{"id":"aaf14e7d-95e5-45ce-a0a6-620148030c73","arxiv_id":"2608.07448","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Foraging sea turtle paths show long-memory loops that confine them to a feeding region and produce transient superdiffusion, captured by a data-driven stochastic model.","lead":"Scientists used satellite tracks of ten loggerhead sea turtles to build a random-motion model with long memory. The turtles' paths show large loops that keep them inside a feeding area and create a type of superdiffusive movement called anomalous looping.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Oscillatory VACF—the central evidence for anomalous looping—may be an artifact of interpolation and boundary confinement; no null-model test rules out a passive confined random walk producing the same signal.","rationale":"The reader's weakest assumption (interpolation artifacts) is real but only half the story. The more dangerous confound is that the oscillatory VACF—the paper's central observable—is exactly what one expects from a confined trajectory regardless of active looping. A particle in a bounded domain has a velocity autocorrelation that must become negative at the boundary-return timescale; with a roughly periodic crossing of a ~100 km domain at ~2 km/h, the half-period matches the reported 165–204 h. The paper's own model includes an asymmetric harmonic potential (SI Eq. S7) but the noise fed into it already contains the fitted cosine term, so the simulation cannot validate the origin of the oscillations. Credit is due for ruling out ocean currents with Lagrangian tracers (SI Sec. 2.5), ruling out chirality via turning angles (SI Sec. 1.8), and ruling out odd diffusivity via cross-correlations (SI Sec. 1.10); these are genuine independent checks. However, none of them compares against the natural null hypothesis of a memoryless (or short-memory) random walk in the same environment. The proposed test would settle this: if the null-model VACF shows oscillations comparable to Fig. 3, then the 'active looping' interpretation is unsupported and the central claim collapses to a confinement artifact; if not, the claim is strengthened. Because the preprint currently lacks this control, and the ten-turtle claim is overstated relative to the five turtles with reliable VACFs (SI Sec. 1.9), a conditional verdict remains appropriate, with the null-model comparison and data/code release as explicit conditions.","tokens_in":24524,"tokens_out":4640,"duration_ms":48729,"concrete_test":"Construct a null model: simulate a 2D overdamped Langevin/OU process (or correlated random walk) confined in x by the same asymmetric harmonic potential used in SI Eq. S7, with the same speed distribution and no memory beyond a short exponentially decaying VACF. Sample it at the same irregular times as Mokamba's Argos fixes, apply the same three-stage cleaning and linear interpolation (SI Sec. 1.3), and compute the longitudinal VACF. If the null VACF exhibits oscillations with amplitude and period comparable to Fig. 3 (i.e., peak negative correlation below roughly half the zero-lag value, period within a factor of 2 of 165–204 h), the oscillatory VACF is not sufficient evidence for active looping.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's central claim rests on the persistent oscillations in the longitudinal VACF (Fig. 3; Methods 'Data analysis'; SI Sec. 1.9). This is the only quantity that distinguishes 'anomalous looping' from a confined random walk. Two artifact sources are not excluded. First, linear interpolation of irregular Argos fixes at the mean sampling interval (Methods; SI Sec. 1.3) creates artificial straight-line segments whose velocity is constant over each gap; this can imprint periodic structure in the VACF at lags related to the gap pattern. Second, and more fundamentally, the foraging region is bounded in x (Fig. 1); the VACF of any stationary process confined by reflecting boundaries develops negative correlations and can oscillate at the boundary-collision timescale, without any internal looping memory. The authors rule out currents, chirality, and odd diffusivity (SI Secs. 1.8, 1.10, 2.5), but never test a null model of a passive particle with the same speed distribution and the same boundaries. Moreover, the stochastic model (SI Sec. 2.2) feeds the fitted oscillatory VACF directly into the Cholesky-synthesised noise; reproducing loops is therefore a consistency check, not independent confirmation. The 'sufficiency' claim in the Discussion is thus close to tautological unless the oscillatory VACF survives a null-model comparison.