{"id":"740cdeab-b268-4d78-b50a-c08460bb24db","arxiv_id":"2412.13600","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A Bluetooth RSSI and Extended Kalman Filter system, paired with a cloud assignment step, matched construction workers to active power tools with up to 98.6% accuracy in indoor and outdoor tests.","lead":"This paper presents and tests a system that matches construction workers with the power tools they are using by measuring Bluetooth signal strength from wearable badges and tool tags. If it works at scale, it could automate tool usage tracking, safety compliance, and maintenance reporting on job sites without costly indoor positioning infrastructure.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Matching accuracy is validated only at easy separations: the true operator is always within 0.5 m while competing workers are placed at 0.5–3 m, so the 98.6% result never stresses the nearest-operator premise or the 0.75 m trust rule.","rationale":"The reader's conditional verdict already identified the 0.75 m assumption and the risk that close clustering breaks matching confidence. This stress-test sharpens that point by showing the experiments never vary that assumption: the true operator is always inside 0.5 m and all competitors are outside it, so the data cannot distinguish a robust nearest-operator rule from a trivial separation artifact. A leave-one-out recalibration would address the in-sample path-loss fit also noted by the reader, but it would not test the matching decision under close spacing; the missing-condition experiment is the more load-bearing check for the central matching claim. If the close-spacing trial succeeds, the concern is resolved and conditional acceptance stands. If it fails, the paper should state explicitly that the method targets sparse work layouts and does not claim reliability when non-operators stand within the 0.75 m threshold.","tokens_in":19603,"tokens_out":11112,"duration_ms":101809,"concrete_test":"Run the §V-B indoor matching scenarios with an additional condition: place a non-operator bystander at 0.25, 0.5, and 0.75 m from the active tool while the operator works at their natural distance, and report overall accuracy and SURE precision per spacing. If accuracy at 0.5 m or 0.75 m separation falls materially below the reported 89.7% indoor figure, the headline matching claim is contingent on sparse spacing and needs a caveat.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing weakness is that the matching evaluation does not include the regime in which the method has to make a non-trivial decision. In the experiments of §V-B, the true operator is by construction within arm's reach of the tool while all competing persons are positioned at 0.5, 1, 2, or 3 m; §VI-B confirms that 'the distance between a user and its asset during the experiments was always below 0.5 m.' Consequently, no trial places a non-operator closer to the active tool than the operator, and none places a non-operator within the 0.75 m margin used in Eq. (13) to declare a SURE match. The reported indoor accuracy of 89.7%, indoor SURE precision of 98.6%, and outdoor accuracy of 98.6% are therefore measured under a favorable separation distribution where the nearest-distance rule is nearly always sufficient. Given the distance estimator's 1.63 m standard deviation (Table II), close-spacing cases could easily invert the ordering of two candidates and produce a confident wrong match, but the data do not contain such cases. This is a missing-condition problem, not merely a generalization concern: the central claim 'nearest operator is the user' is never empirically challenged.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an asset-operator matching system for industrial tools that combines BLE RSSI-based distance estimation on low-power wearables with a cloud-side nearest-operator assignment algorithm. The wearable runs a one-state Extended Kalman Filter (EKF) over RSSI measurements to estimate distance to active BLE tags attached to tools; the cloud algorithm solves a linear assignment problem and labels matches as SURE or UNSURE based on a 0.75 m margin. The authors built a prototype with existing hardware (SMART TAG and ECOTRACK), validated it indoors and outdoors on a construction site, and report a median distance estimation error of 0.49 m, an indoor matching accuracy of 89.7% with 98.6% SURE-precision, and an outdoor accuracy of 98.6%. The main claim is that accurate asset-user matching is feasible without RTLS infrastructure using low-power BLE devices.","tokens_in":19910,"tokens_out":3229,"duration_ms":45511,"significance":"If the reported results hold independently, the work addresses a real industrial gap with a low-cost, low-power solution: it combines activity detection on the tool, RSSI-based ranging on a badge, and cloud assignment, and it validates on physical hardware in a realistic setting. The power measurements and the dynamic tool-swapping experiments are concrete strengths, as is the comparison with prior Bayesian-filter RSSI ranging work. The central limitation is that the quantitative evidence is partly in-sample: the path-loss parameters, measurement noise covariance, and initial distance clamps are fitted on the same RSSI data used to report the 0.49 m median error, and the matching experiments do not include the regime in which the nearest-operator premise is actually challenged. The core idea is sensible and plausibly correct, but the paper needs additional