{"id":"c0ecabcb-a43c-4798-8097-5fb9742815d3","arxiv_id":"2607.07061","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Persistent homology signatures of the solid phase in synthetic porous media correlate with diffusion-release regimes even after stratifying by target porosity, and classify early-fast, late-release, and long-tail behavior with 0.64–0.76 test accuracy.","lead":"This paper uses a shape-counting technique called persistent homology to describe the solid skeleton of porous materials and links those shape fingerprints to how fast dissolved substances escape. It finds that even at the same porosity, samples with more complex solid topology tend to release solute more slowly, and the fingerprints can help sort release behaviors with a simple classifier.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unmeasured actual porosity may confound the claimed topology–release association within target-porosity strata.","rationale":"The reader's weakest assumption is exactly this porosity confound, and it is the most load-bearing because it attacks the internal validity of the central comparison. The paper itself acknowledges only approximate target matching, so the burden is on the authors to show realized porosity is not driving the effect. Other concerns (reproducibility, missing comparisons) are secondary. I agree with CONDITIONAL; the paper's conclusions are plausible but conditional on resolving this confounder.","tokens_in":11831,"tokens_out":4799,"duration_ms":49461,"concrete_test":"For each generated sample, compute the exact porosity from the union of spheres (or from the voxelized geometry used for FEM). First, compare actual-porosity distributions across the three regime groups within each target-porosity stratum (e.g., Kruskal-Wallis). If they differ, the claim is confounded. Second, include actual porosity as a covariate in the multinomial logistic regression or as an additional feature; if PH features no longer improve classification or the effect sizes in Table 3 drop to negligible, the central claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that release depends on multiscale solid organization beyond pore-space amount—rests on the porosity-controlled analysis in §4.2. That analysis stratifies by target porosity, but §4.1 only states that samples are generated to 'approximately match' a prescribed target porosity and gives no measurement or report of the realized porosity. Because the solid phase is random, actual porosity presumably varies within each target stratum. If actual porosity is lower (more solid) in long-tail samples than in early-fast samples, then the larger H0/H1 persistence sums observed in long-tail samples (Table 4) may reflect a difference in solid fraction rather than multiscale organization. The paper provides no check that actual porosity is balanced across regimes; without that, the headline inference is vulnerable to a straightforward confounder.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes persistent homology (PH) of the solid phase as a fast, geometry-based descriptor for diffusive release from porous media. Synthetic samples are generated from overlapping spherical grains in six structural classes at target porosities of 40%, 60%, and 80% and at two voxel resolutions (16^3 and 32^3). Release curves are computed with a tetrahedral finite-element approximation of a mixed Dirichlet–Neumann diffusion problem. Release-curve regimes (early-fast, late-release, long-tail) are defined by quantiles of curve summaries. Within each target porosity, the authors compare 18 PH summaries across the three regimes using Kruskal–Wallis tests with Benjamini–Hochberg correction and report large effect sizes, especially for persistence sums. They also train a multinomial logistic regression on the PH features and report test accuracies between 0.641 and 0.759 at fixed porosities. Timing experiments show PH feature extraction is orders of magnitude faster than the finite-element release solve.","tokens_in":11966,"tokens_out":6733,"duration_ms":69803,"significance":"If the central claim holds—that solid-phase topology is associated with release behavior beyond pore-space amount—this is a potentially useful, interpretable screening descriptor for porous materials. The statistical machinery is generally appropriate: nonparametric tests, multiple-testing correction, effect sizes, stratified train–test splits, and balanced classes. The direction of the effect is consistent across porosity levels and grid resolutions, and the paper is appropriately cautious at the end of §4.2 that the analysis shows association, not causation. The main weakness is that the 'beyond pore-space amount' claim rests on stratification by target porosity, while actual realized porosity is never measured or reported. Since both the PH features and the release curves are functions of the same random geometry, the reported associations could be confounded by differences in realized porosity within target-porosity strata. The paper also evaluates classification only on deliberately selected extreme release-regime groups, which may overstate the screening utility. These issues are fixable within the manuscript's scope.","major_comments":[{"comment":"The central claim—that release behavior depends on multiscale solid organization 'even within each target-porosity level'—is supported only by stratifying on target porosity. However, §4.1 states that samples are generated only to 'approximately match a prescribed target porosity,' and no realized porosity is reported or used. Since sphere placements are random, realized porosity almost surely varies within each target stratum. If actual porosity is lower (more solid) in long-tail samples than in early-fast samples, the larger H0/H1/H2 persistence sums in Table 4 could reflect differences in solid fraction rather than in multiscale organization. Please report the distribution of realized porosity by regime and porosity level, and add a sensitivity analysis, e.g., including realized porosity as a covariate in the §4.3 multinomial