{"id":"0fcc0ffb-7a0b-4e95-a687-b915f08e2731","arxiv_id":"2602.07252","paper_version":2,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Introduces an online change-point detection framework that operates intrinsically on distribution-valued data via tangent-space linearization in the 2-Wasserstein space.","lead":"The paper develops a method to detect changes in streaming batches of data where each batch is a full probability distribution, by embedding them into a tangent space around a central distribution using Wasserstein geometry. General readers might care because many real monitoring tasks involve data whose shape or spread matters more than its average.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Fréchet barycenter estimation from finite initial data may be unstable, breaking the tangent-space linearization for non-trivial shifts","rationale":"The reader's weakest assumption already isolates the same point. Because the central empirical claim (reduced delay at matched ARL0) is demonstrated only under the maintained assumption that the barycenter is well-estimated and the linearization remains valid, a direct sensitivity check on initial-sample size would either confirm robustness or show that the performance gain is conditional on favorable initialization.","tokens_in":1658,"tokens_out":366,"duration_ms":31175,"concrete_test":"Re-run the synthetic experiments of §5.1 with initial-window lengths N0 = 50, 100, 200 while keeping the post-change shift size fixed; recompute empirical ARL0 and average detection delay for the proposed tangent-space detector versus the moment-based baseline. If the delay advantage disappears or ARL0 deviates by >20% for N0 ≤ 100, the barycenter-estimation assumption is the binding limitation.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The method defines the tangent space at an estimated pre-change Fréchet barycenter computed from the initial segment. For the adapted multivariate CUSUM or EWMA statistics to retain their nominal ARL0 and power, this barycenter must be close to the true one and the post-change distributions must lie in a neighborhood where the logarithmic map is a faithful linearization. In the 2-Wasserstein space over non-Gaussian or high-dimensional supports, finite-sample barycenter estimation error (even with the usual iterative algorithms) can rotate the tangent basis and inflate variance, so that the claimed reduction in detection delay at matched ARL0 no longer holds once the initial window is realistic.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript proposes an online change-point detection method for distribution-valued streaming data by embedding each batch into the tangent space of the 2-Wasserstein space at an estimated pre-change Fréchet barycenter and then applying adapted multivariate CUSUM and EWMA statistics. It claims theoretical guarantees for the resulting detectors together with empirical evidence of reduced detection delay at matched ARL0 relative to moments-based and model-free baselines on both synthetic and real-world examples. Reproducible code is provided.","tokens_in":1806,"tokens_out":472,"duration_ms":36159,"significance":"If the central claims hold, the work offers a principled extension of change-point detection beyond Euclidean summaries, preserving geometric information that moment-based methods discard. The explicit provision of open code is a clear strength that supports reproducibility and further scrutiny. The approach could find use in applications where data arrive as empirical distributions rather than fixed-dimensional vectors.","major_comments":[{"comment":"§3.1–3.2: The tangent-space linearization is defined at the Fréchet barycenter estimated from the initial window. No finite-sample error bounds or sensitivity analysis are given for how barycenter estimation error propagates into the tangent coordinates or into the nominal ARL0 of the adapted CUSUM/EWMA statistics; this assumption is load-bearing for the claimed reduction in detection delay.","section":"§3.1–3.2"},{"comment":"§5: The experimental results report favorable detection delays, yet the section supplies neither the number of Monte Carlo replications, standard-error bars on the delay and ARL0 figures, nor explicit rules for data exclusion or hyper-parameter selection; without these details the robustness of the performance comparison cannot be verified.","section":"§5"}],"minor_comments":[{"comment":"Notation for the logarithmic map and tangent vectors is introduced without a short reminder of the precise definition used; a one-sentence clarification would aid readers unfamiliar with Wasserstein geometry.","section":"§2"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within scope for stat.ME. The citation list is reasonably current but could be strengthened by explicit comparison with the most recent Wasserstein-based CPD literature."