{"id":"2dc80bed-8753-4da3-b1a2-a5c4466c2464","arxiv_id":"2506.07094","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A CIR bridge with closed-form mean and variance is proposed and fitted to sub-hourly ayu migration counts, capturing the observed bursty intermittency.","lead":"The paper introduces a new stochastic process, a Cox-Ingersoll-Ross bridge pinned to zero at the start and end of each day, and fits it to 10-minute counts of migrating juvenile ayu in the Nagara River. If the model holds, it offers a tractable way to describe bursty, intermittent migration patterns and to plan fish counting schemes.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Empirical validation is missing where it matters: model burst statistics are never compared to observed 10-min counts, and pointwise theoretical moments are fitted against aggregated 10-min data.","rationale":"The reader's weakest assumption identified the discrete-count/data-mapping issue as the soft spot; I agree that this is where the central empirical claim is least secure, and I sharpen it in two ways. The pointwise-versus-binned variance mismatch is a concrete technical defect in the fitting procedure for sigma, and the absence of any empirical burst comparison means the paper's headline phenomenon--intermittent bursts--is never actually validated against data. Both are testable and do not affect the theoretical contribution, which appears plausible. The reader's CONDITIONAL verdict therefore remains appropriate: the paper should be accepted conditionally on carrying out these empirical checks, not rejected outright, since the theoretical construction and the closed-form moment results are independent of the data-fitting issue. I mark agreement as 'partial' because my primary concern is more specific than the reader's general statement: the problem is not merely that counts are discrete, but that the fitting target is the wrong moment (pointwise variance) and that burst statistics are never benchmarked against the observed time series.","tokens_in":30514,"tokens_out":15954,"duration_ms":171524,"concrete_test":"Extract empirical burst statistics from the raw 10-min counts using the exact definitions of Section 4.4.3 (normalize each day's counts by S_k, apply threshold X=0.01 and minimum duration T=0.02), then compare the empirical distributions of burst counts and burst durations to the model-generated distributions in Tables 8-9 via a two-sample Kolmogorov-Smirnov test with bootstrap resampling, and repeat with the doubled thresholds. If the empirical histograms lie outside the model's sampling variability at the 5% level, the claim that the CIR bridge reproduces the observed intermittency fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central empirical claim that model 2 'reasonably models' the intermittent sub-hourly fish counts rests on two load-bearing steps that are not supported by the paper. First, parameter fitting compares the pointwise theoretical variance (9) with the empirical variance of 10-min aggregated counts. The observed counts are integrals of the rate process over 10-minute bins, not point observations of Y_s at the bin center; the variance of a bin integral is not Var[Y_s] and is generally smaller, especially for an intermittent high-volatility process. This can bias the fitted sigma upward and undermine the high-volatility diagnosis made through (25)-(26). Second, the burst analysis in Section 4.4.3 reports only model-simulated burst counts and durations (Tables 8-9, Figures 12-13); no empirical burst statistics are extracted from the same 10-min data using the same threshold and duration definitions and compared to the model output. Therefore the statement that the model 'can handle both sparse- and dense-burst cases' is unverified against the phenomenon it claims to describe. These issues are addressable, but until they are checked the empirical half of the central claim is not established; the theoretical construction may well be sound.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper introduces a 'CIR bridge' as a pathwise continuous nonnegative solution to the time-inhomogeneous SDE (2), with zero initial and terminal values, and proves that under condition (4) the solution is unique and the mean and variance are given in closed form (Propositions 1-2). It also derives a conditional moment-generating function (Proposition 3) and applies the model to 10-minute counts of juvenile ayu migration in the Nagara River, fitting the parameters a, σ, and the shape of h by least squares to the theoretical mean and variance, and reporting simulated burst statistics. The central claims are that the bridge is well-posed for all volatility regimes, that the fitted model corresponds to a high-volatility (intermittent) regime, and that the model reasonably describes the sub-hourly fish counts.","tokens_in":30786,"tokens_out":13302,"duration_ms":138011,"significance":"The theoretical part of the paper is a useful contribution to the diffusion-bridge literature: it provides an affine CIR-type bridge with closed-form moments, valid in principle for high-volatility regimes, and it employs the iVi scheme, which is simple and positivity-preserving. The explicit formulas (8)-(9) and the simulation algorithm are strengths, and the application is original in its sub-hourly target scale. However, the empirical validation has two load-bearing gaps: the moment fitting ignores the bin-integral structure of the 10-minute count data, and the burst analysis is not compared to the observed data. If these gaps are closed, the paper could be a solid contribution; as