Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Deciphering Delivery Mobility: A City-Scale, Path-Reconstructed Trajectory Dataset of Instant Delivery Riders

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A new Beijing dataset reconstructs 79,648 multi-stop delivery tours from order records, claiming to faithfully capture delivery riders' mobility at city scale.

desk verdict A genuinely useful open dataset of reconstructed delivery tours, but the validation is weaker than the 'high fidelity' claim because the source records may themselves be simulated. read the letter →

arxiv 2507.11584 v1 pith:WEFFAMKW submitted 2025-07-15 physics.soc-ph

Chengbo Zhang , Yonglin Li , Zuopeng Xiao This is my paper
classification physics.soc-ph
keywords instantdeliveryriderstrajectorydatasetpathreconstructionwaveurbanmobilitygigeconomyBeijingcyclingroutesimulation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper presents a city-scale dataset of 79,648 multi-stop delivery tours ("waves") completed by 986 instant-delivery riders in Beijing, reconstructed from anonymized order records by simulating cycling-optimized routes between each consecutive action stop. The authors claim the resulting trajectories faithfully represent the complex, multi-stop character of on-demand delivery work, and support this with Pearson correlations of 0.92 for route distance and 0.79 for route duration against the source records. If the reconstruction holds, the dataset gives researchers an open resource for studying rider mobility, workload, and urban logistics at a granularity previously unavailable.

What carries the argument

The central object is the "delivery wave": a complete multi-stop tour assigned to a single rider, defined by a chronologically ordered sequence of action steps (assign, pickup, delivery) with an expectation time for each. The mechanism is a path-reconstruction framework that turns each wave into a series of point-to-point cycling-optimized routing queries to a major online map service, then concatenates the returned high-resolution polylines and interpolates timestamps along them, producing one continuous trajectory per wave.

What would settle it

Collecting GPS traces from a sample of Beijing delivery riders during February 2020 and comparing each trace, segment by segment, to the reconstructed path for the same order wave would settle the claim: if actual riders routinely deviate from the map service's cycling-optimized routes, or if the source waves cannot be matched to any real rider, the dataset's fidelity claim fails.

Watch

Extended reading notes

Core claim

The central claim is that a rider's delivery tour can be recovered as a continuous spatiotemporal trajectory by decomposing each "wave" — a bundle of orders assigned to one rider — into chronologically ordered assign, pickup, and delivery actions, then querying an online map service for cycling routes between successive actions and concatenating the returned polylines with interpolated timestamps. The paper argues this wave-based reconstruction captures the true multi-stop routing decisions of instant delivery riders, unlike datasets that treat each order as an isolated origin-destination trip. The validation against the source data's recorded route distances and wave durations yields Pearson r = 0.92 and r = 0.79, which the authors take as evidence of high fidelity for describing delivery mobility.

Load-bearing premise

The load-bearing premise is that the source competition dataset's recorded distances and durations are real observed travel metrics; the organizers' statement that the data is 'desensitized simulated data' leaves open the possibility that these ground-truth values were synthetically generated, in which case the validation would only show the map service agrees with a simulator.

Editorial extensions

If this is right

  • Researchers can use the dataset to identify delivery corridors, hotspots, and route-choice behaviors that differ from general urban traffic.
  • The trajectories provide an objective basis for quantifying rider workload, work scheduling, and their links to the urban daily rhythm.
  • The data can inform data-driven policies on the environmental footprint and safety risks of instant delivery.
  • Because each wave records the full multi-stop sequence, the dataset supports modeling of order bundling and its efficiency gains.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the competition organizers' phrase "desensitized simulated data" means the order records were synthetically generated rather than observed, then the reported correlations show agreement with that simulator, not with real rider behavior, and the dataset's fidelity claim would need independent ground-truth testing.
  • The reconstruction assumes riders follow the map service's cycling-optimized path at a constant speed; real riders may take shortcuts, ride on sidewalks, or wait at restaurants, so the derived timestamps likely underestimate wave duration in ways that could bias workload studies.
  • A natural testable extension would be to validate the reconstruction against a small sample of GPS-tracked delivery riders on the same order logs, measuring path overlap and timing error.
  • The same wave-to-trajectory pipeline could be applied to other cities or platforms if similar batched-order records exist, making the method a general route-reconstruction recipe for gig-economy mobility.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents a city-scale dataset of reconstructed delivery trajectories for 79,648 delivery waves performed by 986 riders in Beijing. The authors take an open Eleme/Tianchi dataset of delivery tasks, sort each wave's action steps chronologically, and call the Amap routing API to generate cycling-optimized paths between consecutive actions. The resulting GeoJSON dataset contains route LineStrings and action points, and is validated by comparing reconstructed total distances and durations against source-recorded metrics (Pearson r=0.92 and r=0.79), plus spatial and temporal pattern checks. The central claim is that the dataset 'faithfully represents' real delivery-rider mobility and is therefore useful for studying rider behavior and urban logistics.