},"author_rebuttal":{"model":"grok-4.3","summary":"We are grateful to the referee for their constructive feedback and for recognizing the potential significance of our work on intrinsic change-point detection for distribution-valued data. We address each of the major comments below.","responses":[{"response":"The referee correctly identifies that our theoretical guarantees assume the pre-change Fréchet barycenter is fixed or estimated consistently from the initial window. While we do not provide explicit finite-sample bounds in the current manuscript, the asymptotic theory ensures that the estimation error becomes negligible as the initial window size increases, preserving the control of ARL0 and the detection properties. To strengthen the presentation, we will include a sensitivity analysis in the revised manuscript, consisting of additional simulations that vary the initial window size and quantify the impact on detection delay and ARL0. This will demonstrate the robustness of the approach in finite samples.","revision_made":"yes","referee_comment":"[§3.1–3.2] §3.1–3.2: The tangent-space linearization is defined at the Fréchet barycenter estimated from the initial window. No finite-sample error bounds or sensitivity analysis are given for how barycenter estimation error propagates into the tangent coordinates or into the nominal ARL0 of the adapted CUSUM/EWMA statistics; this assumption is load-bearing for the claimed reduction in detection delay."},{"response":"We agree that these experimental details are essential for verifying the robustness of our comparisons. In the revised version, we will add the number of Monte Carlo replications performed, include standard error bars on all reported figures, and provide explicit descriptions of the hyper-parameter tuning procedure and any data preprocessing or exclusion rules applied in the experiments.","revision_made":"yes","referee_comment":"[§5] §5: The experimental results report favorable detection delays, yet the section supplies neither the number of Monte Carlo replications, standard-error bars on the delay and ARL0 figures, nor explicit rules for data exclusion or hyper-parameter selection; without these details the robustness of the performance comparison cannot be verified."}],"tokens_in":1324,"tokens_out":444,"duration_ms":47874,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"This paper's main contribution is a method for online change point detection that operates on distribution-valued data by linearizing around a Fréchet barycenter in Wasserstein space and then using adapted multivariate detectors. It is new in combining the intrinsic geometry of the 2-Wasserstein space with classical sequential analysis tools for streaming settings. The work does well in providing both theoretical guarantees and empirical evidence from synthetic and real data that shows improved detection delays compared to baselines that rely on moments or model-free approaches. Releasing the code is also a plus for reproducibility. The potential soft spot is in the estimation of the pre-change Fréchet barycenter from finite initial observations. If this estimate is off, especially with complex or high-dimensional distributions, the tangent space linearization may not hold accurately, which could undermine the claimed performance advantages and the control of ARL0. The stress test raises a fair point here that deserves checking in the full derivations and experiments. Overall, this is for researchers focused on statistical monitoring of non-Euclidean or distributional data streams. A reader interested in extending change detection beyond Euclidean assumptions will find practical value and ideas to build on. The paper shows clear thinking on the problem and engages with relevant literature, so it deserves a serious referee. I recommend sending it for peer review.","headline":"The paper adapts Wasserstein tangent spaces for online CPD on distributions and demonstrates better detection performance, but finite-sample barycenter stability needs scrutiny.","tokens_in":2271,"tokens_out":331,"would_cite":true,"duration_ms":33967,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":{"model":"grok-4.3","evidence":[{"relation":"unclear","rs_module":"IndisputableMonolith/Cost/FunctionalEquation.lean","rs_theorem":"washburn_uniqueness_aczel","paper_passage":"mapping each empirical distribution to a tangent space relative to a pre-change Fréchet barycenter, yielding a reference-centered local linearization of 2-Wasserstein space"},{"relation":"unclear","rs_module":"IndisputableMonolith/Foundation/BranchSelection.lean","rs_theorem":"branch_selection","paper_passage":"Proposition 3.4 (Radial Isometry) … W₂²(¯µ, µ) = ∥v(µ)∥²_{L²(¯µ)}"}],"headline":"Wasserstein tangent-space CPD via Fréchet barycenter and OT linearization lies outside RS forcing chain","alignment":"orthogonal","rationale":"The paper's core construction (mapping empirical measures to L2(¯µ) tangent fields via barycentric projection of OT plans, then applying MFPCA + Hotelling T²/SPE with ARL0 calibration) is standard Wasserstein geometry for distribution-valued streams. It invokes no J-cost, cosh(ρ ln φ), ratio-symmetric forcing, φ-ladder, 8-tick periodicity, or parameter-free constant derivation. RS modules (e.g., Cost.FunctionalEquation, Foundation.AlphaCoordinateFixation, Foundation.BranchSelection) derive J(x) = ½(x + x⁻¹) − 1 and its geometric consequences from a single distinction; the present work operates in an unrelated statistical domain with no overlap or