{"id":"573fbc7e-953b-4173-9d36-8639ef311e5f","arxiv_id":"1908.05846","paper_version":4,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"Pedestrian-carried BLE receivers can detect e-scooter encounters and reveal campus hotspots where pedestrians and e-scooters interact most, though the link to safety is assumed rather than measured.","lead":"Researchers used smartwatches carried by pedestrians to detect Bluetooth signals from rental e-scooters, logging 1,800 algorithm-detected and thousands of user-reported encounters on two university campuses over three months. The result is a crowd-sensing template for mapping where and when pedestrians and e-scooters come into close contact, aimed at informing micromobility safety policy.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The safety conclusion rests on the Section 3 postulate that encounter frequency/density proxies collision risk, but this proxy is never validated against any outcome measure; the paper is a useful exposure study, not a validated safety measurement.","rationale":"I read the paper in good faith as a first-of-its-kind descriptive field study. The method—passively capturing e-scooter BLE beacons from pedestrian-worn smartwatches to detect proximate encounters—is plausible and the data collection is nontrivial: 77 participants over three months, on two campuses, with both predicted (EP) and observed (EO) encounters, plus heart-rate and survey data. The paper is transparent about several limitations, including the university setting, the exclusion of privately owned e-scooters, and sampling bias. The concern I identify is the same one the Reader flagged: the Section 3 postulate that encounter density/frequency is a good proxy for pedestrian safety risk. This is genuinely load-bearing because all the safety conclusions, hotspot identifications, and planning implications are built on it. There is no external outcome variable—collisions, injuries, near-misses, or even validated disruption reports—against which the encounter metric is checked. I do not think this is a fatal flaw, because the paper explicitly calls it a postulate and positions itself as a preliminary blueprint, but it does mean the central safety claim is conditional rather than established. I considered other potential concerns, such as the heuristic thresholds in the encounter-detection algorithm (Section 4.4.1: 1-second window, >=4 BLE packets, 300-second merging) and the comparison of BLE signal strength across providers with different transmit powers. These are real but secondary; they affect the precision of the encounter statistics, not the validity of the safety interpretation. Because the Reader's verdict is already CONDITIONAL and my analysis does not move it, I recommend no change.","tokens_in":20690,"tokens_out":2414,"duration_ms":27439,"concrete_test":"Run a validation sub-study on the same two campuses: during a comparable one-month period, have a separate sample of pedestrians (or the same participants) log ground-truth safety-relevant incidents—near-misses, forced evasive actions, sidewalk blockages, falls, or any collision with an e-scooter—using a simple time-stamped and GPS-tagged report. Then compare the spatial and temporal distribution of these reported incidents against the paper's EP/EO hotspot maps, using a simple relative-risk or correlation analysis (e.g., incident rate in high-encounter segments/time slots versus low-encounter ones, with confidence intervals). If the high-encounter zones do not show statistically elevated incident rates, the Section 3 postulate is unsupported and the paper should be read as an exposure study, not a safety study.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim of the paper is that crowd-sensed BLE encounters can be used to benchmark pedestrian safety. The load-bearing step is the Section 3 postulate: 'a higher density/concentration or frequency (or both) of such close vehicle-pedestrian encounters is indicative of a higher probability or potential for vehicle-pedestrian collisions.' Everything that follows in RA1-RA3 and the safety-oriented conclusions (hotspots, 'potentially unsafe zones,' infrastructure recommendations) depends on this postulate. The paper provides no external validation connecting encounter counts, densities, or BLE-derived closeness to actual collisions, injuries, near-misses, or even pedestrian-reported disruptions. The descriptive encounter statistics and hotspot maps may be accurate, but without a demonstrated link between encounter exposure and harm, the safety interpretation does not follow. The elevated-heart-rate observation in Section 4.4.2 is suggestive but is not a validated outcome measure: there is no baseline comparison, no control for other causes of heart-rate elevation, and no evidence that the 60% figure corresponds to danger rather than ordinary attention. The paper even labels the key step as a postulate, and Section 6.1 acknowledges scope limitations, but it does not test the proxy against any outcome data. Thus the weakest assumption is not an internal inconsistency but an unvalidated external link: encounter frequency is adopted as a safety metric because it is measurable, not because it has been shown to predict harm. If this proxy is wrong or is only a measure of exposure, then the policy recommendations—building bike lanes, changing shuttle schedules, redesigning intersections—are aimed at exposure rather than at demonstrated risk.