat is actually new: the application of the authors' non-normal-dynamics framework to genome-resolved gut time series, with a battery of diagnostics tailored to dissociate oscillatory from non-normal stochastic structure. The lead-lag covariance asymmetry I(τ) with a peak at tens of hours is a direct datum, not dependent on the Jacobian. The co-membership cluster recovery (98.3%/93.4% agreement with the cyc7plus partition, 24–38 points above a surrogate null) is a clever and unusual test, and it does suggest the data contain some low-dimensional shared structure. The falsifiable clock-knockout prediction is a good way to end.\n\nThe soft spots are in proportion to how much they matter. First, the main text claims the surrogate test establishes a low-frequency excess 'in both animals', but Appendix Table S5 reports no FDR-significant bin for Mouse A in either window, and for Mouse B only in the full record, not after the 72-h exclusion. That is a direct internal contradiction, and the abstract's wording inherits it. Second, the local Jacobian Â_k = Y_k X_k^+ uses three effective input vectors in a space of 118–181 dimensions. Â_k is then a rank-≤3 object, its commutator has rank at most 6, and the top two eigenvectors can be dominated by noise amplified through X^+. The synthetic calibration (Fig. 9) draws from the same linear model class the theory assumes, so it cannot rule out a common diurnal drive producing both the phase clusters and the reaction-mode signs under a normal (or zero) Jacobian. The surrogate null for the cluster agreement only destroys cross-MAG phase coherence; it does not exclude a shared host-driven signal with per-MAG delays. Third, the data source is an overlapping-author preprint used as both the empirical input and the external benchmark, and the theoretical tools are mostly the authors' own prior work. None of that is disqualifying by itself, but it raises the circularity burden.\n\nWho gets value from this? Someone working on circadian rhythms in microbial communities, or on non-normal stochastic dynamics, should read it as a provocative alternative to the oscillator default — but with the understanding that the central inference is not yet supported. It would make a good reading-group case study in underdetermined high-dimensional inference. I would not cite it in the next year until the Jacobian estimator is validated against a nonlinear generative model or the code/data are released.\n\nMy recommendation: send it to peer review. The claim is important enough and the methods novel enough that a careful referee should have a chance to sort out the spectral contradiction and push for an independent validation of the Jacobian. Expect heavy revision.","headline":"A provocative negative claim about microbiome circadian rhythms that hangs on a 3-sample Jacobian — intriguing cluster recovery, but the inference is too underdetermined to carry it.","tokens_in":25272,"tokens_out":3607,"would_cite":false,"duration_ms":43345,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper argues that apparent circadian rhythms in genome-resolved gut microbiome data stem from transient non-normal amplification of noise in a stable system, not from microbial oscillators.","keywords":["circadian rhythms","gut microbiome","non-normal dynamics","pseudo-coherence","metagenome-assembled genomes","lead-lag imbalance","cross-feeding guilds","time-frequency analysis"],"falsifier":"A persistent ridge in the Morlet scalogram of either animal, surviving long-time averaging, would falsify the no-oscillator claim; the decisive test is a clock-gene knockout cohort where pseudo-coherence predicts a weakened 24-hour component but preserved transient amplification and lead-lag asymmetry, while an autonomous-oscillator picture predicts a surviving coherent near-24-hour ridge.","tokens_in":24043,"feed_emoji":"🦠","tokens_out":6789,"duration_ms":75158,"temperature":0.7,"pith_summary":"This paper argues that the apparent circadian rhythm in genome-resolved gut microbiome data is not produced by microbial oscillators. Re-analyzing hourly two-week mouse-gut time series at the level of metagenome-assembled genomes, it finds none of the standard signs of a sustained oscillator: the time-frequency scalograms have no persistent ridge, the lagged covariance is strongly time-asymmetric with an imbalance peak at tens of hours, and surrogate tests show only a weak time-averaged low-frequency excess. The authors propose that the observed intermittent synchrony, two-cluster structure, and low-frequency spectral build-up are signatures of a different regime: a linearly stable but strongly non-normal stochastic system in which transient geometric amplification funnels noise into a low-dimensional 'reaction' subspace. If correct, this changes the default interpretation of microbiome rhythmicity from host-entrained or autonomous oscillators to a trophic cross-feeding architecture that amplifies fluctuations, and it predicts a specific dissociation in