Significance. If the fidelity claim were established, this would be a valuable open resource: it provides wave-level, multi-stop trajectories rather than disconnected OD pairs, and it ships with public data (figshare) and code (GitHub), which are concrete strengths. The paper also gives a clear workflow and schema. However, the significance depends entirely on two premises: (1) that the source Eleme records are observed ground truth rather than synthetic data, and (2) that riders actually follow the API-computed cycling routes. The manuscript itself quotes the organizers describing the source as 'desensitized simulated data,' which directly threatens premise (1). The reported correlations cannot resolve this because they are computed against the same source used to build the routes. No independent GPS, survey, or platform data is used. The dataset is therefore, at present, best described as a plausible synthetic trajectory generation pipeline conditioned on the source wave structure, not a validated record of real rider paths.

major comments (4)
  1. [Methods, 'Delivery record datasets'] The paper quotes the competition organizers as stating that the source data is 'desensitized simulated data,' then immediately reinterprets this as meaning that personal identifiers were removed or anonymized. These are different claims: de-identification applies to observed records, whereas 'simulated' indicates that the records themselves may be synthetically generated. This distinction is load-bearing because the Technical Validation treats the source fields r_dis_all and r_dur_all as 'ground-truth travel metrics.' If the source is synthetic, the reported r=0.92 and r=0.79 only show that Amap routing is internally consistent with the synthetic generator, and they cannot establish fidelity to real rider behavior. The authors must clarify the source data's actual provenance (ideally with the organizers' original wording in full) or supply an independent validation set, such as GPS traces or platform-side records not used in the reconstruction.
  2. [Technical Validation, Fig. 2] The validation is not independent: the same Eleme wave definitions provide both the action sequence used to build the routes and the 'ground-truth' metrics against which the routes are compared. Furthermore, the two Pearson coefficients are presented without sample sizes, confidence intervals, or diagnostics for systematic bias, and only wave-level totals are compared. On a log-log scatter, a high r can mask consistent short-trip overestimation or long-trip underestimation. The authors should report Bland-Altman limits of agreement, mean absolute percentage error, and confidence intervals, and ideally compare segment-level or action-level distances and durations rather than only aggregate wave totals.
  3. [Path-reconstruction framework, 'delivery wave' definition and 'expect_time' sorting] The action sequence within each wave is sorted by expect_time timestamps, but no evidence is provided that riders execute actions in exactly the scheduled order. If actual pickup or delivery order deviates from expected times, the concatenated route sequence may not reflect the rider's real tour. The manuscript should either justify the use of expected times with evidence or discuss how sensitive the reconstructed routes are to this ordering assumption.
  4. [Path-reconstruction framework and Discussion of route-choice assumptions] The entire trajectory generation assumes that a rider follows the Amap cycling-optimized path for each consecutive action step. The manuscript provides no GPS traces, field observations, or independent platform data to test this assumption, and it does not discuss how e-bike use, shortcuts, or multi-order pickup strategies might cause deviations. As a result, the abstract's claim that the dataset has 'high fidelity' and 'faithfully represents' rider mobility is stronger than what is demonstrated. The dataset should be framed as route-simulated trajectories conditioned on the source wave data, with the route-choice assumption stated as an explicit limitation and, if possible, tested on a small independent sample.
minor comments (5)
  1. [Fig. 7 caption] The caption contains 'the average order in a bundle is x,' which is a literal placeholder; replace 'x' with the computed average value.
  2. [Throughout, typos] There are multiple typos and grammatical errors, for example 'deliverd' (Fig. 7 caption), 'fluction' (Fig. 7 paragraph), 'differentiate with weekdays' (Fig. 4 paragraph), and 'the peakaccuringat11amatweekdays' (Usage Notes). These should be corrected in a careful proofreading pass.
  3. [Fig. 2] The log-log scatter plots lack axis labels and units; please label the axes explicitly for both pairs (nav_dis vs r_dis_all and nav_dur vs r_dur_all).
  4. [Table 1, r_time_lst] The field r_time_lst is described as 'real UNIX timestamps for each action,' but the method section says actions are sorted by 'expect_time'; clarify whether the timestamps in the dataset are actual or expected times, and define 'real' in this context.
  5. [Technical Validation, duration discussion] The sentence attributing the weaker duration correlation to waiting time says this introduces a 'reduced bias' of estimated travel time; the intended meaning is likely a downward bias, and the phrasing should be corrected.

