{"id":"e0e60c1d-bbeb-4acb-9143-c8eb7489e347","arxiv_id":"1908.07208","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"In the CrowdLimits experiment, heterogeneous crowds in bidirectional and intersecting flows sustained flow at densities above 6 people per square meter, and the maximum global flow decreased as movement complexity increased.","lead":"Scientists ran a large laboratory experiment in which 130 to 140 diverse pedestrians walked through corridors and right-angle crossings at very high densities, and measured the maximum number of people who could pass per second. The results suggest that more complex movement patterns lower the maximum sustainable flow, which matters for designing safe stations, stadiums, and event spaces.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The global qmax values may be transient loading peaks rather than sustainable capacities, because inflow is stepped every minute and the paper reports no stationarity check.","rationale":"The reader's weakest assumption identifies the stationarity of each one-minute inflow interval as the load-bearing condition for interpreting the measured density-flow pairs as fundamental-diagram points. My independent reading reaches the same conclusion: the paper's central claim, that maximum sustainable flow decreases with complexity, depends on the four global qmax values in Table 4 being equilibrium capacities. The experimental protocol steps inflow every minute and gives no evidence of equilibration; the text even states that some high-inflow runs were halted early. This is not a manufactured concern—it is the difference between reporting capacity and reporting a transient response. I considered other potential objections, such as the two repetitions per scenario and the mismatch between the stated 99th percentile and the table's 95th percentile label. Those are real but secondary; the stationarity issue alone is enough to make the headline claim conditional rather than fully established. The paper is honest about being preliminary, and the dataset appears genuine and carefully extracted, so I do not recommend rejection. The appropriate verdict remains CONDITIONAL: the conclusion is plausible but requires a stationarity analysis and replicate-level statistics before the quantitative values can be accepted.","tokens_in":12164,"tokens_out":3585,"duration_ms":40807,"concrete_test":"Recompute the global fundamental diagram using only stationary portions of each one-minute inflow interval. For each run, compute 5-second moving-window estimates of global density and flow, then test for stationarity within each interval by requiring that the density trend be statistically indistinguishable from zero and that the mean of the first half and last half of the interval agree within 10%. Recalculate Table 4 qmax using only intervals and time windows that pass this stationarity test. If the ordering of the four scenarios changes, or if most intervals fail the test, the complexity conclusion is not supported by the current analysis. As a second check, report the time at which qmax occurs relative to the start of the inflow step and to the end of the run, especially for runs stopped early.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central conclusion—that capacity decreases with increasing complexity—rests on the values in Table 4: global qmax of 0.65, 0.86, 0.23, and 0.54 P/m/s for bidirectional 50-50, bidirectional 80-20, intersecting 50-50, and intersecting 80-20. For these numbers to be interpretable as points on a fundamental diagram, each one-minute inflow step must allow the crowd to reach a quasi-stationary state before the next step begins. Section 3.3 explains that inflow was increased every minute, and Section 3.2 notes that at the highest flow rates some runs were stopped early because outflow dropped and queues could not be replenished. The paper does not report any check that density and flow stabilized within each minute. If the relaxation time after an inflow increase is comparable to or longer than one minute, the measured density-flow pairs trace the transient loading path, not the equilibrium fundamental diagram. This concern is especially acute for the intersecting 50-50 scenario, whose global qmax of 0.23 is far below the other scenarios; that low value could reflect an unfinished build-up rather than a genuinely lower sustainable capacity. A secondary imprecision reinforces the worry: Section 5.2 states that maxima are computed as the 99th percentile, while Table 4 is labeled as the 95th percentile. Without knowing which percentile was used, and without error bars or replicate-level results, the ordering of the four maxima is not yet a robust quantitative claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports the CrowdLimits laboratory experiment, in which 130-140 heterogeneous participants performed bidirectional and 90-degree intersecting pedestrian flows at 50-50 and 80-20 directional splits, with inflow stepped upward every minute. From automatically tracked and manually corrected trajectory data, the authors compute local and global Voronoi-based densities, speeds, and flow rates, and summarize the fundamental diagrams through 95th/99th percentile maxima. They report that global maximum flow is highest