{"id":"5126a882-823b-41ee-8fa4-8c00c6a7cdd3","arxiv_id":"2501.13688","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"In Didelphis aurita, fine-scale movements follow a truncated Lévy flight, but weekly displacements are Gaussian-like because many truncated steps rapidly converge to normal diffusion.","lead":"Researchers tracked a small marsupial at daily and weekly scales and found that its fine-scale movements look like Lévy walks while its weekly movements look like ordinary diffusion. This suggests that movement phases in animals can emerge from a simple statistical effect of adding many steps, without needing extra ecological explanations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'no ecological constraints' claim conflicts with the Section 3.2 filter that excludes animals leaving fragments; the Rayleigh weekly result may be an artifact of that filter, so the unfiltered analysis must be reported.","rationale":"I read the paper's central claim as the abstract's assertion that the two movement phases are produced by aggregation of daily truncated Levy steps without ecological constraints. The most load-bearing condition for that claim is that the weekly Rayleigh distribution is observed under the same process that generates the daily step distribution. The Section 3.2 filter breaks this condition: it selects settled, fragment-faithful individuals, which is an ecological constraint, and it removes exactly the long displacements that would create heavy tails. The unbounded simulation in Section 4.3 has no such filter, so the close match in Figure 5 could be coincidental or driven by the filter rather than by pure scale aggregation. The reader's concern that n is fitted to sigma is real but secondary: fitting n makes the one-week crossover a calibration rather than an independent prediction, but even an independently measured n would not rescue the no-ecological-constraints claim if the unfiltered weekly distribution is heavy-tailed. I agree with the CONDITIONAL verdict: the small-scale analysis and the convergence mechanism are plausible, but the headline inference requires the unfiltered analysis or a careful rephrasing that explicitly limits the claim to settled animals. The proposed test would settle whether the filter is the cause of the observed Brownian behavior.","tokens_in":11833,"tokens_out":6070,"duration_ms":56686,"concrete_test":"Rerun the Section 4.2 analysis on the complete radio-tracking dataset, including all weekly displacements from individuals that left their initial fragment or dispersed, without the Section 3.2 filter. Compute the Hill tail index and the Kolmogorov-Smirnov p-value against the Rayleigh distribution. If alpha < 3 or the Rayleigh fit is rejected, the Brownian crossover is an artifact of excluding dispersers; if the unfiltered distribution is still Rayleigh with similar sigma, the no-ecological-constraints claim survives. Also report how many of the 43 weekly displacements come from each individual and how many dispersal events were excluded, to assess whether sample size or pseudoreplication drives the result.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.2 states that the radio-tracking data were filtered to keep only movements of individuals that did not leave the forest fragment where they were first detected, disregarding animals that left the fragment for dispersal. Section 5 then concludes that the Levy-to-Brownian crossover arises without the necessity of introducing ecological constraints. This is internally inconsistent: the filter is itself an ecological/behavioral constraint, and it removes the longest displacements that would contribute heavy tails. The empirical Rayleigh fit (sigma = 54 +/- 4 m, KS p = 0.166) and Hill tail index alpha > 3 are computed on this filtered sample, so they may reflect the selection rule rather than pure aggregation of small-scale steps. The simulation in Section 4.3 is an unbounded random walk with no fragment boundary or dispersal filter, so its agreement with the filtered empirical distribution is not a test of the no-ecological-constraints claim; it is a comparison between an unbounded process and a conditioned sample. The central claim therefore rests on the untested assumption that the unfiltered weekly displacement distribution would also be Rayleigh.