it stands, the empirical half of the central claim is not established.","major_comments":[{"comment":"The observed data are 10-minute counts, which are integrals of the rate process over the bin interval, not point observations of Y_s. The theoretical mean and variance in (8)-(9) apply to the instantaneous rate Y_s. For a process with volatility as high as the fitted values, Var(∫_s^{s+Δ} Y_u du) is generally smaller than (Δ)^2 Var(Y_s), so least-squares fitting of σ to the empirical variance of the bin counts can bias σ upward. Since the high-volatility diagnosis (25)-(26) depends on the fitted σ, this is a load-bearing issue. I suggest deriving the moments of the bin integrals from the SDE (or from the explicit moment formulas) and fitting those, or quantifying the bias numerically for the fitted process.","section":"Section 4.3.2, Eqs. (8)-(9), Figs. 7-8"},{"comment":"The burst statistics (number and duration of bursts per day) are reported only for simulated paths of the fitted model. The same burst definition (thresholds X* and T* applied to the normalized process) should be applied to the empirical 10-minute counts, which are available for the same days used in fitting. Without an empirical burst histogram or at least a summary of observed bursts, the statement that the model 'can handle both sparse- and dense-burst cases' is not a comparison to the phenomenon. The paper should include this empirical benchmark.","section":"Section 4.4.3, Tables 8-9, Figs. 12-13"},{"comment":"Proposition 1 assumes condition (4), but the second example in (5), which is exactly model 2 used in the application (24), has h(t)=ε/t + 1/(1-t). This function is unbounded near t=0, whereas the upper bound in (4) is finite at t=0 if it is the displayed h_0/(T-t)+ω. Thus model 2 does not satisfy the stated assumptions of Proposition 1, and the proof in Appendix A.1 only controls h near the terminal time T. The authors should either extend the well-posedness theorem to allow initial singularities of the form ε/t, or state and prove a separate well-posedness result for model 2. This is needed because the application relies on model 2.","section":"Section 3.1, Eq. (4)-(5); Section 4.3.1, Eq. (24)"},{"comment":"The derivation of the mean and variance ODEs (27) and (29) assumes that expectations can be differentiated under the integral and that the first two moments are finite. In the high-volatility case with unbounded h near T, these facts are not immediate and should be justified, e.g., by a truncation argument or by using the explicit solution of the regularized SDE and then passing to the limit. The martingale convergence step is also sketched; providing the localization details would make the proof complete.","section":"Appendix A.1, Eqs. (27)-(29)"}],"minor_comments":[{"comment":"The normalization uses the observed daily total S_k as a scaling factor. It should be stated explicitly that S_k is treated as a known, non-random constant for each day and that the dimensionless model (23) is an additional modeling assumption; the robustness of the fitted parameters to this normalization would be worth a brief discussion.","section":"Section 4.3.1, Eq. (22)"},{"comment":"The burst thresholds X* and T* are defined in dimensionless units; it would be helpful to report the corresponding physical values (fish per unit time and minutes) so the burst definition is interpretable for fish counters.","section":"Section 4.4.3"},{"comment":"The convergence study reports maximum errors but does not state an empirical order of convergence; a short comment on how the error scales with sample size and time step would improve the presentation.","section":"Section 4.4.2, Tables 5-7"},{"comment":"Several equations in the full text are garbled (e.g., (13), (31)-(37)); the final manuscript should be carefully proofread for formula rendering.","section":"Throughout"},{"comment":"The statement 'Mathematical models that deal with the fine (sub-daily) dynamics of migrating fish populations have not been studied' is strong; the preceding literature review is adequate but could soften the claim or cite any related sub-daily models.","section":"Section 1.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is a single-author work and the author has several closely related papers (e.g., [30,31,54]). It would be useful for the editor to consider whether the novelty relative to those works is sufficiently highlighted. Also, the data are from one site and two years; the generalizability of the fitted parameters is unclear, but this is a standard limitation. The empirical gaps (bin aggregation, missing empirical burst comparison) are the main barriers to acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The theoretical core of this paper is real and worth taking seriously. The SDE (2) with unbounded drift and diffusion near the terminal time is a legitimately different construction from the h-transform CIR bridges in the cited literature, and the closed-form mean and variance in Proposition 2 are genuinely useful. The paper also gives an explicit moment-generating function, a sensible numerical scheme, and careful convergence tables for the iVi discretization. As a piece of applied probability, this is a solid addition to the toolkit for nonnegative, pinned, intermittent processes.