{"id":"c17b9dff-4812-48be-914c-3d400b3a1c7f","arxiv_id":"2607.26798","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Navigation performance in shallow gradients equals a closed-form function of the feedforward and feedback transfer entropies, holding for simulated E. coli without fitted parameters.","lead":"This paper derives exact relations between a cell's navigation performance and the flow of information from its sensors and back through its actuators, and shows these relations predict simulated E. coli chemotaxis without fitting. If correct, it means navigation in shallow gradients is controlled by two information flows alone, not by the cell's detailed molecular parameters.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equation 3, the spatial-sensing BEST, is internally inconsistent: as TFB→0 the printed expression diverges while performance must vanish; direct algebra from Eq. 1 yields 4 TFF √TFB, not Eq. 3.","rationale":"The reader's weakest assumption focused on the E. coli test and the single-exponential H timescale. That is a legitimate concern about extrapolation to real bacteria, but it is secondary. The most load-bearing issue is internal to the theory: Equation 3, the spatial-sensing BEST, is demonstrably wrong as printed. The limit TFB→0 should give P→0, but the printed equation diverges. This is not a matter of interpretation or parameter selection; it is an algebraic inconsistency that undermines the paper's central claimed equality for spatial sensing. The temporal-sensing BEST (Eq. 7) appears consistent, and the weak-feedback limit 4TFF√TFB follows from Eq. 1, so the general framework may be salvageable, but the paper as written cannot be accepted without correcting Eq. 3 and re-verifying the derivation. The proposed concrete test—re-deriving Eq. 3 from Eq. 1—will settle whether it is a typographical error or a deeper flaw. The verdict should remain CONDITIONAL, with the explicit condition that Eq. 3 be corrected and validated.","tokens_in":17584,"tokens_out":13459,"duration_ms":127455,"concrete_test":"Re-derive Eq. 3 by substituting TFF = [sqrt(1+...)-1]/2 and TFB = [sqrt(1+...)-1]/2 into Eq. 1, taking k→0, and simplify. Check whether the result is Eq. 3 or 4 TFF√TFB. Also evaluate Eq. 3 at TFB=0; if nonzero/infinite, the equation is wrong.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Equation 3 is the first central result, stating that in shallow gradients P/P0 = TFF * (1/2)[sqrt((2TFB+1)^2−1) + 1/sqrt((2TFB+1)^2−1)]. This cannot be correct. For small TFB (weak actuation gain J), sqrt((2TFB+1)^2−1) ≈ 2√TFB, so the bracket ≈ √TFB + 1/(4√TFB), which diverges as TFB→0. Thus the right-hand side diverges while the performance P must vanish when the actuator is turned off (J=0 ⇒ TFB=0). The stated weak-feedback bound in Eq. 4 is P/P0 ≈ 4 TFF√TFB, and direct substitution of the transfer-entropy formulas into Eq. 1 in the small-k limit also gives 4 TFF√TFB, not Eq. 3. The printed equality is therefore internally inconsistent; the correct bracket appears to be 2√((2TFB+1)^2−1)/(2TFB+1), which reduces to 4√TFB for small TFB. This error affects the paper's headline claim for spatial sensing and must be resolved.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops two analytic minimal models of stochastic cellular navigation—spatial and temporal sensing—and claims that in the linear-response regime of shallow gradients, navigation performance is fully determined by the feedforward and feedback transfer entropies T_FF and T_FB. These 'Behavioral Equations of State' (BESTs) are stated as Eq. (3) for spatial sensing and Eq. (7) for temporal sensing. The temporal-sensing BEST is tested against stochastic simulations of E. coli chemotaxis using the exact TE-PWS algorithm, and the authors report agreement without fitting or scaling parameters. The paper also argues that single-step transfer entropy is insufficient and discusses an optimal feedback information in the spatial model.","tokens_in":17815,"tokens_out":12329,"duration_ms":118026,"significance":"The idea that bidirectional information transmission provides a system-independent equation of state for navigation is conceptually appealing and, if correct, yields an experimentally testable data collapse across cell types and mutants. The temporal-sensing derivation uses a controlled perturbation theory and is corroborated by exact numerical simulations, and the E. coli test is a strong, falsifiable validation. However, the spatial-sensing BEST as printed in Eq. (3) is mathematically inconsistent, which calls into question the associated claims about optimal feedback information. The overall framework remains significant, but the spatial-sensing leg of the paper requires correction.","major_comments":[{"comment":"Eq. (3) is internally inconsistent and contradicts the model. As T_FB→0 (J→0), the bracket diverges