independent validation before the headline numbers can be taken as evidence of real-world performance.","major_comments":[{"comment":"The distance accuracy claim is evaluated in-sample. The path-loss parameters (n = 1.011, RSSI(x0) = -45.6 dB) are fitted to the 55,097 RSSI measurements in Fig. 3, the measurement noise covariance R_k = 43.53 in Eq. (6) is set using the same measurement set, and the initial distance clamps in Eq. (7) are chosen from that fitted curve. The 2338 distance estimates used to report the median error of 0.49 m are then drawn from the same overall dataset. An independent test set, cross-validation, or a separate calibration session is needed before the 0.49 m figure can be treated as a realistic accuracy estimate.","section":"§VI-A, §VI-B, Eq. (1), Eq. (6), Eq. (7)"},{"comment":"The matching experiments do not exercise the nearest-operator premise in a discriminating way. In the reported tests the true operator is always within 0.5 m of the active tool, while all competing persons are placed at distances of 0.5, 1, 2, or 3 m; the paper states in §VI-B that 'the distance between a user and its asset during the experiments was always below 0.5 m.' Consequently, no trial places a non-operator closer to the asset than the true operator, and no trial places a non-operator within the 0.75 m trust margin of Eq. (13). Given the distance estimator's standard deviation of 1.63 m (Table II), close-spacing cases could invert the order of candidates, but such cases are absent from the evaluation. The reported accuracy of 89.7% indoor and 98.6% outdoor therefore reflects a favorable separation distribution and does not validate the central claim that the nearest operator is the user.","section":"§V-B, §VI-B, §VI-C"},{"comment":"The trust-level classifier is calibrated by an assumed minimum interpersonal distance of 0.75 m, which the paper itself states 'comes from the assumption that two users do generally not stand closer to each other.' No sensitivity analysis is provided for this threshold, yet the headline SURE-precision figures depend directly on it. The authors should report how the precision, recall, and accuracy of matching vary with the threshold, or otherwise justify the threshold from data.","section":"§IV-B, Eq. (13)"}],"minor_comments":[{"comment":"The abstract reports a 'median accuracy of 0.49 m' while Table II and the text state 'Median Error 0.49 m'; the wording should be consistent to avoid implying a different metric.","section":"Abstract, §VI-B"},{"comment":"The conclusion states an indoor accuracy of 87.7%, while §VI-C reports 89.7%. This numerical inconsistency should be reconciled.","section":"§VI-C vs §VII"},{"comment":"The 'exhaustiveSearch()' step is described only in prose. For the stated scalability assumption of at most 15 active assets, an explicit enumeration is acceptable, but the pseudo-code should state the complexity or the search-space size.","section":"Algorithm 1, Line 9"},{"comment":"The notation 'v_max^2 / X^2_{1,c}' is terse; a brief explanation of the chi-squared scaling factor and the chosen confidence level would improve readability.","section":"§IV-A, Eq. (3)"},{"comment":"The confusion matrices would benefit from a caption that defines the rows and columns, since the text refers to 'dark blue color means many results' without an explicit mapping of colors to counts.","section":"Fig. 7"},{"comment":"The downsampling of 0.5 s advertisements to 7 s intervals to create a larger dataset is a useful technique, but it should be stated clearly that the resulting 2338 estimates are not fully independent samples for statistical purposes.","section":"§V-B"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a solid applied IoT paper, but the two load-bearing quantitative claims—the 0.49 m distance accuracy and the matching accuracy figures—rest on in-sample parameter fitting and a favorable experimental separation regime. These are fixable within the manuscript's scope by adding out-of-sample validation and targeted close-spacing experiments, so I recommend major revision rather than rejection. The involvement of Hilti co-authors is transparent and appropriate given the industrial context; I see no citation or novelty concerns beyond the experimental validation gap."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The stress-test note holds up, and it is the key thing to know about this paper. The system is real, the writing is clear, and the matching formulation (nearest active-asset assignment with a trust level) is a legitimate new contribution for applied IoT. The authors built a physical prototype, measured power consumption carefully, and compared with prior RSSI ranging work in a useful table. The outdoor results are genuinely strong.\n\nThe soft spots are in the validation, not the concept. First, the 0.49 m median distance error is an in-sample number: the path-loss parameters (RSSI(x0) = −45.6 dB, n = 1.011) and the measurement noise covariance R_k = 43.53 are fitted to the same 55,097 RSSI values used to compute the error. There is no held-out set, and the paper even states the variance of the raw RSSI is 48.92 dB², so the filter is essentially tuned to this environment. Second, and more importantly, the matching evaluation never challenges the central premise. In every indoor and outdoor test, the true operator is always within 0.5 m of the tool while competing workers are placed at 0.5, 1, 2, or 3 m. Section VI-B confirms this explicitly. That means no trial has a non-operator closer to an active tool than the operator, and no trial places a non-operator within the 0.75 m margin that triggers an UNSURE classification. The 98.6% SURE precision is therefore measured in a regime where the nearest-distance rule is trivially sufficient. With a distance estimator standard deviation of 1.63 m, a close-spacing case could easily invert the ordering, but the data never contain such a case.