for bidirectional 80-20 (0.86 P/m/s), lower for bidirectional 50-50 (0.65 P/m/s), lower still for intersecting 80-20 (0.54 P/m/s), and lowest for intersecting 50-50 (0.23 P/m/s), and conclude that the capacity of pedestrian infrastructures decreases with increasing complexity.","tokens_in":12363,"tokens_out":1918,"duration_ms":21267,"significance":"If the directional ordering of these maxima is reliable, the paper would provide rare empirical evidence on how movement-base-case complexity degrades the capacity of pedestrian infrastructure under high densities, extending earlier homogeneous-population studies to a heterogeneous European population. Strengths of the manuscript include the deliberate heterogeneity of the participant pool, the use of the Voronoi method for density and speed, and the manual verification/correction of all trajectories, which gives high-quality trajectory data. The claimed conclusion, however, rests on a small number of runs and on an untested quasi-stationarity assumption, so the quantitative claims should be regarded as provisional rather than established.","major_comments":[{"comment":"The central load-bearing assumption is that each one-minute inflow step allows the crowd to reach a quasi-stationary state, so that density-flow pairs lie on a fundamental diagram rather than on a transient loading path. The paper states in Section 3.3 that inflow was increased every minute and in Section 3.2 that some runs were stopped early when outflow dropped and queues could not be replenished, but no check is reported that density and flow stabilized within each minute. This is especially relevant for the intersecting 50-50 scenario, whose global qmax of 0.23 P/m/s may reflect an unfinished build-up rather than a genuinely lower sustainable capacity. The authors should either provide evidence of stabilization (e.g., time series within each minute, or a comparison of the first and second half of each interval) or reinterpret the reported maxima as transient loading values, which would weaken the capacity-decrease conclusion.","section":"Section 3.3 and Section 6.2"},{"comment":"There is a direct inconsistency between the text and the results table: Section 5.2 states that rho_max, q_max, and v_max are computed as the 99th percentile, while Table 4 is labeled as the 95th percentile. Since the reported global maxima are the primary evidence for the paper's central claim, the authors must state which percentile was actually used and recompute the table consistently. The ordering in Table 4 could change if the two percentile definitions are mixed across scenarios or if the 95th and 99th percentiles behave differently across the four scenarios.","section":"Section 5.2 and Table 4"},{"comment":"Each of the four base scenarios has only two experimental repetitions (e.g., bidirectional 50-50 is run twice on Day 1 and intersecting 50-50 twice on Day 2), yet Section 6 presents the results without error bars, replicate-level ranges, or significance tests. With n=2 per scenario and no measure of run-to-run variability, the quantitative ordering in Table 4 (0.65, 0.86, 0.23, 0.54) is not robust. The paper itself call the results preliminary in Section 7. The authors should report per-run maxima and, at minimum, bootstrap or resampling intervals, or explicitly downgrade the conclusion to a directional trend pending a larger sample.","section":"Table 3 and Section 6"}],"minor_comments":[{"comment":"The captions of Figures 5 and 6 are identical, but Figure 6 shows global fundamental diagrams while Figure 5 shows local ones; the caption for Figure 6 should say 'global' instead of 'local'.","section":"Figure 5 and Figure 6 captions"},{"comment":"The text contains an empty table reference (\"TABLE \") after the sentence about increasing flow rates every minute; the intended table appears to be Table 2, and the cross-reference should be added.","section":"Section 3.2"},{"comment":"The Munkres reference in the reference list is dated 1997, but the text cites it as 1957; please verify the correct publication year for the assignment problem algorithm.","section":"References"},{"comment":"There are several typographical issues, including 'increasi ng' in the abstract, 'emprical' in the Helbing et al. reference, and inconsistent spacing around table numbers; a careful language edit would improve readability.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper presents a valuable dataset and a plausible directional claim, but the statistical support is thin: two runs per scenario, no error quantification, and an untested stationarity assumption. The percentile mismatch in Section 5.2 vs Table 4 is a concrete, checkable inconsistency that the authors must resolve. I recommend major revision rather than rejection because the issues are addressable with additional analysis of the existing trajectory data."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nShort version: the CrowdLimits data are real and the experiment is the first to combine a heterogeneous European sample, densities above 6 P/m2, and bidirectional and intersecting base cases in one controlled setup. That part is worth your time. The paper's headline claim, that capacity decreases with complexity, is directionally consistent with Table 4, but the support is thinner than the abstract suggests.