because of the term 1/sqrt((2T_FB+1)^2-1) ~ 1/(2√T_FB), while the performance P must vanish when actuation is turned off. Direct substitution of T_FF and T_FB into Eq. (1) in the small-k limit gives P/P0 = 4 T_FF√T_FB, which is precisely the first bound in Eq. (4) and the expression used in the Discussion for spatial sensing. Thus Eq. (3) is not the shallow-gradient limit of the model. The derived optimum T*_FB=(√2−1)/2 and the non-monotonic curve in Fig. 3A are artifacts of this erroneous expression and must be revised.","section":"§'Performance depends solely on bidirectional information transmission', Eq. (3)"},{"comment":"The temporal-sensing BEST (Eq. (7)) rests on a second-order perturbation expansion of stochastic Riccati equations. The Methods state that the full derivation is provided in the SI, but the SI is not included in the manuscript. The authors also disclose that the expressions were obtained with AI assistance and independently verified. Because Eq. (7) is a central result, the derivation must be available for review; without it, the analytic content of the temporal BEST cannot be fully verified, although the final expression matches the numerical simulations in Fig. 4C.","section":"Materials and Methods, 'Nonlinear perturbation theory...'"}],"minor_comments":[{"comment":"The definition of a single velocity correlation timescale H relies on choosing parameters for which the velocity autocorrelation function is a single exponential. This restricts the test to a subset of navigators; the universal claim for multi-timescale velocity dynamics is not tested. The authors should state this scope condition more prominently.","section":"E. coli section, 'BEST predicts chemotaxis in E. coli'"},{"comment":"The printed definitions of T_FF and T_FB contain garbled mathematical expressions ('[p 1 + (2Gk/H2)(Dv/Df )1/2 −1]/2'); they should be typeset unambiguously.","section":"Equation (3) vicinity"},{"comment":"The sentence 'Eq. 1 shows that in shallow gradients, navigation is controlled by bidirectional information flow alone' is imprecise: Eq. (1) is written in elementary parameters; the information-only statement follows only after substitution. Please clarify the wording.","section":"Performance depends solely on bidirectional information transmission, p.5"}],"recommendation":"major_revision","confidential_remarks":"The central conceptual claim is interesting, and the temporal-sensing/E. coli portion is the strongest part. The error in Eq. (3) is serious but appears correctable—the correct small-gradient expression is already stated in Eq. (4) and the Discussion. The SI with the full perturbation derivation and, ideally, simulation code should be supplied. The explicit AI-assistance disclosure is not itself a problem, but it makes the missing SI less acceptable. The paper fits the scope of the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The headline in your report is right: this is a novel and potentially important framework. The idea that navigation performance in the linear regime can be expressed as a function of feedforward and feedback multi-step transfer entropies alone—without any reference to the elementary parameters—is genuinely new. The temporal-sensing BEST (Eq. 7), P/P0 = 4[ρ/(1+ρ)]√TFF√TFB, is a clean result, and the E. coli simulation shows a collapse with no fitted parameters. That is real evidence and deserves credit.\n\nThe soft spots are real, though, and one of them is load-bearing. Equation 3, the spatial-sensing BEST, cannot be correct as printed. As TFB → 0, the bracket diverges as 1/√TFB, so the right-hand side diverges while the performance must vanish when the actuator gain goes to zero. Direct substitution from Eq. 1 in the same small-TFF limit gives P/P0 = 2 TFF √((2TFB+1)^2−1)/(2TFB+1)^2, which reduces to 4 TFF√TFB for weak feedback and TFF/TFB for strong feedback—the bounds that Eq. 4 claims to derive from Eq. 3. So the printed equality is internally inconsistent; the formula appears to have a typo in the bracket. This error sits in the first central result of the paper, and it cannot be waved away. If the authors produce the correct expression, the framework may stand; but as submitted, the spatial claim is not supported.\n\nOther concerns are less severe but worth stating. The full temporal perturbation derivation is in an SI that is not included, and the Methods acknowledge the expressions were developed with ChatGPT/Claude and independently verified. That is not disqualifying, but I would need to see the calculations to believe the second-order perturbation theory. The E. coli test selects parameter values whose velocity autocorrelation is a single exponential, then defines H from that; that makes the test less universal than the abstract implies. The normalization of transfer entropies also looks inconsistent across sections, which matters because the BEST relations depend on those dimensionless definitions.