clock-gene-knockout experiments.","feed_headline":"Gut microbiome's daily rhythm may be noise amplification, not a clock","feed_subtitle":"Hourly genome-resolved mouse-gut data show transient amplification, not a persistent circadian ridge.","key_machinery":"The central device is the commutator of the estimated local Jacobian, B = Â Âᵀ - Âᵀ Â, computed from short sliding windows of four hourly samples. Because the dynamics are assumed to have real negative eigenvalues, non-orthogonal eigenvectors make perturbations transiently grow; diagonalising this traceless symmetric commutator isolates the rank-two subspace of maximal non-normal amplification and defines the two modes — the non-normal mode that injects noise and the reaction mode onto which fluctuations are redirected. All diagnostics (support, phase coherence, lead-lag imbalance, and the phase-agnostic co-membership recovery) flow from this rank-two geometric object.","core_discovery":"At genome resolution, both animals' time series carry no persistent time-frequency ridge; the time-averaged spectrum is enhanced at low frequencies and depleted at intermediate frequencies; the lagged covariance is markedly asymmetric with a peak near tens of hours; and a phase-free clustering based only on the sign of an inferred reaction mode recovers the same two functional guilds (primary polysaccharide-degrading Bacteroidota versus secondary butyrate/propionate-fermenting Bacillota A) that a phase-based method finds, with 98.3% and 93.4% agreement, well above surrogate null maxima. The paper interprets this conjunction as evidence for pseudo-coherence: geometric, non-normal amplificatio","pith_inferences":["Because the estimator relies on short windows, a stress test worth running is re-estimation at denser sampling or with a state-space filter; if the rank-two structure is stable, the geometric reading is strengthened.","The re-reading of 'night' and 'day' clusters as a cross-feeding delay suggests that published circadian microbiome datasets sampled at lower resolution may have been interpreting a trophic pipeline delay as a clock; re-running these diagnostics on such data would reveal how general the pseudo-coherent regime is.","The same diagnostic battery could be applied to neural or ecological data with directed interactions, predicting that their rhythmic-looking output may also be pseudo-coherent rather than oscillatory."],"forward_implications":["MAG-level rhythmic structure should not be read as oscillator evidence: the null hypothesis to beat is a stable, strongly non-normal stochastic process.","The two anti-phased microbial clusters become the upstream and downstream halves of a directed cross-feeding cascade, not circadian subpopulations.","A positive, quasi-stationary entropy production rate follows from the reaction-mode support, placing the microbiome in a non-equilibrium steady state sustained by host-supplied free energy.","Clock-gene-knockout cohorts should show a selective dissociation: the 24-hour clock-locked component weakens while transient amplification, lead-lag asymmetry, and mode-based guild recovery persist."],"fun_headline_variants":["Gut microbiome's daily rhythm may be noise amplification, not a clock","Microbiome cycles are pseudo-coherence: no persistent oscillator exists","Daily gut patterns aren't a clock—just transient amplification","Gut rhythm is an illusion: no circadian ridge in genome-resolved data"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The argument rests on local interaction matrices inferred from only three effective hourly differences in over a hundred species; if those matrices are dominated by estimation noise, the inferred non-normal modes and the cluster recovery built on their signs would be artifacts.","fun_headline_variants_meta":{"raw":{"variants":["Gut microbiome's daily rhythm may be noise amplification, not a clock","Microbiome cycles are pseudo-coherence: no persistent oscillator exists","Daily gut patterns aren't a clock—just transient amplification","Gut rhythm is an illusion: no circadian ridge in genome-resolved data"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000665,"raw_usage":{"total_tokens":2901,"prompt_tokens":804,"completion_tokens":2097,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":548,"completion_tokens_details":{"reasoning_tokens":2021}},"tokens_in":548,"tokens_out":2097,"duration_ms":16073,"temperature":1.0,"reasoning_tokens":2021,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T01:00:49.578266+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A persistent ridge in the Morlet scalogram of either animal, surviving long-time averaging, would falsify the no-oscillator claim; the decisive test is a clock-gene knockout cohort where pseudo-coherence predicts a weakened 24-hour component but preserved transient amplification and lead-lag asymmetry, while an autonomous-oscillator picture predicts a surviving coherent near-24-hour ridge.","supporting_citations":[],"review_version":1}