contradiction.","tokens_in":64269,"confidence":"high","tokens_out":373,"duration_ms":13433,"cache_read_input_tokens":38528,"cache_creation_input_tokens":0},"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Treating streaming batches as points in 2-Wasserstein space and mapping them to a tangent plane at the initial barycenter lets standard detectors catch shape changes that moments miss.","keywords":["change point detection","Wasserstein space","distribution-valued data","online monitoring","Fréchet barycenter","tangent space","sequential detection"],"falsifier":"A simulation in which the post-change distributions lie far from the initial barycenter, the tangent detector fails to flag the shift within the claimed delay, yet a direct nonlinear Wasserstein-distance monitor detects it promptly.","tokens_in":2556,"feed_emoji":"📈","tokens_out":702,"duration_ms":29833,"temperature":0.7,"pith_summary":"The paper develops an online change-point procedure that works directly on batches of data whose empirical distributions vary over time. Rather than first collapsing each batch to a vector of moments or features, it keeps the full distributional information by embedding the batches in the space of probability measures equipped with the 2-Wasserstein distance. A single reference barycenter is estimated from the first observations; every later empirical distribution is then projected onto the tangent space at that barycenter, turning the nonlinear geometry into ordinary Euclidean vectors. Classical multivariate monitoring statistics can therefore be run on these vectors. The resulting procedure is shown to register shifts in shape or geometry with shorter delay than moment-based or model-free alternatives while preserving the same average run length under no change.","feed_headline":"Tangent map detects distribution shifts with less delay","feed_subtitle":"Projecting batches onto the local Euclidean plane at the pre-change barycenter lets standard charts flag shape changes sooner than moment or","key_machinery":"Reference-centered tangent-space representation of 2-Wasserstein space, which converts the nonlinear geometry of distributions into a Euclidean vector field on which classical change-point detectors operate.","core_discovery":"Changes in the law of a stochastic process taking values in the space of probability measures are detected by estimating a pre-change Fréchet barycenter from initial data, mapping each new empirical distribution to the tangent space at that barycenter, and applying adapted sequential monitoring statistics to the resulting tangent vectors.","pith_inferences":["Periodic recomputation of the reference barycenter could extend reliable operation over long streams whose distributions drift slowly.","The same linearization step may be useful for other optimal-transport distances or for data whose geometry is given by a different metric.","In applications where only a small initial window is available, the quality of the barycenter estimate becomes the dominant factor limiting early detection."],"forward_implications":["Detection delay decreases for shifts that alter variance, multimodality, or support while low-order moments stay fixed.","The same average run length under no change is maintained because the tangent vectors inherit the asymptotic behavior of the original monitoring statistics.","Theoretical control of false-alarm probability follows from the local Euclidean structure once the barycenter is fixed.","The method applies to any data type whose batches admit empirical distribution estimates, including images, sensor histograms, and compositional observations."],"fun_headline_variants":["Tangent maps at Frechet barycenter detect distribution shifts","Mapping to tangent space reveals changes in distribution streams","Pre-change barycenter linearizes Wasserstein space for CPD","Adapted statistics on tangent fields monitor distribution batches"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"Distributional changes remain close enough to the initial reference barycenter that the tangent-space approximation stays accurate and the barycenter itself can be estimated reliably from the first observations.","fun_headline_variants_meta":{"raw":{"variants":["Tangent maps at Frechet barycenter detect distribution shifts","Mapping to tangent space reveals changes in distribution streams","Pre-change barycenter linearizes Wasserstein space for CPD","Adapted statistics on tangent fields monitor distribution batches"]},"model":"grok-4.3","cost_usd":0.009124,"raw_usage":{"total_tokens":4058,"prompt_tokens":602,"num_sources_used":0,"completion_tokens":53,"cost_in_usd_ticks":91237000,"prompt_tokens_details":{"text_tokens":602,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":3403,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":602,"tokens_out":53,"duration_ms":43184,"temperature":1.0,"reasoning_tokens":3403,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-22T11:17:56.886915+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A simulation in which the post-change distributions lie far from the initial barycenter, the tangent detector fails to flag the shift within the claimed delay, yet a direct nonlinear Wasserstein-distance monitor detects it promptly.","supporting_citations":[],"review_version":1}