\n\nThe authors are honest about the regime restrictions: they state shallow gradients, weak actuation, and the need for fast sensing in steeper gradients. So this is not an overreach in the discussion; the problem is the technical error in Eq. 3 and the missing SI.\n\nMy take: the paper deserves a serious referee, not a desk reject. The temporal result and the E. coli test are strong enough to warrant a careful round. But the referee needs to demand the correct form of Eq. 3 (or an explanation for the divergence), the SI, and a fuller treatment of the H-selection. I would not cite this version in my own work until the spatial relation is fixed.","headline":"The spatial BEST is wrong as printed, but the temporal result and E. coli test are strong enough to warrant a careful revision.","tokens_in":18326,"tokens_out":8619,"would_cite":false,"duration_ms":74221,"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":"Navigation performance in shallow gradients is governed by two information flows alone.","keywords":["chemotaxis","transfer entropy","navigation performance","sensory-motor feedback","stochastic control","E. coli","Behavioral Equation of State"],"falsifier":"Measure P, T_FF, T_FB and rho in cells with a clearly multi-exponential velocity autocorrelation under shallow-gradient, weak-actuation conditions. If their normalized performance deviates from P/P0 = 4[rho/(1+rho)] sqrt(T_FF T_FB) while T_FF and T_FB are correctly computed from full trajectories, the universality claim is refuted. Equivalently, find two navigators with the same T_FF, T_FB and rho but different elementary parameters whose performance differs.","tokens_in":17440,"feed_emoji":"🧭","tokens_out":8411,"duration_ms":87474,"temperature":0.7,"pith_summary":"The paper tries to establish that a cell's navigation performance in shallow gradients is fully determined by how much information flows from the sensory input to the sensory output and back from the output through actuation to the future input, not by the specific gains, noises, or rates of the underlying biochemical circuitry. For both spatial and temporal sensing, the authors derive exact equalities, called Behavioral Equations of State (BESTs), connecting normalized performance to feedforward and feedback transfer entropies plus the relative speed of sensing and actuation. If true, this means information transmission is the organizing principle for stochastic navigation, and a simple data collapse should be observable across mutants and cell types. The authors support the claim with analytical solution of two minimal models and parameter-free simulation of E. coli chemotaxis.","feed_headline":"Bidirectional information flow alone predicts how cells navigate","feed_subtitle":"Two transfer entropies set drift and localization in shallow gradients; E. coli tests match with no fitting.","key_machinery":"The central objects are the feedforward and feedback transfer-entropy rates, defined over full signal trajectories rather than single time points (multi-step transfer entropy). The spatial-sensing derivation uses stochastic control theory with spectral factorization of power spectra to obtain causal prediction variances; the temporal-sensing derivation combines path-measure transformation formulas with a perturbation expansion of stochastic filtering equations, valid to second order in the actuator gain. For the nonlinear E. coli test, transfer entropies are computed exactly by the TE-PWS (path-weight sampling) algorithm.","core_discovery":"Stochastic navigation in shallow gradients is claimed to obey a Behavioral Equation of State (BEST): performance is fully determined by two history-dependent information flows—feedforward transfer entropy T_FF (sensory input to output) and feedback transfer entropy T_FB (output through actuation to future input)—independent of gain, noise, or actuation strength. For temporal sensing, P/P0 = 4[rho/(1+rho)] sqrt(T_FF) sqrt(T_FB), rho = sensory speed / velocity relaxation rate; for spatial sensing, an analogous equality (Eq. 3) holds. These equalities are derived analytically in two minimal models and verified in E. coli chemotaxis simulations without fitting.","pith_inferences":["If the BEST holds broadly, navigation performance could be monitored non-invasively by measuring two information flows from trajectory data, turning information into a practical diagnostic for chemotaxis and other sensorimotor loops.","The different scaling with feedforward information (linear for spatial, square-root for temporal) suggests that trajectory statistics alone could reveal which sensing strategy a cell uses, without requiring morphological measurements.","The optimal feedback information found in E. coli simulations implies a possible evolutionary pressure on actuator gain; one