\n\nWhat's genuinely good: the trajectory data appear carefully produced, with manual correction to 100% of tracks, and the authors are appropriately circumspect in places, calling the results 'preliminary' and noting that the local fundamental diagrams do not show the classic mountain shape. The finding that flow continues at very high densities in a Western population is a useful replication of Helbing's Hajj results and goes beyond the Asian student samples in Lian and Cao.\n\nWhere the soft spots are: first, no statistical support for the ordering of the four qmax values. Two repetitions per scenario, no error bars, no significance test, and a percentile mismatch—Section 5.2 says 99th, Table 4 is labeled 95th. That alone prevents a quantitative reading. Second, and more load-bearing, the stationarity assumption. Inflow is stepped every minute, and the paper never checks that density and flow stabilize within that interval. If relaxation takes longer than a minute, the qmax values are transient loading peaks, not sustainable capacities. That concern bites hardest on the intersecting 50-50 scenario, whose 0.23 P/m/s could be an unfinished build-up rather than a genuinely low capacity. The paper's own description of runs stopped early because outflow dropped and queues could not be replenished adds to the worry.\n\nI don't think the central argument collapses—the pattern across all four scenarios is consistent with complexity increasing friction—but the evidence as presented does not yet support the strong conclusion in Section 7. The fix is straightforward: report per-replicate values, add a stationarity diagnostic (e.g., density and flow as a function of time within each minute), and settle the percentile question.\n\nWho should read this: anyone working on pedestrian fundamental diagrams or crowd safety will want to know this dataset exists. It deserves a serious referee, not a desk reject, but the referee should be instructed to ask for the missing checks before publication.\n\nBest,\n[Name]","headline":"A genuinely new high-density pedestrian experiment with a plausible central pattern, but the capacity-decreases-with-complexity claim needs a stationarity check and error bars before the numbers can be taken quantitatively.","tokens_in":12958,"tokens_out":2672,"would_cite":true,"duration_ms":27153,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Pedestrian capacity drops as movement complexity rises","keywords":["pedestrian movement dynamics","fundamental diagram","laboratory experiment","heterogeneous crowd","capacity","flow ratio","bidirectional flow","intersecting flow"],"falsifier":"A replication that holds each inflow level constant for several minutes and checks whether density and flow stabilise within the first minute would settle whether the capacity differences are steady-state phenomena; if the maxima shift with the step duration, the complexity conclusion would need revision.","tokens_in":11903,"feed_emoji":"🚶","tokens_out":8602,"duration_ms":68823,"temperature":0.7,"pith_summary":"This paper tries to establish that the capacity of a pedestrian space—the maximum sustainable flow through a cross-section—shrinks as the movement task becomes more complex. Using a laboratory crowd of 130 to 140 heterogeneous pedestrians, the authors measured global maximum flow rates of 0.65, 0.86, 0.23, and 0.54 people per metre per second for a bidirectional 50-50 corridor, a bidirectional 80-20 corridor, an intersecting 50-50 crossing, and an intersecting 80-20 crossing. They conclude that capacity decreases when flows must cross at right angles and when counterflows are more balanced, because both conditions reduce the chances to avoid collisions. The result matters because most existing capacity values come from homogeneous, unidirectional crowds at lower densities, whereas real crowds are heterogeneous, dense, and often move in crossing patterns.","feed_headline":"Pedestrian capacity drops as movement complexity rises","feed_subtitle":"Lab crowds of 130-plus show intersecting and evenly split flows sustain far lower maximum flow rates than simple corridors.","key_machinery":"The central object is the fundamental diagram built from individual pedestrian trajectories captured by ceiling cameras. Density and speed are computed with the Voronoi method: each pedestrian's local density is the inverse of the area of their Voronoi cell, and global density and speed are averages over the measurement area. The experiment manipulates two factors—movement base case (bidirectional corridor vs. 90-degree intersection) and flow ratio (50-50 vs. 80-20)—while raising the inflow every minute. The load-bearing comparison is the maximum of the global flow rate, taken as the 99th percentile, across the four scenarios; that maximum is the paper's operational definition of capacity.","core_discovery":"The central discovery is that the fundamental diagram, the density-flow relation for pedestrian movement, is not one universal curve: its shape and its peak depend on the movement base case and the flow ratio. The maximum sustainable global flow rate—the 99th percentile of the global flow—was highest in the bidirectional 80-20 corridor (0.86 P/m/s) and lowest in the intersecting 50-50 