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes satellite tracking data from ten loggerhead sea turtles foraging off West Africa, focusing on one representative individual (Mokamba). It reports that the speed distribution has a Rayleigh center and exponential tail, that the longitudinal velocity autocorrelation function (VACF) shows persistent oscillations, and that the mean-squared displacement is transiently superdiffusive (α ≈ 1.68) before localizing. The authors construct an overdamped generalized Langevin equation in which Gaussian velocities are correlated by a fitted oscillatory VACF via Cholesky decomposition and confined by an asymmetric harmonic potential. Simulated trajectories reproduce large-scale loops, the VACF, and the MSD scaling, which the authors interpret as a new movement mode, 'anomalous looping,' distinct from Lévy walks and attributed to active, non-Markovian memory rather than currents, chirality, or odd diffusivity.","tokens_in":24840,"tokens_out":5580,"duration_ms":54882,"significance":"If the central claim holds, the paper makes a useful contribution by connecting movement ecology, active matter, and anomalous diffusion via a data-driven generalized Langevin approach. Its strengths include the breadth of analysis across ten individuals, the explicit checks against ocean currents, chirality, and odd diffusivity, and the scale-matching argument linking loop periods to the foraging-region width. However, the key evidence—the oscillatory VACF and the claim that it is actively generated—is not yet secure: the simulated VACF is fed into the model rather than predicted, and plausible artifacts from interpolation and passive confinement are not excluded. The novel classification of 'anomalous looping' is therefore plausible but needs stronger falsifiable tests before it can be accepted as a distinct movement mode.","major_comments":[{"comment":"The simulated VACF in Fig. 3 is not an independent prediction: the covariance matrix used in the Cholesky decomposition (SI Sec. 2.2, Eqs. S5–S6) is constructed from the same fitted experimental VACF, so the agreement between simulation and data in Fig. 3 holds by construction. The central sufficiency claim—that oscillatory VACFs plus confinement generate looping and localisation—therefore rests entirely on the simulated MSD and trajectory morphology. I recommend adding a null-model control with the same boundary potential and speed distribution but an exponentially decaying (non-oscillatory) VACF, and showing that it fails to reproduce the looping and the intermediate-time superdiffusion; without such a control, the evidence does not distinguish active looping from a confined correlated random walk.","section":"Data-driven stochastic modelling; SI Sec. 2.2"},{"comment":"The central oscillatory VACF is computed on data that were linearly interpolated to a regular grid at the mean sampling interval after removing large gaps (Methods; SI Sec. 1.3). Linear interpolation produces artificial straight-line segments whose constant velocities can imprint periodic structure in the VACF at lags related to the gap pattern. The paper does not test whether the observed oscillations survive alternative treatments, such as computing the VACF directly on the irregular raw data with slotting or simulating a known process and passing it through the same cleaning and interpolation pipeline. This test is needed to establish that the oscillations in Fig. 3 are a property of the turtle's movement rather than of the resampling procedure.","section":"Methods, Data analysis; SI Sec. 1.3"},{"comment":"A passive particle confined by reflecting or harmonic boundaries can develop negative velocity correlations and oscillations at the boundary-return timescale, so confinement alone may produce the qualitative shape of the longitudinal VACF. The authors rule out currents, chirality, and odd diffusivity (SI Secs. 1.8, 1.10, 2.5) but do not test the null model of a passive, or exponentially correlated, random walk with the same speed distribution and the same asymmetric boundary potential. Adding such a test is necessary to support the claim that the oscillations reflect actively generated loops; as written, the 'active' interpretation is not uniquely supported.","section":"Results; SI Sec. 1.9"},{"comment":"The main text states that the main results, including oscillatory VACFs, were confirmed for all ten turtles, but SI Sec. 1.9 reports that Kamoka and Mingo do not show periodic oscillations and that Bemvinda and Kika have long gaps that compromise the VACF, and the stochastic model is constructed for only five of the ten turtles (Mokamba, Catchupa, Goody, Olympia, Papaya). This selection should be stated transparently in the main text, and the fraction of turtles exhibiting the oscillation signature should be quantified; otherwise the generality claim is stronger than the evidence supports.","section":"Results; SI Secs. 1.9, 2.4"},{"comment":"The confining potential parameters were 'adapted by trial and error to match the experimental results' (Methods, step 5), and the MSD exponents α_1 and α_2 are reported without uncertainty estimates. The agreement between simulated and experimental MSD in Fig. 4 is therefore at least partly a fit rather than a prediction. Please provide bootstrap or fitting uncertainties for the exponents and show the sensitivity of α_sim to the boundary stiffness parameters k1 and k2.","section":"Methods, Computer simulations; Fig. 4"}],"minor_comments":[{"comment":"The sentence listing turtles for stochastic modelling repeats 'Goody' twice ('Catchupa, Goody, Olympia, Papaya and Goody'); one name is presumably meant to be another turtle or the list should contain only the four distinct individuals.","section":"SI Sec. 2.4"},{"comment":"Reference entries [51]–[55] contain the placeholder text 'Duplicate citation removed for clean compilation' and should be deleted before publication.","section":"References"},{"comment":"The caption contains the typo 'Diffferent lines'; it should read 'Different lines'.","section":"Fig. 3 caption"},{"comment":"The