\n\nThere is also a promised baseline that never appears: Section VI-B says the accuracy of EKF and cumulative average (CA) algorithms are compared, but no CA results are ever reported. That is a straightforward omission. Minor point: the conclusion gives indoor accuracy as 87.7% while Section VI-C reports 89.7% for the same metric; presumably a typo.\n\nNone of this kills the paper. It is a competent engineering contribution with a plausible central claim, and the authors are honest about some limitations (the 0.75 m threshold is stated as an assumption, and future work mentions sensitivity analysis). But the current evidence supports \"the system works when the operator is clearly closest,\" not the stronger claim that it reliably identifies the operator under realistic ambiguity. A serious referee should send this to revision and ask for a stress test with closer competitor spacing, the CA baseline, and a proper train/test split on the path-loss model.","headline":"A solid applied IoT paper with a real prototype and a novel matching formulation, but the headline matching accuracy is validated only in an easy regime where the nearest-operator premise is never stressed.","tokens_in":20404,"tokens_out":2467,"would_cite":false,"duration_ms":22378,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Matching construction tools to their operators can be done from Bluetooth signal strength alone, with a reported outdoor accuracy of 98.6%.","keywords":["iot","edge computing","bluetooth low energy","rssi distance estimation","extended kalman filter","asset tracking","construction site","activity recognition"],"falsifier":"Take one active tool and two operators at measured distances of 0.4 m and 1.0 m, run the published EKF and assignment solver on fresh RSSI data, and compare match results to video ground truth. A wrong SURE decision in more than a small fraction of trials, or a systematic reversal of the distance ordering, would falsify the nearest-operator rule as implemented.","tokens_in":19443,"feed_emoji":"🔧","tokens_out":8484,"duration_ms":70506,"temperature":0.7,"pith_summary":"Asset-operator matching in industry usually requires either a biometric reader on each tool or a full real-time locating system. This paper argues that a much cheaper cue suffices: whoever stands closest to a handheld tool while it is active is the person using it. The authors build a system consisting of a BLE tag on the tool that detects and broadcasts the tool's active state, a wearable badge carried by each worker that estimates the distance to every active tag from received signal strength, and a cloud algorithm that assigns each active tool to its nearest operator. The estimates are smoothed with an Extended Kalman Filter, so only small amounts of filtered data are sent over the cellular link. If the claim holds, construction, manufacturing, and healthcare fleets can obtain per-operator usage logs, safety-training checks, and maintenance schedules from infrastructure they may already partly own.","feed_headline":"Proximity alone matches workers to power tools 98.6% of the time","feed_subtitle":"Low-power badges and tool tags infer who used each device from measured proximity, without beacon infrastructure.","key_machinery":"The load-bearing mechanism is the nearest-operator premise together with a one-state Extended Kalman Filter (EKF) and a cloud-side assignment solver. The EKF, defined by a no-motion motion model $x_k = x_{k-1} + w_k$ and a logarithmic path-loss observation model $h(x) = \\operatorname{RSSI}(x_0) - 10n\\log_{10}(x/x_0)$, runs on the wearable badge and converts noisy RSSI readings into a filtered distance per active tag. The assignment solver then minimizes $\\sum_i\\sum_j\\sum_t a_{ijt}\\, d(i,j,t)$, subject to each active asset receiving exactly one operator and each operator remaining fixed for the duration of a continuous active session, and it labels a match UNSURE whenever the distance to the second-nearest operator is below 0.75 m.","core_discovery":"On its own terms, the paper claims that RSSI-based proximity alone can resolve who-is-using-which-asset, and that the resolution is accurate enough for real deployment: a median distance-estimation error of 0.49 m, an indoor matching accuracy of 89.7%, and an outdoor matching accuracy of 98.6%, with 100% precision among high-confidence outdoor matches. The argument is that absolute distance accuracy is not required; only the ordering of operators by distance to the active asset matters, and the system's two algorithms preserve that ordering. The matching accuracy is high when the distance gap between the nearest and second-nearest operator exceeds the 0.75 m trust threshold, and the classifier intentionally withholds confidence (\"UNSURE\") when it does not.","pith_inferences":["Editorial extension: the method's reliance on relative ordering rather than absolute distance suggests that per-wearable RSSI calibration could remove systematic bias and push the usable range beyond the tested 0.5–6 m.","Editorial extension: the 0.75 