crossing (0.23 P/m/s), with bidirectional 50-50 at 0.65 and intersecting 80-20 at 0.54. This ordering supports the paper's hypothesis that capacity is governed by the scope for collision avoidance: the more demanding the geometry (crossing vs. opposing streams) and the more evenly the flows are split, the more avoidance manoeuvres are needed and the less throughput the space sustains. The paper also reports that local fundamental diagrams lack the classic single-peaked shape, while global diagrams show it, and that flow continues even at densities around 7 people per square metre, reproducing in a controlled European heterogeneous crowd a phenomenon previously seen only in field observations.","pith_inferences":["The four capacity points suggest a rough 'complexity penalty' could be quantified—an inference the authors do not draw—allowing planners to discount corridor capacities when crossings or balanced counterflows are introduced.","Because inflow was increased every minute, the measured maxima may mix transient and quasi-stationary states; if longer dwell times shift the maxima, the capacity ordering could change, so this is a natural robustness check.","The absence of instructions in these four scenarios means the crowd was relatively 'normal'; the assignment conditions (crossing, fast walking) that the authors plan to analyse would test whether goal-orientation and unpredictability amplify the complexity effect.","The local-versus-global discrepancy hints that pedestrians may temporarily move faster through dense patches to maintain global throughput; a microscopic model reproducing this behaviour would explain why local scatter is large while global flow remains smooth."],"forward_implications":["Facility capacity guidelines that quote a single number for corridors and crossings will overestimate throughput for intersecting layouts and balanced counterflows.","Since local fundamental diagrams lack a clean peak, capacity estimates should be based on global, area-averaged quantities rather than on single-pedestrian measurements.","Very high densities (above 6 P/m²) do not force a full stop in forward movement over short periods, so crowd-safety and evacuation models should not assume a hard jam density.","The density at which capacity is reached differs by scenario, so demand-capacity checks should use scenario-specific fundamental diagrams rather than a universal curve."],"supporting_citations":[{"why":"Provides the empirical basis that 50-50 and 80-20 flow ratios are respectively the most and least advantageous for bidirectional capacity, motivating the chosen scenarios.","marker":"Kretz et al. (2006)"},{"why":"The prior comparison of unidirectional, bidirectional, and intersecting flows, whose conclusion of small capacity differences this study directly opposes.","marker":"Cao et al. (2017)"},{"why":"Field observations at the Hajj that first showed flow continuing at densities above 6 P/m², the phenomenon this experiment reproduces under laboratory conditions.","marker":"Helbing et al. (2007)"},{"why":"Laboratory recreation of high-density four-directional intersecting flow with a homogeneous Asian student population, which this study extends to a heterogeneous European population.","marker":"Lian et al. (2015)"},{"why":"Supplies the Voronoi-based definitions of density and speed used to construct the fundamental diagrams.","marker":"Steffen & Seyfried (2009)"},{"why":"Used together with a pilot run to set the minimum inflow rate that would still guarantee continuous flow in each scenario.","marker":"Zhang et al. (2014)"}],"fun_headline_variants":["Pedestrian capacity falls as flow complexity grows","Intersecting crowds cut max pedestrian flow by up to 73%","Complex movement shapes capacity of pedestrian spaces","Lab users show high-density flow survives complex crossings"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The experiment steps the inflow up every minute and treats the resulting density-flow pairs as points on a fundamental diagram, which assumes the crowd reaches a quasi-stationary state within each minute; if one minute is too short, the measured maxima capture transients rather than sustainable capacity.","fun_headline_variants_meta":{"raw":{"variants":["Pedestrian capacity falls as flow complexity grows","Intersecting crowds cut max pedestrian flow by up to 73%","Complex movement shapes capacity of pedestrian spaces","Lab users show high-density flow survives complex crossings"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000353,"raw_usage":{"total_tokens":1966,"prompt_tokens":1034,"completion_tokens":932,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":650,"completion_tokens_details":{"reasoning_tokens":872}},"tokens_in":650,"tokens_out":932,"duration_ms":10291,"temperature":1.0,"reasoning_tokens":872,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:22:20.478037+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A replication that holds each inflow level constant for several minutes and checks whether density and flow stabilise within the first minute would settle whether the capacity differences are steady-state phenomena; if the maxima shift with the step duration, the complexity conclusion would need revision.","supporting_citations":[],"review_version":1}