model, or stratifying by binned realized porosity. Without","section":"§4.1 and §4.2 (Abstract; Table 4)"},{"comment":"The reported classification accuracies (0.641–0.759) are computed on labels obtained by selecting extreme quantiles of release-curve summaries from 720 samples per porosity condition: top 180 by early release, then top 180 by final unreleased fraction, then top 180 by late gain; the remaining 180 samples are discarded. This extreme-group design makes the three classes artificially well separated, so an accuracy above the 1/3 chance level overstates the model's ability to screen a natural distribution of candidate geometries. To support the stated screening use case, evaluate the classifier on the unassigned 180 samples or on all 720 samples (e.g., by adding an 'unassigned' class or by predicting a continuous curve summary), and report those results alongside the extreme-regime accuracy. At minimum, state this selection explicitly as a limitation of §4.3.","section":"§4.3 (Table 5)"}],"minor_comments":[{"comment":"The grid notation '163' and '323' should be 16^3 and 32^3; use superscripts to avoid ambiguity.","section":"§4.1, Table 5"},{"comment":"The numerical values of t_end, t_20%, t_50%, the diffusion coefficient D, and the time step Δt are not reported. These are needed to reproduce the release curves and the regime definitions.","section":"§4.1"},{"comment":"The text says structural classes are 'sampled approximately uniformly' and porosity is 'approximately match[ed]'; please describe the sampling/placement procedure and report the actual class counts and porosity ranges per condition.","section":"§4.1, Table 1"},{"comment":"Only group means are reported. Since n = 540 per porosity level, the very small corrected q-values partly reflect sample size; reporting standard deviations or confidence intervals would help readers judge the magnitude of separation.","section":"Table 4"},{"comment":"A code/data availability statement would improve reproducibility. If the finite-element solver and PH pipeline are not released, please state how the reported numbers can be obtained.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The main obstacle is the realized-porosity confound identified in the stress-test note. The fix is straightforward—compute and report realized porosity from the existing sphere/voxel geometry, show balance across regimes, and run a sensitivity analysis with realized porosity as a covariate. If this is done, I expect the central claim to become defensible and the paper could proceed to acceptance after the classification-selection caveat in §4.3 is also addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dave — quick take on arXiv:2607.07061. The paper is a competent, honest pilot study: it uses solid-offset persistent homology summaries to distinguish diffusive release regimes (early-fast, late-release, long-tail) in synthetic sphere packs, and shows the association persists within fixed target-porosity levels. That specific combination is new relative to the PH-for-permeability/trapping literature. The statistics are careful: Kruskal–Wallis with Benjamini–Hochberg correction, effect sizes, stratified splits, and the authors explicitly say association is not causation. Timing comparisons are fair. All that earns a serious referee.\n\nThe soft spots are real but addressable. The main one is the porosity control. Section 4.1 says samples are generated to 'approximately match' a prescribed target porosity, but actual porosity is never measured or reported. Realized porosity presumably varies within each stratum, and if long-tail samples are denser (lower porosity), the larger persistence sums could reflect solid fraction rather than multiscale organization. That confounder sits right under the central claim. The paper needs a reported porosity balance check or a covariate adjustment. Second, no code, data, or simulation parameters (D, Δt, tend, grain distributions) are provided, so the experiments are not reproducible as written. Third, no comparison against classical geometry descriptors (measured porosity, tortuosity, chord-length distributions), so the practical screening claim is not fully demonstrated. Finally, the release-regime labels are selected as extreme quantiles of Q(t) within each porosity level; that is not circular, but it does limit the claim to distinguishing tails, and the 0.64–0.76 accuracy should be read with that in mind.\n\nI don't see any fatal flaw in the math or statistics. The nerve-theorem justification for the weighted alpha filtration is sound. The paper deserves peer review, but it should be sent back for the porosity and reproducibility fixes. I wouldn't cite it until those are addressed.","headline":"Solid, honest pilot study of persistent homology as a release-regime descriptor, but the unmeasured porosity confounder must be fixed before the central claim is safe.","tokens_in":640,"tokens_out":1491,"would_cite":false,"duration_ms":34285,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["55N31","65M60"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper argues that persistent homology summaries of the solid phase in porous media—counts and lifetimes of loops, cavities, and connected components—are strongly associated with diffusive release behavior even after stratifying by targ","keywords":["persistent homology","porous media","diffusive release","topological data analysis","release-curve classification","finite element method","porosity control","drug release screening"],"falsifier":"Measure the true pore volume of each generated sample (e.g., by voxel counting), then re-run the Kruskal–Wallis and classification analyses with actual porosity as a covariate or stratifier; if the PH–release associations vanish, the central claim is refuted.","tokens_in":11653,"feed_emoji":"🔬","tokens_out":6811,"duration_ms":58839,"temperature":0.7,"pith_summary":"The paper sets out to show that the shape and multiscale organization of the solid phase in a porous medium—not just how much pore space there is—drives how quickly a