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a three-month field study in which 77 pedestrian participants carried custom-instrumented smartwatches that passively captured Bluetooth Low Energy (BLE) advertising packets emitted by rental e-scooters (Bird and Lime) on two university campuses. From these data the authors define two encounter types: predicted encounters (EP, derived from BLE packet streams via a sliding-window detector) and observed encounters (EO, participant-confirmed encounters with directional information). They then compute spatio-temporal encounter statistics, analyze BLE signal strength as a proxy for encounter closeness, and correlate encounters with functional road classifications, class schedules, and participant heart rate. The central safety claim is that zones and times with high encounter density or frequency are 'potentially unsafe' for pedestrians, a conclusion built on the explicit postulate in Section 3 that close e-scooter–pedestrian encounters are indicative of collision probability.","tokens_in":20980,"tokens_out":3654,"duration_ms":38636,"significance":"If the encounter-to-risk link were validated, the paper would make a novel and practical contribution: it demonstrates that pedestrian-carried commodity BLE receivers can identify nearby rental e-scooters and that crowd-sensed encounter data can produce useful exposure maps. The deployment is nontrivial, the encounter-detection pipeline is clearly described, and the paper is transparent in labeling its central risk proxy as a postulate rather than hiding it. However, the current manuscript is better characterized as a study of pedestrian exposure to e-scooters than as a validated safety measurement. Its headline conclusions—'potentially unsafe zones,' 'collisions are more likely,' and specific infrastructure recommendations—follow directly from the unvalidated postulate and are not independently supported by outcome data. The descriptive and methodological contributions are worth publishing, but the safety interpretation needs either external validation or systematic reframing.","major_comments":[{"comment":"The core safety claim relies entirely on the Section 3 postulate that higher encounter density/frequency indicates higher collision probability or potential. The paper provides no external validation of this proxy against actual collisions, injuries, near-misses, or even pedestrian-reported disruptions. Consequently, statements such as 'this indirectly suggests that collisions are more likely to occur in high-encounter atomic segments' (Section 5.1) and 'e-scooter related pedestrian collisions are more likely to occur in spatio-temporal zones with high encounter counts' (Section 5.3) are restatements of the postulate, not empirical findings. To make the safety claim load-bearing, the authors should either validate the proxy against an outcome measure (e.g., hospital or police records, participant-reported close calls near identified hotspots) or systematically revise the manuscript to frame the results as encounter-exposure findings and use safety language only where outcomes were actually measured.","section":"Section 3, postulate; Sections 5.1–5.3 conclusions"},{"comment":"The comparisons of BLE signal strength between low- and high-encounter segments/time periods/spatio-temporal zones are presented as supporting the collision-risk conclusion, but no confidence intervals or significance tests are reported. Because per-encounter RSSI is noisy and the plotted distributions appear strongly overlapping, it is not clear whether the star-marked mean differences are statistically reliable. The authors should report effect sizes with confidence intervals or bootstrap tests, and also report the number of encounters contributing to each comparison. Without this, the closeness-based claims are not quantitatively supported.","section":"Section 5.1, Figure 9; Section 5.2; Section 5.3, Figure 9"},{"comment":"The elevated-heart-rate analysis used to support the 'startled pedestrian' narrative is not a validated outcome measure. There is no baseline or control condition (e.g., heart rate during similar walking without e-scooter encounters), no adjustment for physical activity, stair climbing, or other causes of heart-rate elevation, and no description of how the personalized thresholds were computed beyond 'most frequently occurring pulse rate(s).' The 60% figure is reported without a denominator or a statistical comparison. At best this analysis is exploratory; the authors should either add a proper control analysis or explicitly downgrade this finding to a hypothesis-generating observation.","section":"Section 4.4.2, heart-rate analysis"},{"comment":"The designation of 'high-encounter' atomic segments (EP > 25 and EO > 5) and the corresponding spatio-temporal zone thresholds appear arbitrary, and no sensitivity analysis is provided. Since the paper's practical recommendations target these hotspots, the authors should test whether the spatial and spatio-temporal conclusions are robust to reasonable changes in the thresholds (e.g., varying EP and EO cutoffs, or using per-mile normalized counts as motivated by Table 3). Without this, the hotspot list may reflect threshold choices rather than stable features of the encounter distribution.","section":"Sections 5.1 and 5.3, hotspot thresholds"}],"minor_comments":[{"comment":"The encounter-detection parameters (1-second