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies movement of the marsupial Didelphis aurita at two temporal scales: daily step-length data from spool-and-line tracking and weekly displacement data from radio-tracking. The daily step-length distribution is best described by a truncated Pareto (truncated Levy) distribution with exponent μ = 1.36 and cutoff b = 103.1 m, selected by AIC over Pareto and exponential alternatives. The weekly displacement distribution is reported as consistent with a Rayleigh distribution (σ = 54 ± 4 m, KS p = 0.166), and a Hill tail index above 3 is taken as evidence against a heavy-tailed weekly distribution. The authors then simulate truncated-Pareto random walks with parameters from the small-scale fit, show that the displacement distribution converges to Rayleigh for a number of steps around 68, and that this number reproduces the empirical weekly σ exactly. They conclude that the crossover from Levy to Brownian behavior occurs after about one week as a purely statistical effect of aggregation, without ecological constraints.","tokens_in":12040,"tokens_out":2391,"duration_ms":23955,"significance":"If the central claim holds, the paper provides a clean and parsimonious mechanism for scale-dependent movement regimes: the same individual-level step distribution generates Levy-like small-scale movement but Brownian-like large-scale displacement, with a crossover time set by the truncation scale. The small-scale analysis is a strength: the AIC comparison strongly supports the truncated Pareto over alternatives on 2239 steps, and the projection-based segmentation is a principled way to define steps. The simulation framework is transparent and the match with the weekly σ is visually compelling. However, the broad claim of 'no ecological constraints' and the specific claim that one week corresponds to 68 steps are weakened by the filtering of the radio-tracking data and by the fact that n is fitted, not independently measured. The paper's value depends on resolving these two issues, because they directly affect the interpretation of the weekly Rayleigh result and the predictive nature of the crossover.","major_comments":[{"comment":"The filtering described in Section 3.2 is an ecological constraint in itself: the authors keep only movements of individuals that did not leave their original forest fragment, explicitly disregarding animals that left for dispersal. This removes the longest displacements, which are exactly the events that could generate heavy tails in the weekly distribution. The Rayleigh fit (σ = 54 ± 4 m, KS p = 0.166) and the Hill-index claim α > 3 are computed on this filtered sample. The simulation in Section 4.3, by contrast, is an unbounded random walk with no fragment boundary or dispersal filter. Comparing an unbounded process to a conditioned sample does not test the claim that the Rayleigh behavior arises 'without the necessity of introducing ecological constraints' (abstract and Section 5). The authors must report the unfiltered weekly displacement distribution and repeat the Rayleigh fit and Hill-index estimation on it, or otherwise quantitatively show that the filter does not affect the conclusions.","section":"Section 3.2 and Section 5"},{"comment":"The 'prediction' of a one-week crossover is not independent: the number of steps n is selected so that the simulated σ equals the empirical σ exactly ('for a number of steps fixed to 68, σ_MLE = 54, exactly the same of the empirical case'). Because n is a free parameter fitted to the same dataset, the agreement does not validate the model's temporal extrapolation. To support the claim that one week corresponds to 68 steps, an independent estimate of the number of steps per week is needed, for example from turning-point rates or activity budgets collected in the same study system, or the claim should be explicitly demoted from a prediction to a consistency check.","section":"Section 4.3"},{"comment":"The convergence to Rayleigh for n ≈ 68 is an expected consequence of the finite variance of the truncated Pareto distribution, not a distinctive signature of the specific parameters. The paper shows that a Rayleigh distribution becomes a good approximation for n greater than about 40, which is a general property of sums of independent finite-variance variables. The novel empirical content is therefore the estimated crossover time, and that content depends entirely on the fitted n. The manuscript should clarify that the simulation does not independently constrain the crossover time; it only demonstrates that if animals make about 68 steps per week, then the weekly displacement would be Rayleigh.","section":"Section 4.3 and Figure 4"}],"minor_comments":[{"comment":"The tail index α and the likelihood ratio R are plotted without confidence intervals or bootstrap uncertainties; adding these would strengthen the visual claim of a crossover near n ≈ 68.","section":"Figure 4"},{"comment":"The statistic R, defined as the ratio of log-likelihoods of simulated data and i.i.d. Rayleigh draws, is nonstandard; the paper should explain how its values should be interpreted and whether any threshold is meaningful.","section":"Section 4.3"},{"comment":"The Data Availability section states that data will be archived 'if the paper is accepted'; for reproducibility, the data and code should be made available during review or in a permanent repository at submission.","section":"Section 2 and Data Availability"},{"comment":"Some reference entries are incomplete or combine multiple citations (e.g., [36]); please ensure each reference is complete and formatted consistently.","section":"References"},{"comment":"There are occasional typographical inconsistencies, such as inconsistent spacing around accents and equations; a careful proofread would improve readability.