text refers to the 'non-Makovian nature' of a stochastic process; the intended word is 'non-Markovian'.","section":"SI Sec. 1.9"},{"comment":"The symbols α_1 and α_2 are used for the simulated and experimental MSD exponents but are not defined in the caption or the main text; please define them in the caption.","section":"Fig. 4 caption"},{"comment":"The phrase 'from insects to birds, marine predators, mammals and even humans' should read 'from insects to birds, marine predators, mammals, and even humans' for grammatical parallelism.","section":"Abstract"},{"comment":"The expression 'whisker 3.0' in the box-plot description is not standard; please clarify that the whiskers extend to 3.0 times the interquartile range.","section":"SI Sec. 1.3"}],"recommendation":"major_revision","confidential_remarks":"I agree with the reader's conditional assessment: the paper is methodologically interesting and likely salvageable, but the central evidence for 'anomalous looping' is not yet robust. The most important fixes are the null-model comparisons (non-oscillatory VACF with the same boundaries; passive confined walker) and the robustness checks against interpolation artifacts. I also encourage the editor to ask the authors to align the main-text generality claims with the five-turtle subset used for modelling. If the authors add the missing controls and quantify uncertainty, the paper could become a solid contribution to movement ecology and statistical physics."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know: this paper has a genuinely new empirical observation—oscillatory velocity autocorrelations in foraging sea turtle tracks, with a loop-based mechanism for transient superdiffusion—but the evidence for it rests on a signal that might be an artifact of the data processing. The model 'validation' is largely a consistency check, not independent confirmation. Still, there is enough substance that it deserves a serious referee; it just needs heavy revision.\n\nWhat is actually new: the claim that foraging turtles move in large-scale loops detected via persistent oscillations in the longitudinal VACF, distinct from Lévy walks, is not in the prior literature. They also do a decent job ruling out ocean currents (Lagrangian tracer simulations), chirality (turning-angle symmetry), and odd diffusivity (cross-correlation). The generalized Langevin framework is borrowed from bumblebee work, but the specific finding, 'anomalous looping,' is new.\n\nThe paper is careful in places: multiple turtles, SI with per-turtle analyses, scale matching between loop half-period and foraging-region width. Goody and Papaya issues are honestly discussed.\n\nNow the soft spots, which are serious. First, the data are irregular Argos fixes, linearly interpolated at the mean sampling interval. That can imprint artificial straight-line segments and corrupt the VACF at lags related to gap patterns. Second, the foraging region is bounded; the VACF of a passively confined random walk can develop negative correlations and oscillate at the boundary-collision timescale. The authors never test a null model—say, a confined Ornstein-Uhlenbeck process with the same speed distribution and same boundaries—to see if it produces the same oscillatory VACF and MSD. Without that, the central claim is fragile.\n\nThird, the model feeds the fitted experimental VACF directly into the Cholesky covariance matrix, so the loop reproduction in Fig. 3 is by construction. The boundary potentials are tuned by trial and error to match the MSD plateau. The 'sufficiency' statement in the Discussion is close to tautological.\n\nFourth, the ten-turtle claim does not match the modeling: the SI says five turtles were used for stochastic modeling, while the main text says all main results were confirmed across ten. The MSD exponents have no error bars, and the reference list contains placeholder entries ([51]–[55], 'Duplicate citation removed for clean compilation'). Data and code are not released.\n\nBottom line: if the oscillatory VACF is real and robust to null-model testing, this is a useful new movement mode. As it stands, it is a plausible but under-supported empirical claim. I would send it to peer review—the empirical pattern is important enough to warrant scrutiny—but I would not cite it as established until the artifacts are ruled out and the analysis is reproduced on raw data.\n\nRecommendation: engage seriously; request data/code release, a null-model comparison, and reconciliation of the turtle counts.","headline":"Plausible new movement mode, but the oscillatory VACF at the heart of it may be an interpolation/boundary artifact; worth refereeing, not worth citing yet.","tokens_in":25352,"tokens_out":2520,"would_cite":false,"duration_ms":24715,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["82C31","60J60"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that foraging loggerhead turtles move in actively generated large-scale loops whose oscillatory velocity correlations confine them to a feeding region and produce superdiffusion over days.","keywords":["anomalous diffusion","animal movement ecology","velocity autocorrelation function","generalised Langevin equation","superdiffusion","loggerhead sea turtle","active particles","data-driven stochastic modelling"],"falsifier":"Recompute the longitudinal velocity autocorrelation