sliding window with 80% overlap, N ≥ 4 packets, 300-second merge interval, maximum four encounters per scooter per participant per day) are described as 'empirically determined,' but the calibration procedure and any validation of these thresholds against independently labeled encounters are not reported. A brief description of the calibration data and its accuracy would materially improve reproducibility.","section":"Section 4.4.1"},{"comment":"The sentence '105 participants who participated for at least 15 days (on average) for the 30-day study' is unclear: it does not specify whether the intended participation window was 30 days or three months, and how the 15-day average was computed. Please clarify the protocol.","section":"Section 4.2, Participants"},{"comment":"The visual display of BLE signal strength comparisons would be clearer with box plots or error bars; the star-mean alone does not convey the variance or sample size, which is especially important given the significance-testing concern in the major comments.","section":"Figure 9"},{"comment":"The limitation about privately owned e-scooters not emitting BLE packets is relevant, but the authors should also note that the study is restricted to Bird and Lime vehicles and that other providers' beacon intervals or identifiers may differ, limiting generalizability to other fleets.","section":"Section 6.1, Limitations"},{"comment":"The survey question numbering skips item 5; please renumber or correct the omission.","section":"Appendix A"},{"comment":"The definition of Total Encounters per Segment (TES) says 'sum of all detected... encounters in a network segment,' but the column header TES is repeated for each functional class; clarify whether TES is per-segment or aggregated over all segments in that class, and state clearly how MEM is computed (it appears the MEM values are much larger than TES for some rows, so the units are unclear).","section":"Table 3"},{"comment":"The statement 'at least twenty atomic segments in both campuses have a relatively high number of encounters: EP > 25 and EO > 5' should state the exact number of high-encounter segments and how the threshold was chosen, particularly because the threshold is later used for safety-relevant claims.","section":"Section 5.1"}],"recommendation":"major_revision","confidential_remarks":"The paper's core methodological contribution—pedestrian-side BLE detection of e-scooters—is sound and likely of interest to the ubiquitous-computing and urban-sensing communities. The referee's central concern is that the safety conclusions are entailed by an unvalidated postulate. I would be comfortable with a major revision that either adds an outcome-based validation (even a small one, such as correlating hotspots with known incident reports) or systematically rephrases the claims in terms of encounter exposure, reserving safety language for the heart-rate and survey results with appropriate hedges. A revision that merely adds a limitations paragraph without changing the conclusions would not be sufficient."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth your time if you work on micromobility or urban sensing. The real contribution is the pedestrian-perspective BLE encounter detector, and that part is believable; the safety conclusions are where the paper outruns its data.\n\nWhat is new: rental e-scooters continuously broadcast BLE beacons, and the authors show you can catch those on a pedestrian's smartwatch, filter them into encounters with a sliding-window detector, and map hot spots from the pedestrian side. Prior work used surveys, hospital data, and provider APIs; nobody had done field sensing of this kind. The three-month deployment on two campuses with 77 participants is real work, and the descriptive findings—encounters clustering hard (95% of segments with five or fewer), high-encounter zones with measurably closer BLE proximity, midday peaks tracking class schedules—are plausible and useful for planners.\n\nThe paper is also candid. Section 3 explicitly labels the load-bearing assumption a 'postulate': that encounter density and frequency indicate collision probability. Section 6.1 admits the scope, sampling, and data-sharing limits. That honesty helps you see the weakness clearly.\n\nThe weakness is real, though. The postulate is never tested against collisions, injuries, near-misses, or pedestrian-reported disruption, so calling hot spots 'potentially unsafe' is partly true by definition. The heart-rate observation (elevated in ~60% of one-foot encounters) has no baseline and no control for other causes; treat it as suggestive. The encounter comparisons carry no confidence intervals or significance tests, the detection thresholds are heuristic with no sensitivity analysis, and there is no code or data to re-check. None of this kills the method; it just means the paper is an exposure study, not a validated safety measurement.