\n\nThe application is the weaker half, and the stress-test note points at the right problems. The observed 10-minute counts are integrals of the rate process over bins, not point observations of Y_s at the bin center. Comparing the theoretical variance at a point to the empirical variance of bin counts is not apples-to-apples, and for an intermittent high-volatility process the discrepancy can be large enough to bias the fitted sigma. That concern is real, and it matters because the high-volatility diagnosis is one of the paper's headline empirical claims. The burst analysis is also only model-simulated: the paper reports simulated burst counts and durations but never extracts the same statistics from the observed 10-minute data using the same threshold definitions. So the claim that the model 'can handle both sparse- and dense-burst cases' is unverified against the phenomenon it is meant to describe. The fitting-then-validating-on-the-same-moments circularity is standard but still weakens the empirical conclusions; there are no error bars and no held-out days.\n\nThese are addressable flaws, not fatal ones. The math does not depend on the fish data, and the paper is honest about many of its limitations. The proof of Proposition 1 is sketched in places—existence on [0,T-delta) via a cited regularization and the terminal behavior via moment bounds plus a martingale argument—but nothing here looks broken. I also checked the citation pattern; the relevant CIR bridge literature is cited, and the self-citations are legitimate given the author's prior work on this fish system.\n\nWho is this for? Mathematicians working on affine bridges will get a clean new example with closed-form moments. Ecologists or quantitative fisheries people will get a useful model class, but they should treat the empirical validation as provisional until the aggregation issue is fixed and the burst comparison is actually done.\n\nMy recommendation: send it to peer review. The theoretical contribution deserves referee time, but the referee should insist on addressing the bin-aggregation problem, adding empirical burst statistics, and preferably releasing data and code. This is a conditional accept trajectory, not a reject.","headline":"A genuinely new tractable CIR bridge with closed-form moments, attached to an empirical application that is suggestive but not yet properly validated.","tokens_in":31292,"tokens_out":1857,"would_cite":true,"duration_ms":26364,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H10","60J60","92D25"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that a Cox–Ingersoll–Ross bridge pinned to zero at sunrise and sunset is well-posed with closed-form mean and variance, and that the fitted high-volatility bridge reproduces the intermittent 10-minute migration counts of…","keywords":["CIR bridge","Cox-Ingersoll-Ross process","diffusion bridge","sub-hourly fish migration","ayu Plecoglossus altivelis altivelis","on-off intermittency","burst statistics","closed-form moments"],"falsifier":"Re-estimate the fitted models on days with very large total counts and compare the observed variance of the normalized 10-minute counts to the theoretical variance (9). If the observed variance consistently exceeds the theoretical one by more than Monte Carlo error, the square-root diffusion coefficient understates the burstiness and the central fit claim would need revision.","tokens_in":30323,"feed_emoji":"🐟","tokens_out":8206,"duration_ms":81243,"temperature":0.7,"pith_summary":"This paper introduces a stochastic bridge built from the Cox–Ingersoll–Ross process and argues that it is a well-posed model for sub-hourly fish migration counts. In the model, the number of fish passing a river observation point between sunrise and sunset is a nonnegative diffusion that starts and ends at zero, with drift and volatility functions that become unbounded near sunset. The paper proves existence, uniqueness, and continuity of the solution for a broad class of such functions, and derives closed-form time-dependent averages and variances that make parameter fitting direct. Fitted to 10-minute counts of juvenile ayu in the Nagara River, the bridge falls in the high-volatility regime, so its sample paths show intermittent bursts similar to the observed on–off pattern of counts.","feed_headline":"CIR bridge reproduces 10-minute fish-count bursts","feed_subtitle":"Sunrise-to-sunset diffusion with closed-form statistics matches intermittent ayu counts in the Nagara River.","key_machinery":"The central object is the CIR bridge, the solution of SDE (2): a time-inhomogeneous, localized Cox–Ingersoll–Ross process with a square-root diffusion coefficient $\\sigma\\sqrt{h(t)X_t}$ and an unbounded drift and diffusion as $t$ approaches the terminal time. Its tractability comes from being an affine process under a time change: with $H'=h$, the bridge is a time-changed CIR process with a time-dependent source, which yields the closed-form average (8), variance (9), and conditional moment-generating function (11)–(13). The function $h$ controls how the biological clock accelerates near sunset, and condition (4) ensures the terminal value is zero; a recently developed one-step numerical method preserves nonnegativity even when the net volatility is unbounded.","core_discovery":"The central claim is that the SDE $dX_t=(a(t)-h(t)X_t)\\,dt+\\sigma\\sqrt{h(t)X_t}\\,dB_t$ with $X_0=0$ and a suitable unbounded $h$ near $T$ is a well-posed CIR bridge: it has a unique pathwise continuous, almost surely nonnegative solution with terminal limit $0$ regardless of the volatility size, and its mean (8) and variance (9) are available in closed form. Applied to each day's 10-minute fish counts after normalization by that day's total, the fitted bridge with $h(s)=1/s+1/(1-s)+\\varepsilon$ tracks the empirical average, standard deviation, and coefficient of variation, and the fitted parameters satisfy the