and the time-averaged mean-squared displacement from the raw irregular telemetry without linear interpolation, using variable-lag estimators on the original gap-separated segments; if the persistent oscillations in $\\langle v_x(0)v_x(\\tau)\\rangle$ disappear or the intermediate-time exponent drops to $\\alpha\\le 1$, the central claim fails. A complementary test is a new deployment with high-resolution GPS logging at sub-minute intervals: a foraging turtle whose velocity autocorrelation decays monotonically and whose path shows no loops would falsify the loop-based mechanism.","tokens_in":24308,"feed_emoji":"🐢","tokens_out":15854,"duration_ms":147645,"temperature":0.7,"pith_summary":"This paper sets out to show that foraging loggerhead sea turtles move in large, actively generated loops that keep them inside a bounded feeding area while making their spreading superdiffusive on time scales of days. From satellite tracks of ten turtles, the authors extract speed distributions with Gaussian centres and exponential tails, sustained oscillations in the velocity autocorrelation functions, and a mean-squared displacement growing as $t^\\alpha$ with $\\alpha\\simeq 1.68$ before the motion localises. They then construct a data-driven stochastic generalised Langevin equation whose noise has exactly those oscillatory autocorrelations and which is confined by an asymmetric potential, and show that it reproduces the loops, the superdiffusion, and the eventual confinement. On this basis they identify anomalous looping as a movement mode distinct from L\\'evy walks, where superdiffusion instead arises from heavy-tailed jump lengths, and they rule out ocean currents, chirality, and odd diffusivity as explanations. If the claim is right, long-term velocity memory deserves the same status as step-length distributions in models of animal movement.","feed_headline":"Foraging turtles show a new movement mode: anomalous looping","feed_subtitle":"Satellite tracks show oscillatory velocity correlations that confine turtles to feeding grounds while making their search superdiffusive…","key_machinery":"The central object is the time-averaged velocity autocorrelation function $C(\\tau)=\\langle v(0)v(\\tau)\\rangle_t$ for each spatial component, fitted by $f(t)=A\\exp(-Bt)+C\\cos(Dt)$. The exponential part is ordinary short-time noise decay; the persistent cosine part is the statistical signature of loops. The carrying mechanism is an overdamped generalised Langevin equation $\\dot{x}=F(x)+\\zeta(t)$, where $\\zeta$ is coloured Gaussian noise whose covariance is the fitted VACF, generated by Cholesky decomposition of the covariance matrix, and $F(x)$ is an asymmetric harmonic potential in the longitudinal direction. A scale-matching step links the oscillation half-period $T_{x,1/2}$ to the mean displacement $\\sqrt{\\langle x^2(T_{x,1/2})\\rangle}$, showing that the loops are about as wide as the longitudinal extent of the foraging region.","core_discovery":"The paper's central claim is that the turtle's foraging path is statistically encoded by an oscillatory velocity autocorrelation function, and that this oscillation together with spatial confinement is sufficient to generate the large-scale looping and the approach to localisation visible in the data. The authors build the simplest model with those ingredients: velocities drawn from Gaussians whose covariance is fixed by the fitted form $C(\\tau)=Ae^{-B\\tau}+C\\cos(D\\tau)$, and an asymmetric harmonic potential representing the coastline on one side and open water on the other. Simulated trajectories reproduce the empirical superdiffusive exponent and the looped, eventually localised path. The supporting analysis excludes passive advection by ocean currents (Lagrangian tracers disperse differently), chirality (turning-angle distributions are symmetric about zero), and odd diffusivity (no cross-correlation between $v_x$ and $v_y$). The name anomalous looping captures the mode: anomalous because of transient superdiffusion, looping because of the sustained velocity correlations, and distinct from L\\'evy walks because the velocity statistics are not heavy-tailed.","pith_inferences":["Editorial inference: if oscillatory velocity autocorrelations are a generic loop signature, existing archival tracking datasets on other species -- many currently classified as L\\'evy or correlated-random-walk movers -- could be re-examined for VACF oscillations without collecting new data.","The model's two ingredients suggest a concrete algorithmic recipe for bounded-area search: an agent that imposes sinusoidal velocity correlations and a soft reflecting boundary can generate effective superdiffusion without heavy-tailed step lengths.","Editorial caution: this manuscript's compiled reference list contains 'Duplicate citation removed' placeholders, and some in-text citation numbers no longer map cleanly to the printed entries; the bibliographic identity of load-bearing references should be verified against the original sources.","Because the empirical statistics come from irregularly sampled Argos locations that were linearly interpolated, the decisive next test is high-resolution GPS or accelerometer tracking: sustained VACF oscillations at sub-hourly sampling would confirm the mechanism, while a monotonically decaying VACF would implicate interpolation