dissolved tracer escapes. By computing persistent homology of the solid-offset filtration and solving a finite element diffusion problem on synthetic sphere packings, the author finds that within fixed target porosity levels, samples with richer topological features (more and longer-lived loops and cavities) tend to exhibit delayed, long-tailed release. Eighteen persistent-homology summary features classify three release regimes—early-fast, late-release, long-tail—with test accuracy between 64% and 76%, far above the 33% chance level, and the features are computed tens to hundreds of times faster than the diffusion simulation. A sympathetic reader would take this as evidence that persistent homology is a viable lightweight, interpretable screening descriptor for porous structures in applications like drug delivery and contaminant transport.","feed_headline":"Pore topology predicts diffusive release beyond porosity","feed_subtitle":"PH signatures classify early, late, and long-tail release regimes at 64–76% accuracy, far faster than simulations.","key_machinery":"The load-bearing object is the solid-offset filtration: the solid phase (union of spheres) is thickened by a distance r, and persistent homology records the birth and death of connected components (H0), loops (H1), and void-like regions (H2) as r grows; the weighted alpha complex computes the same diagrams efficiently. From each diagram, six scalar summaries (count, finite count, persistence sum, mean persistence, maximum persistence, mean birth) yield 18 features that carry the whole argument. The release curve Q(t) is generated independently by a conforming P1 finite element solve of the mixed Dirichlet–Neumann diffusion equation on a voxel-derived tetrahedral mesh, and the three regimes (","core_discovery":"On its own terms, the paper's central discovery is that the multiscale topology of the solid phase carries release-behavior information that total porosity does not. In porosity-controlled comparisons (40%, 60%, 80% target porosity), Kruskal–Wallis tests show that persistence-sum and count summaries for H0, H1, and H2 differ significantly across release regimes, with the largest effect sizes for H0/H1 persistence sum (η²_H ≈ 0.27–0.54). Group means show a consistent gradient: early-fast samples have the smallest persistence counts and sums, long-tail samples the largest, with late-release in between. A multinomial logistic regression using only the 18 PH summaries achieves 0.641–0.759 test a","pith_inferences":["If the association holds on experimentally measured microstructures, then porosity alone—the standard design metric in many drug-delivery papers—is an incomplete predictor, and PH summaries could be reported alongside porosity as a standard descriptor.","The implied mechanism (loops and cavities acting as topological traps that slow escape) could be tested by comparing PH features against pore-network tortuosity or effective diffusivity on the same geometries.","A direct extension would be to predict the entire release curve Q(t) from PH features using a regression model, and to check whether adding PH features to porosity reduces prediction error compared with porosity alone.","The self-referential structure (regimes defined from the same simulation and labels fitted on them) means the 64–76% accuracy partly reflects internal consistency; the true test is external validation on experimental release data."],"forward_implications":["At any fixed porosity, two microstructures can have markedly different release curves, so porosity-based screening alone can mis-rank candidates.","Persistent-homology summaries can flag long-tail or slow-release candidates early in a design loop, reserving expensive FEM simulations for a shortlist.","The direction of association gives a design heuristic: to achieve fast release, keep the solid phase topologically simple; to delay release, introduce multi-scale loop/cavity structures.","Combining PH features with release-curve regression (not just regime labels) may yield fast surrogate models for Q(t) over large candidate libraries.","The classification accuracy (0.641–0.759) provides a baseline; adding further geometry descriptors or using the full persistence diagrams (not 18 summaries) may push accuracy higher."],"fun_headline_variants":["Pore topology, not just porosity, governs release curve","Beyond porosity: pore topology predicts diffusion release","Multiscale pore structure forecasts long-tail release","Topological descriptors classify release regimes quickly","Pore shape analysis predicts diffusive release faster than simulation"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The porosity-controlled result compares samples that only approximately match a prescribed target porosity, and true porosity is never measured or reported; if actual porosity varies within a target stratum and correlates with topology, the claimed 'beyond porosity' effect could be a hidden porosity effect.","fun_headline_variants_meta":{"raw":{"variants":["Pore topology, not just porosity, governs release curve","Beyond porosity: pore topology predicts diffusion release","Multiscale pore structure forecasts long-tail release","Topological descriptors classify release regimes quickly","Pore shape analysis predicts diffusive release faster than simulation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000168,"raw_usage":{"total_tokens":1069,"prompt_tokens":685,"completion_tokens":384,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":429,"completion_tokens_details":{"reasoning_tokens":312}},"tokens_in":429,"tokens_out":384,"duration_ms":4222,"temperature":1.0,"reasoning_tokens":312,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T08:06:13.210489+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the true pore volume of each generated sample (e.g., by voxel counting), then re-run the Kruskal–Wallis and classification analyses with actual porosity as a covariate or stratifier; if the PH–release associations vanish, the central claim is refuted.","supporting_citations":[],"review_version":3}