high-volatility condition. The paper therefore claims that sub-hourly upstream migration of $P$. altivelis at the study site is an intermittent, high-volatility phenomenon, with sample paths exhibiting several burst events per day whose statistics match the sparse and dense burst patterns seen in the data.","pith_inferences":["The paper leaves implicit that treating the observed counts as noisy observations of the latent bridge, for example through a Poisson or overdispersed count link, would be a natural next test; if counts are more overdispersed than the diffusion variance allows, the fitted burst statistics would change.","The normalization by each day's total count removes the dominant day-to-day scale but assumes the daily total is known exactly; a hierarchical extension with random daily totals would propagate counting uncertainty into the bridge parameters.","Because the terminal time is fixed and the endpoint is exactly zero, days with migration after sunset, before sunrise, or with missing observation windows would need a relaxed or randomized endpoint; the paper itself names the fixed terminal time as a limitation.","The burst analysis counts events relative to a fixed threshold height and duration; defining bursts relative to the local mean or to the day's scale would be a straightforward robustness check that could sharpen or overturn the intermittency picture."],"forward_implications":["The closed-form average and variance let the model parameters be identified by least squares, without simulation-based likelihoods or extra approximation errors.","Because the bridge is well-posed in all volatility regimes, it can be used as a building block for any nonnegative quantity that vanishes outside a fixed time interval, such as intermittent river discharge.","The fitted high-volatility bridge predicts that 0 to 2 burst events are typical per day, with 6 to 9 possible in extreme cases and average burst duration near one hour, so manual counting protocols that ignore this intermittency could be badly biased.","The seasonal dependence of migration can be tracked through the fitted parameters of daily bridges, and the analysis points to water temperature as the environmental indicator worth measuring along the river.","The numerical scheme reproduces the theoretical mean and variance accurately when sample size and time step are balanced, so the model is directly computable rather than only an abstract object."],"supporting_citations":[{"why":"Supplies the CIR process existence, uniqueness, and nonnegativity results that the well-posedness proof builds on.","marker":"[71]"},{"why":"Defines the classical Cox–Ingersoll–Ross process that the bridge generalizes.","marker":"[55]"},{"why":"Shows that classical h-transform CIR bridges are limited to low-volatility cases, motivating the new time-change formulation.","marker":"[63]"},{"why":"Provides the one-step numerical scheme whose nonnegativity preservation and convergence are extended to the CIR bridge.","marker":"[64]"},{"why":"Supplies the time-inhomogeneous affine process generator used to derive the conditional moment-generating function.","marker":"[101]"},{"why":"Contributes the 10-minute ayu migration observations that define the empirical sub-hourly target.","marker":"[20]"},{"why":"Provides the burst-threshold statistics used to quantify intermittency in sample paths.","marker":"[94,95]"}],"fun_headline_variants":["CIR bridge captures 10-minute fish bursts","CIR bridge model for sub-hourly fish migration","CIR bridge matches intermittent fish counts","CIR bridge reproduces fish migration bursts"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The 10-minute fish counts are treated as exact, error-free readings of a continuous diffusion state variable, and the process is forced to hit exactly zero at sunrise and sunset; if the discrete count scale or the day-to-day normalization distorts this mapping, the fitted parameters and burst statistics would not carry over to the real counts.","fun_headline_variants_meta":{"raw":{"variants":["CIR bridge captures 10-minute fish bursts","CIR bridge model for sub-hourly fish migration","CIR bridge matches intermittent fish counts","CIR bridge reproduces fish migration bursts"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000277,"raw_usage":{"total_tokens":1648,"prompt_tokens":941,"completion_tokens":707,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":557,"completion_tokens_details":{"reasoning_tokens":650}},"tokens_in":557,"tokens_out":707,"duration_ms":7604,"temperature":1.0,"reasoning_tokens":650,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T05:42:03.612173+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-estimate the fitted models on days with very large total counts and compare the observed variance of the normalized 10-minute counts to the theoretical variance (9). If the observed variance consistently exceeds the theoretical one by more than Monte Carlo error, the square-root diffusion coefficient understates the burstiness and the central fit claim would need revision.","supporting_citations":[{"cited_title":"L., Nguyen, D","cited_arxiv_id":null,"evidence_quote":"Shows that classical h-transform CIR bridges are limited to low-volatility cases, motivating the new time-change formulation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the time-inhomogeneous affine process generator used to derive the conditional moment-generating function."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Contributes the 10-minute ayu migration observations that define the empirical sub-hourly target."}],"review_version":1}