\n\nAudience: transportation and urban-informatics researchers who want a cheap encounter-sensing blueprint, and anybody teaching proxy validation. Recommendation: send it to peer review. It deserves referee time, and the right reviewers will push the authors to either validate the proxy or soften the claims.","headline":"The pedestrian-perspective BLE encounter-sensing method is genuinely new and plausible; the safety claims outrun the data because the risk proxy is a stated postulate, never validated.","tokens_in":21569,"tokens_out":3719,"would_cite":true,"duration_ms":35612,"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-carried Bluetooth receivers can identify nearby rental e-scooters from their beacon signatures and use the density of close encounters as a benchmark of pedestrian safety.","keywords":["micromobility","pedestrian safety","electric scooters","Bluetooth Low Energy","crowd-sensing","wearables","encounter detection","spatio-temporal analysis"],"falsifier":"Install fixed video cameras at a sample of low-, medium-, and high-encounter segments and independently code actual collisions, near-misses, sudden swerves, and pedestrian disruptions over the same weeks the encounter counters run; if the encounter-density ranking does not match the coded incident ranking, the safety proxy fails even though the encounter detection itself may be accurate.","tokens_in":20500,"feed_emoji":"🛴","tokens_out":14950,"duration_ms":135375,"temperature":0.7,"pith_summary":"This paper tries to establish that pedestrians can serve as living sensors for e-scooter safety. Volunteers wore smartwatches during their normal campus routine for a 30-day window; the watches passively recorded Bluetooth Low Energy beacons that rental e-scooters emit continuously, and the authors turned those signals into detected encounters between a pedestrian and a scooter. From three months of such crowd-sensed data on two university campuses, they show that encounters are strongly concentrated in space and time—near dormitories, shuttle stops, and building entrances, around midday on weekdays, and on sidewalks rather than shared paths—and that close encounters are even more common in high-encounter zones. The paper's stated safety postulate is that higher encounter frequency and density indicate higher collision potential, so these maps can benchmark pedestrian safety. If the method and the postulate hold, urban planners gain a low-cost, pedestrian-side way to monitor conflicts with micromobility vehicles without access to scooter-provider data.","feed_headline":"Pedestrians' smartwatches can map e-scooter encounter hotspots","feed_subtitle":"Three months of Bluetooth signals reveal where and when e-scooters crowd walkers—no scooter-company data needed.","key_machinery":"The load-bearing object is the BLE advertising packet that commercial e-scooters emit continuously for app connectivity and unlocking, and the load-bearing definition is the predicted encounter: a sequence of at least four such packets from one scooter within a one-second sliding window, with gaps under 300 seconds merged into the same encounter and more than four encounters with the same scooter per day discarded to prevent stationary parking bias. Packet regularity separates close scooters from distant ones, since reception becomes inconsistent beyond roughly 20-25 feet, and received signal strength acts as a distance proxy calibrated against known provider models. Participant answers to prompted yes/no questions add ground truth on whether the scooter is moving, whether it is ahead or behind, and its direction, while heart-rate readings are used to detect startle responses.","core_discovery":"The paper's central claim is that a pedestrian's smartwatch can, by passive listening alone, recover a usable record of when and where rental e-scooters come close. Because commercial e-scooters broadcast BLE advertising packets at regular intervals, and because those packets carry provider-distinguishable identifiers, the authors define an encounter operationally: at least four packets from the same scooter in a one-second window, with nearby windows merged if less than five minutes apart, and repeated stationary proximity to the same scooter capped at four encounters per day. Applied to the campus dataset, this operation produced roughly 1,800 predicted encounters alongside participants' real-time reports of moving scooters, which add direction of approach and line of sight. The resulting spatial and temporal distributions are extremely uneven: a small number of road segments and time slots carry most of the encounters, and those high-encounter segments and periods show stronger received BLE signal strength, meaning scooters were physically closer on average. The authors take this convergence as evidence that the crowd-sensed encounter metric captures conditions under which collisions or disruptions are most probable, and they present the full pipeline—passive BLE capture, encounter detection, heart-rate confirmation, and hotspot mapping—as a blueprint for pedestrian-centered micromobility safety studies.","pith_inferences":["Beyond the paper, the same passive-listening pipeline could be applied to other Bluetooth-emitting vehicles and personal devices, making encounter mapping a general urban-sensing primitive rather than an e-scooter-specific tool.","Beyond the paper, separating parked-scooter obstructions from moving-scooter pass-bys would likely give planners very different safety signals, since the study's encounter counts mix the two.","Beyond the paper, a calibration study overlaying these BLE encounters with video-coded near-misses or injury records could convert the qualitative 'potentially unsafe' label into a quantitative risk threshold, directly testing the paper's weakest assumption.","Beyond the paper, city-scale aggregation of such pedestrian-sensed encounters could produce a dynamic sidewalk-conflict layer that complements