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's central claim is internally inconsistent with its own filtering procedure: the weekly Rayleigh distribution is obtained after excluding dispersal events, which is an ecological/behavioral constraint, yet the conclusion states that no ecological constraints are necessary. This is not a matter of presentation but of the scientific claim. The second issue is that the crossover time is fitted, not predicted, because n is chosen to match the empirical σ. Both issues are fixable within the manuscript's scope if the authors report the unfiltered analysis and either obtain an independent estimate of n or reframe the claim as a consistency check. The small-scale analysis is solid and the paper has value, but in its current form the abstract and discussion overstate the predictive power of the results."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Edgardo,\n\nQuick take: this is a real empirical contribution with a core idea that is easy to state—truncated Lévy steps, once aggregated, can produce Rayleigh displacements after about a week—and the small-scale statistics are the best part. But the strong version of the claim, that this happens without ecological constraints, does not survive contact with their own filtering. The paper deserves a serious referee, but a revision is needed.\n\nWhat is new: they combine daily spool-and-line tracks (2239 steps, 141 trajectories) with weekly radio-tracking displacements for the same species, Didelphis aurita. The AIC comparison strongly favors truncated Pareto over Pareto and exponential (ΔAIC 448 and 503), and the fitted parameters (μ=1.36, b=103 m) are in line with their prior work. The simulation analysis of how the displacement distribution's tail index crosses from heavy-tailed to Rayleigh as n grows is well done and clearly presented. The qualitative crossover is plausible and worth testing elsewhere.\n\nWhere it gets soft. First, Section 3.2 states the weekly data were filtered to keep only animals that did not leave their forest fragment, discarding dispersal events. That filter removes exactly the long moves that generate heavy tails. Then the simulation is an unbounded random walk, and the comparison in Figure 5 is between that unbounded process and the conditioned sample. So the empirical Rayleigh result may be an artifact of the selection rule. The authors should report the unfiltered weekly distribution and show it is also Rayleigh; if it is, the claim holds.\n\nSecond, the bridge between scales is not a prediction. In Section 4.3 they write that they \"estimated the best n value which reproduces the P(r)\" of the field data, and n=68 gives σ=54 exactly. Since n is fitted to the target, the \"one week\" crossover is partly a fitting outcome. An independent estimate of nightly step counts, or at least a range of n with uncertainty on the simulated distributions, would make the claim much stronger. As it stands, the crossover time is a consistency check, not an independent prediction.\n\nThird, the \"without the necessity of introducing ecological constraints\" sentence in the abstract and discussion overstates the case, given the filter. That sentence should be rewritten to say \"within the settled-animal sample\" or the filter should be dropped.\n\nMinor: the weekly sample is 43 and the KS p=0.166 is a failure to reject, not proof of Rayleigh, but they do acknowledge this is a single-parameter fit.\n\nWho is this for? Movement ecologists and anyone who works on Lévy vs Brownian debates. I would send it out, but with a request for the unfiltered analysis, a sensitivity analysis on n, and a softened interpretation. As it stands, I wouldn't cite the crossover time as a prediction, but the small-scale fit is citable.","headline":"A genuinely interesting empirical crossover result with solid small-scale statistics, but the 'no ecological constraints' claim is undercut by the dispersal filter and a fitted step count, so it needs revision rather than rejection.","tokens_in":12620,"tokens_out":2358,"would_cite":false,"duration_ms":22376,"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":"For the marsupial Didelphis aurita, daily movement follows a truncated Lévy flight, but weekly displacement is already Brownian—and the crossover arises purely from aggregation, without ecological constraints.","keywords":["movement patterns","spatio-temporal scales","Lévy flight","Brownian motion","marsupials","Didelphis aurita","truncated Pareto distribution","scale of