m UNSURE threshold is a tunable safety valve; a deployment with many workers in close quarters could trade recall for precision by raising it, or use fixture spacing to lower it.","Editorial extension: the same pipeline could be applied to medical equipment and manufacturing machines, provided the asset's active state is detectable from its accelerometer or another low-power sensor.","Editorial extension: a natural test is to re-run the experiments in a second building with different clutter; if the fitted path-loss parameters need to be re-learned, the system would require a short per-site calibration step."],"forward_implications":["Per-operator usage logs, safety-training compliance checks, and maintenance triggers can be generated automatically from BLE-based asset-tracking infrastructure that many sites already run.","Because only the nearest operator is selected, the system tolerates biased distance estimates as long as they preserve the ordering of operators around an asset.","Wearables need to transmit only one aggregated distance per active session, which the power measurements put at 0.39 Wh per day and a 28-day battery life on the tested device.","The EKF's median error of 0.49 m is sufficient for matching at typical power-tool working distances, but the trust classifier should be used to filter out decisions made in dense operator layouts.","In the paper's tests, the dynamic scenario with workers swapping tools mid-session still achieved 98.9% precision, which suggests the session-continuity constraint is not a practical handicap."],"supporting_citations":[{"why":"Shows Bayesian filters improve BLE RSSI distance estimation and supplies the baseline against which the EKF choice is justified.","marker":"[17]"},{"why":"An earlier EKF for RSSI-based indoor localization that is adapted here to one-state ranging on a wearable.","marker":"[41]"},{"why":"A construction-site EKF using a no-motion motion model, the closest prior deployment and the accuracy figure the paper compares against.","marker":"[43]"},{"why":"Defines the no-motion motion model that provides the state-evolution equation for the distance filter.","marker":"[46]"},{"why":"Supplies the standard EKF prediction and update equations used in the implementation.","marker":"[47]"},{"why":"Describes the asset-side tag hardware whose accelerometer and BLE radio carry the activity detection and advertisement.","marker":"[18]"},{"why":"Describes the wearable hardware that runs the EKF and reports distances over the cellular link.","marker":"[22]"},{"why":"Supplies the activity classifier that decides when the tool is active and triggers the matching window.","marker":"[51]"}],"fun_headline_variants":["Proximity alone matches operators to tools with 98.6% accuracy","RSSI ordering, not absolute distance, drives operator-tool matches","Low-power BLE tags hit 98.6% worker-tool match rate outdoors","Proximity-based matching hits 98.6% accuracy via low-power IoT"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the closest person to an active tool is the one using it, and that the fitted RSSI path-loss model plus the 0.75 m separation rule keep that ordering visible in real deployments.","fun_headline_variants_meta":{"raw":{"variants":["Proximity alone matches operators to tools with 98.6% accuracy","RSSI ordering, not absolute distance, drives operator-tool matches","Low-power BLE tags hit 98.6% worker-tool match rate outdoors","Proximity-based matching hits 98.6% accuracy via low-power IoT"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000935,"raw_usage":{"total_tokens":4002,"prompt_tokens":951,"completion_tokens":3051,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":567,"completion_tokens_details":{"reasoning_tokens":2966}},"tokens_in":567,"tokens_out":3051,"duration_ms":19501,"temperature":1.0,"reasoning_tokens":2966,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T12:58:04.339773+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take one active tool and two operators at measured distances of 0.4 m and 1.0 m, run the published EKF and assignment solver on fresh RSSI data, and compare match results to video ground truth. A wrong SURE decision in more than a small fraction of trials, or a systematic reversal of the distance ordering, would falsify the nearest-operator rule as implemented.","supporting_citations":[{"cited_title":"Method for improving indoor positioning accuracy using extended kalman filter,","cited_arxiv_id":null,"evidence_quote":"An earlier EKF for RSSI-based indoor localization that is adapted here to one-state ranging on a wearable."},{"cited_title":"Monitoring harness use in construction with ble beacons,","cited_arxiv_id":null,"evidence_quote":"A construction-site EKF using a no-motion motion model, the closest prior deployment and the accuracy figure the paper compares against."},{"cited_title":"Fernandez-Madrigal and J","cited_arxiv_id":null,"evidence_quote":"Defines the no-motion motion model that provides the state-evolution equation for the distance filter."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the standard EKF prediction and update equations used in the implementation."},{"cited_title":"Optimizing IoT-Based Asset and Utilization Tracking: Efficient Activity Classification with MiniRocket on Resource-Constrained Devices","cited_arxiv_id":"2310.14758","evidence_quote":"Supplies the activity classifier that decides when the tool is active and triggers the matching window."}],"review_version":1}