official crash statistics, which are known to undercount pedestrian and bicycle incidents."],"forward_implications":["Planners could use the encounter maps directly to target fixes—separated bike lanes, sidewalk clearances, or re-sited scooter parking—at the specific segments where encounters cluster.","Because data collection is pedestrian-side and passive, cities could monitor scooter-pedestrian conflict without requesting data from scooter operators, who may have incentives not to emphasize pedestrian impacts.","The alignment of encounters with class schedules implies hotspots are partly predictable, so shuttle frequency and scooter deployment could be adjusted before peak conflict windows.","The finding that close encounters are more common on sidewalks and local streets than on shared-use paths supports separating micromobility traffic from walking space.","Elevated heart rates during close, fast-moving encounters, together with survey interest from 58% of participants, point toward a feasible real-time smartwatch alert application."],"supporting_citations":[{"why":"Supplies the motivating case of a pedestrian injured by an e-scooter, anchoring the claim that close encounters matter for safety.","marker":"[12]"},{"why":"Documents illegal and risky e-scooter riding, supporting the need for pedestrian-side monitoring.","marker":"[13]"},{"why":"Represents the city-level e-scooter findings report whose qualitative, operator-side approach the study positions itself against.","marker":"[16]"},{"why":"Provides the built-environment risk-factor framework used to interpret why encounter locations matter for pedestrian safety.","marker":"[23]"},{"why":"Shows that pedestrian and bicycle crashes are underreported, justifying encounter crowd-sensing as a complementary safety data source.","marker":"[28]"},{"why":"Is the prior observational study of e-scooter parking and rider/non-rider perceptions that the paper extends with continuous encounter sensing.","marker":"[41]"}],"fun_headline_variants":["Smartwatches turn pedestrians into e-scooter proximity sensors","E-scooter hotspots emerge from smartwatch BLE pings","Pedestrian wearables map e-scooter encounters without scooter data","BLE from e-scooters exposes crowded walkways via smartwatches","Crowd-sensed smartwatches reveal e-scooter encounter hot spots"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper's safety conclusions rest on the assumption that encounter frequency and density are a valid proxy for collision risk: no counted encounter is ever linked to an actual collision, injury, or observed near-miss.","fun_headline_variants_meta":{"raw":{"variants":["Smartwatches turn pedestrians into e-scooter proximity sensors","E-scooter hotspots emerge from smartwatch BLE pings","Pedestrian wearables map e-scooter encounters without scooter data","BLE from e-scooters exposes crowded walkways via smartwatches","Crowd-sensed smartwatches reveal e-scooter encounter hot spots"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000279,"raw_usage":{"total_tokens":1706,"prompt_tokens":1041,"completion_tokens":665,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":657,"completion_tokens_details":{"reasoning_tokens":570}},"tokens_in":657,"tokens_out":665,"duration_ms":6326,"temperature":1.0,"reasoning_tokens":570,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:03:18.048917+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Install fixed video cameras at a sample of low-, medium-, and high-encounter segments and independently code actual collisions, near-misses, sudden swerves, and pedestrian disruptions over the same weeks the encounter counters run; if the encounter-density ranking does not match the coded incident ranking, the safety proxy fails even though the encounter detection itself may be accurate.","supporting_citations":[{"cited_title":"Sharing the sidewalk: a case of e-scooter related pedestrian injury,","cited_arxiv_id":null,"evidence_quote":"Supplies the motivating case of a pedestrian injured by an e-scooter, anchoring the claim that close encounters matter for safety."},{"cited_title":"Illegal and risky riding of electric scooters in brisbane,","cited_arxiv_id":null,"evidence_quote":"Documents illegal and risky e-scooter riding, supporting the need for pedestrian-side monitoring."},{"cited_title":"2018 E-Scooter Findings Report,","cited_arxiv_id":null,"evidence_quote":"Represents the city-level e-scooter findings report whose qualitative, operator-side approach the study positions itself against."},{"cited_title":"Pedestrian Safety and the Built Environment: A Review of the Risk Factors,","cited_arxiv_id":null,"evidence_quote":"Provides the built-environment risk-factor framework used to interpret why encounter locations matter for pedestrian safety."},{"cited_title":"Investigating the underreporting of pedestrian and bicycle crashes in and around university campuses - a crowdsourcing approach,","cited_arxiv_id":null,"evidence_quote":"Shows that pedestrian and bicycle crashes are underreported, justifying encounter crowd-sensing as a complementary safety data source."},{"cited_title":"Pedestrians and e-scooters: An initial look at e-scooter parking and perceptions by riders and non-riders,","cited_arxiv_id":null,"evidence_quote":"Is the prior observational study of e-scooter parking and rider/non-rider perceptions that the paper extends with continuous encounter sensing."}],"review_version":1}