aggregation"],"falsifier":"As an independent test, an animal would need to be tracked continuously for a week, counting every turning-point step and its length, so that the weekly displacement can be compared with a Rayleigh distribution generated by summing exactly that many observed steps. If the independently counted n yields a substantially different σ or a non-Rayleigh shape, the purely statistical crossover would not hold.","tokens_in":11581,"feed_emoji":"🐾","tokens_out":8730,"duration_ms":68326,"temperature":0.7,"pith_summary":"In this paper the authors ask whether the same animal's movement can look like a Lévy flight or like Brownian motion depending on how finely you slice its path. Daily step lengths of the marsupial Didelphis aurita follow a truncated Lévy distribution, while distances traveled in a week already match a Rayleigh distribution, the signature of ordinary diffusion. The authors show that this transition is a purely statistical effect of summing many small steps: a simulation using the daily step distribution reproduces the weekly spread without adding any ecological constraints such as home ranges or habitat boundaries. If correct, this means Lévy flights are of little use for describing dispersal in this species over practical periods, and normal diffusion is the appropriate approximation.","feed_headline":"Lévy flight becomes Brownian after one week in marsupial moves","feed_subtitle":"Daily steps of Didelphis aurita look Lévy, but a week of sums already behaves as ordinary diffusion.","key_machinery":"The central object is the truncated Pareto (truncated Lévy) step-length distribution, f(ℓ) ∝ $ℓ^{{-μ}}$ on [a,b], with a = 0.8 m and b = 103.1 m; this finite-variance distribution is what guarantees a crossover to Gaussian/Rayleigh behavior once many steps are summed. The argument is carried by the summation of independent steps in a discrete random walk: the number n controls the shape of the resulting displacement distribution, and the authors track the crossover using two diagnostics—the tail index α (below 3 means heavy-tailed, above 3 means light-tailed) and the ratio R of log-likelihoods of the simulated displacements against a fitted Rayleigh distribution (near 1 means effectively Brownian). The key identity is the central-limit-theorem behavior r ∼ √n for large n, recovered in the simulations for n ≳ 100.","core_discovery":"The central claim is that the truncated Lévy distribution fitted to the daily step lengths of Didelphis aurita—with exponent μ = 1.36 and cutoff b = 103.1 m—sums to a Brownian, Rayleigh-distributed displacement after roughly one week, and that this crossover is produced by aggregation alone. The weekly radio-tracking data give σ = 54 ± 4 m and are statistically indistinguishable from a Rayleigh distribution, with a tail index α > 3 for all tail subsets, ruling out a power-law upper tail. Simulating walks with the fitted daily step distribution, the authors find that for n ≈ 68 steps the simulated σ equals 54 m, exactly the empirical value, and the simulated histogram overlaps the field data. For n > 100 the simulated distribution is effectively Rayleigh, while for n < 40 it still shows heavy tails, documenting a smooth crossover from superdiffusive to normal diffusion as the number of aggregated steps grows. The authors conclude that no distinct ecological mechanisms at different spatio-temporal scales are needed to explain the two movement phases.","pith_inferences":["The one-week crossover time is not an independent prediction: because n is fitted, the conclusion rests on the assumption that a week contains about 68 turning-point steps. A direct count of steps could shift the crossover time and still preserve the general mechanism.","The same aggregation mechanism may explain apparent Lévy/Brownian dichotomies in other taxa where only one scale has been measured; whenever step distributions have finite variance, a crossover should occur at some scale, and its timing is set by the tail exponent and the cutoff.","The paper's framing suggests home ranges may be a consequence of short-term step statistics rather than an external constraint, but this is not tested; a direct test would compare movement in environments with and without physical boundaries."],"forward_implications":["For Didelphis aurita, dispersal distances within habitat fragments can be modeled as normal diffusion, with typical displacement growing as the square root of time, not as a faster superdiffusive process.","Short-term step data, collected over a single night, are sufficient to predict weekly displacement statistics without additional ecological parameters, provided the number of steps is known or estimated.","The crossover time is short because the daily step distribution has a sharp cutoff (b = 103.1 m); species with longer-tailed or weakly truncated step distributions would be expected to show