could test whether wild-type cells sit near this optimum by measuring feedback information across many strains and conditions.","The single-exponential velocity assumption is a natural place to generalize: cells with several motility timescales may require a multi-timescale BEST with rho replaced by a spectrum of ratios."],"forward_implications":["The BEST predicts a data collapse: cells or mutants with different sensory gain, noise, or actuation strength should fall on the same performance-versus-information curve in shallow gradients.","In this regime, navigation can be improved only by increasing feedforward or feedback information or, for temporal sensing, the relative sensing speed rho; elementary parameters matter only through these flows.","Too much feedback information degrades performance, implying an optimal actuator gain—found analytically for spatial sensing and by simulation (around 4 nats) for E. coli—so actuator design must balance responding to signal versus sensory noise.","Single-step transfer entropy is insufficient; any empirical test of the BEST must use history-dependent, trajectory-based transfer entropy.","The same principle may extend to other feedback-driven biological functions such as homeostasis and immune response, as the paper's closing discussion suggests."],"supporting_citations":[{"why":"Defines transfer entropy and its history-dependent, trajectory-based form that the paper uses throughout.","marker":"[21]"},{"why":"Supplies the exact TE-PWS algorithm used to compute transfer entropies in the nonlinear E. coli simulations.","marker":"[24]"},{"why":"Provides the stochastic control/filtering method used to analytically calculate transfer entropy rates in the linear spatial-sensing model.","marker":"[37]"},{"why":"Provides the coarse-grained E. coli chemotaxis model and parameter values used for the parameter-free test.","marker":"[38]"},{"why":"Establishes that E. coli chemotaxis is information-limited, motivating the link between information and performance.","marker":"[15]"},{"why":"Gives a trajectory-based estimator that would allow the BEST's data-collapse prediction to be checked experimentally.","marker":"[33]"},{"why":"Independent framework for bidirectional information in navigation that the paper contrasts with its history-dependent transfer-entropy definition.","marker":"[43]"}],"fun_headline_variants":["Two info flows set navigation accuracy in cells","Behavioral equation of state maps info to cell drift","Bidirectional info flow controls cellular navigation","E. coli navigation predicted by transfer entropy","Cell navigation collapses onto information-based law"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The E. coli test assumes the cell's velocity fluctuations are fully captured by a single exponential decay timescale H; cells with multi-timescale velocity dynamics may not obey the BEST.","fun_headline_variants_meta":{"raw":{"variants":["Two info flows set navigation accuracy in cells","Behavioral equation of state maps info to cell drift","Bidirectional info flow controls cellular navigation","E. coli navigation predicted by transfer entropy","Cell navigation collapses onto information-based law"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000196,"raw_usage":{"total_tokens":1203,"prompt_tokens":758,"completion_tokens":445,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":502,"completion_tokens_details":{"reasoning_tokens":379}},"tokens_in":502,"tokens_out":445,"duration_ms":5237,"temperature":1.0,"reasoning_tokens":379,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T04:21:34.557677+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure P, T_FF, T_FB and rho in cells with a clearly multi-exponential velocity autocorrelation under shallow-gradient, weak-actuation conditions. If their normalized performance deviates from P/P0 = 4[rho/(1+rho)] sqrt(T_FF T_FB) while T_FF and T_FB are correctly computed from full trajectories, the universality claim is refuted. Equivalently, find two navigators with the same T_FF, T_FB and rho but different elementary parameters whose performance differs.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines transfer entropy and its history-dependent, trajectory-based form that the paper uses throughout."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the stochastic control/filtering method used to analytically calculate transfer entropy rates in the linear spatial-sensing model."},{"cited_title":"Sourjik, A","cited_arxiv_id":null,"evidence_quote":"Gives a trajectory-based estimator that would allow the BEST's data-collapse prediction to be checked experimentally."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Independent framework for bidirectional information in navigation that the paper contrasts with its history-dependent transfer-entropy definition."}],"review_version":2}