artifacts."],"forward_implications":["A superdiffusive scaling exponent alone does not identify a L\\'evy walk; the same $\\alpha \\simeq 1.6$--$1.7$ arises here from oscillatory velocity memory inside a bounded region.","Velocity autocorrelation functions become a primary diagnostic: persistent oscillations in $C(\\tau)$ are the statistical fingerprint of looping, even when the path appears random.","The two-ingredient model -- Gaussian velocities with the measured correlations plus a soft boundary -- suffices to reproduce looping, superdiffusion, and localisation, so complex foraging paths need not require elaborate behavioural rules.","The loop size, estimated from the VACF half-period, matches the width of the foraging region (roughly 60--100 km), suggesting the loops are scaled to the habitat.","Anomalous looping is an active, self-generated movement mode, not passive current advection, chiral swimming, or odd-diffusive transport."],"supporting_citations":[{"why":"The satellite-tracking dataset for the ten turtles and the habitat variables that define the foraging region.","marker":"1"},{"why":"The bumblebee-flight studies the paper says it adapts: constructing a generalised Langevin model directly from experimental velocity measurements.","marker":"5,6"},{"why":"The cell-migration and memory-kernel constructions that motivate using non-Markovian velocity correlations as model input.","marker":"7,26"},{"why":"The L\\'evy-walk review that defines the heavy-tailed-jump mechanism from which anomalous looping must be distinguished.","marker":"27"},{"why":"The active-particle review cited for chiral motion, which the paper rules out through symmetric turning-angle distributions.","marker":"19"},{"why":"The odd-diffusivity effect the paper excludes by finding no cross-correlation between orthogonal velocity components.","marker":"45"},{"why":"The active Ornstein-Uhlenbeck particle model, which is recovered when the noise is exponentially correlated Gaussian.","marker":"46"},{"why":"The search-efficiency literature invoked to temper the conjecture that anomalous looping could be an optimal bounded-habitat search strategy.","marker":"47"}],"fun_headline_variants":["Turtle foraging loops generate superdiffusive search","Sea turtles' anomalous looping: a new movement mode","Looping turtles reveal a new search strategy","Anomalous looping in turtles: superdiffusion without Levy","Oscillatory velocity correlations drive turtle looping"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that linearly interpolating the irregular Argos locations onto a regular grid at the mean sampling interval preserves the real movement statistics; if interpolation artifacts create the oscillatory correlations or the superdiffusive scaling, the loop-based mechanism is not established.","fun_headline_variants_meta":{"raw":{"variants":["Turtle foraging loops generate superdiffusive search","Sea turtles' anomalous looping: a new movement mode","Looping turtles reveal a new search strategy","Anomalous looping in turtles: superdiffusion without Levy","Oscillatory velocity correlations drive turtle looping"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000221,"raw_usage":{"total_tokens":1463,"prompt_tokens":971,"completion_tokens":492,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":587,"completion_tokens_details":{"reasoning_tokens":418}},"tokens_in":587,"tokens_out":492,"duration_ms":5877,"temperature":1.0,"reasoning_tokens":418,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T04:25:31.779335+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the longitudinal velocity autocorrelation and the time-averaged mean-squared displacement from the raw irregular telemetry without linear interpolation, using variable-lag estimators on the original gap-separated segments; if the persistent oscillations in $\\langle v_x(0)v_x(\\tau)\\rangle$ disappear or the intermediate-time exponent drops to $\\alpha\\le 1$, the central claim fails. A complementary test is a new deployment with high-resolution GPS logging at sub-minute intervals: a foraging turtle whose velocity autocorrelation decays monotonically and whose path shows no loops would falsify the loop-based mechanism.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The satellite-tracking dataset for the ten turtles and the habitat variables that define the foraging region."},{"cited_title":", author Denisov, S","cited_arxiv_id":null,"evidence_quote":"The L\\'evy-walk review that defines the heavy-tailed-jump mechanism from which anomalous looping must be distinguished."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The active-particle review cited for chiral motion, which the paper rules out through symmetric turning-angle distributions."},{"cited_title":", author Epstein, J","cited_arxiv_id":null,"evidence_quote":"The odd-diffusivity effect the paper excludes by finding no cross-correlation between orthogonal velocity components."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The active Ornstein-Uhlenbeck particle model, which is recovered when the noise is exponentially correlated Gaussian."},{"cited_title":", author Textor, J","cited_arxiv_id":null,"evidence_quote":"The search-efficiency literature invoked to temper the conjecture that anomalous looping could be an optimal bounded-habitat search strategy."}],"review_version":1}