Lévy-like dispersion over longer periods.","The scale-dependence of Lévy versus Brownian modes is an inherent statistical property of aggregation, so claims about a species being a 'Lévy walker' should be qualified by the temporal scale of the data."],"supporting_citations":[{"why":"Supplies the theoretical result on the ultraslow convergence of truncated Lévy flights to Gaussian, which motivates the expectation of a crossover as steps are summed.","marker":"[19]"},{"why":"Introduces the projection method used to define turning-point steps without an arbitrary discretization angle, from which the daily step-length distribution is estimated.","marker":"[2]"},{"why":"Provides the maximum-likelihood and model-comparison framework used to select the truncated Pareto distribution over Pareto and exponential alternatives.","marker":"[15]"},{"why":"Gives the numerical maximum-likelihood procedure for estimating the parameters of the truncated Pareto distribution, including the exponent μ.","marker":"[30]"},{"why":"Prior study of small-scale marsupial movement that this paper extends to a new dataset and connects to large-scale statistics.","marker":"[18]"},{"why":"The dataset used for weekly displacements comes from the cited population ecology study conducted in the same region.","marker":"[28]"},{"why":"The tail-index estimator used to assess the presence of a power-law upper tail in the weekly displacement distribution.","marker":"[35]"}],"fun_headline_variants":["Marsupial movement: Lévy daily, Brownian weekly","Truncated Lévy sums to Brownian in a week for Didelphis","No ecology needed: aggregation alone shifts Lévy to Brownian","Daily Lévy steps sum to ordinary diffusion after ~68 steps","One week of Didelphis aurita steps mimics Brownian motion"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The model assumes the weekly displacement equals the endpoint of a random walk with a fixed number n of independent daily steps, and n is fitted (n = 68) to match the observed weekly spread rather than measured directly from behavior.","fun_headline_variants_meta":{"raw":{"variants":["Marsupial movement: Lévy daily, Brownian weekly","Truncated Lévy sums to Brownian in a week for Didelphis","No ecology needed: aggregation alone shifts Lévy to Brownian","Daily Lévy steps sum to ordinary diffusion after ~68 steps","One week of Didelphis aurita steps mimics Brownian motion"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000713,"raw_usage":{"total_tokens":3255,"prompt_tokens":1039,"completion_tokens":2216,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":655,"completion_tokens_details":{"reasoning_tokens":2125}},"tokens_in":655,"tokens_out":2216,"duration_ms":13443,"temperature":1.0,"reasoning_tokens":2125,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T15:41:00.954637+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"As an independent test, an animal would need to be tracked continuously for a week, counting every turning-point step and its length, so that the weekly displacement can be compared with a Rayleigh distribution generated by summing exactly that many observed steps. If the independently counted n yields a substantially different σ or a non-Rayleigh shape, the purely statistical crossover would not hold.","supporting_citations":[{"cited_title":"Mantegna and H","cited_arxiv_id":null,"evidence_quote":"Supplies the theoretical result on the ultraslow convergence of truncated Lévy flights to Gaussian, which motivates the expectation of a crossover as steps are summed."},{"cited_title":"Scaling laws of marine predator search behaviour","cited_arxiv_id":null,"evidence_quote":"Introduces the projection method used to define turning-point steps without an arbitrary discretization angle, from which the daily step-length distribution is estimated."},{"cited_title":"Power-Law distributions in empirical data","cited_arxiv_id":null,"evidence_quote":"Provides the maximum-likelihood and model-comparison framework used to select the truncated Pareto distribution over Pareto and exponential alternatives."},{"cited_title":"Parameter Estimation for the Truncated Pareto Distribution","cited_arxiv_id":null,"evidence_quote":"Gives the numerical maximum-likelihood procedure for estimating the parameters of the truncated Pareto distribution, including the exponent μ."},{"cited_title":"and Vieira, M.V","cited_arxiv_id":null,"evidence_quote":"Prior study of small-scale marsupial movement that this paper extends to a new dataset and connects to large-scale statistics."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The dataset used for weekly displacements comes from the cited population ecology study conducted in the same region."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The tail-index estimator used to assess the presence of a power-law upper tail in the weekly displacement distribution."}],"review_version":1}