Circularity Check

2 steps flagged · score 4.0 of 10

Fidelity claim rests on reinterpreting 'simulated' source data as anonymized ground truth; the r=0.92/0.79 validations show self-consistency with the source, not independent evidence of real rider mobility.

  1. self definitional [Methods, 'Delivery record datasets']
    "The competition organizers explicitly state that this data is desensitized simulated data, meaning all personal identifiers such as real rider names or contact information were removed or anonymized prior to its release."

    The paper defines 'simulated' as merely 'anonymized,' converting the organizers' caveat that the source is synthetic simulation into a claim that it is observed real-world data. The central fidelity claim and the validation against 'ground-truth travel metrics' both depend on this redefinition. If the Eleme source is a simulator's output, then the high correlations only show that Amap routing agrees with a synthetic generator, not that the dataset represents real delivery riders' mobility. The conclusion is thereby obtained by defining the input as ground truth rather than by demonstrating external correspondence.

  2. other [Methods, 'Path-reconstruction framework'; Technical Validation, 'Correlation with route-level ground-truth records']
    "To enable subsequent evaluation of our generated routes, we also recorded the total duration and grid distance for each wave, as provided in the source data, to serve as comparable metrics. ... comparing ... the total recorded route distance (r_dis_all) versus the API-based reconstructed route distance (nav_dis), and ... total recorded wave duration (r_dur_all) versus the API-based reconstructed route duration (nav_dur)."

    The 'ground-truth' metrics r_dis_all and r_dur_all are inherited from the same Eleme source file from which the wave's ASSIGN/PICKUP/DELIVERY sequence is extracted. The reconstruction is constrained to visit exactly those points in exactly that chronological order, so nav_dis and nav_dur are not independent of the source's own wave metrics. The reported Pearson correlations therefore demonstrate internal consistency between an external routing service and the source data's self-reported distances and durations; they do not independently validate that the reconstructed trajectories faithfully represent real rider behavior. The validation is circular as evidence for the paper's fidelity claim.

full rationale

The reconstruction pipeline itself is not circular in the parameter-fitting sense: it uses an external Amap routing API, contains no fitted parameters, and does not rely on a self-citation chain. The dataset and code are publicly available, and the generation of trajectories from discrete action points is a legitimate external computation. However, the paper's strongest claim—that the dataset 'faithfully represents' on-demand delivery work—is supported primarily by correlating reconstructed metrics with source-data metrics that also define the action sequence. This is compounded by the Methods passage that reinterprets the organizers' explicit 'desensitized simulated data' warning as merely anonymization. Under the literal reading, the source is synthetic, so the r=0.92 and r=0.79 values are not ground-truth validation but self-consistency checks between Amap routes and a simulator. Even under the most favorable reading, the absence of GPS traces, field observations, or independent platform data means the fidelity claim is not externally anchored. These issues make the validation partly circular, but they do not make the entire derivation equivalent to its input, and there is no self-citation or uniqueness-theorem loading. Score 4 reflects partial circularity in the central fidelity argument rather than a fully forced or definitionally empty result.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the authenticity of the Eleme source data and the realism of Amap routing. Neither is independently verified in the paper. There are no fitted free parameters; all quantities come from the source data or the routing API. No new physical or conceptual entities are introduced.

assumptions (3)
  • domain assumption The Eleme Tianchi source dataset accurately represents real delivery wave structure and provides reliable travel metrics for ground-truth comparison.
    Invoked in 'Delivery record datasets' and used as ground truth in Technical Validation. The organizers' 'desensitized simulated data' statement raises the possibility that the data is synthetic, which would invalidate the ground truth.
  • domain assumption Amap's cycling-optimized routing produces paths that delivery riders plausibly take.
    Used throughout the path-reconstruction framework (Methods). No independent GPS or observational validation is provided.
  • domain assumption Linear timestamp interpolation along each navigation segment, using the API's estimated travel time, yields a realistic timing for the trajectory.
    Described in 'Path-reconstruction framework'. The authors acknowledge waiting time at pickups is not modeled.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deciphering Delivery Mobility: A City-Scale, Path-Reconstructed Trajectory Dataset of Instant Delivery Riders." pith.science (2026). https://pith.science/paper/WEFFAMKW

@misc{pith2026250711584,
  author       = {Pith},
  title        = {Pith review of: Deciphering Delivery Mobility: A City-Scale, Path-Reconstructed Trajectory Dataset of Instant Delivery Riders},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WEFFAMKW}},
  note         = {Machine review of arXiv:2507.11584}
}
read the original abstract

The rapid expansion of the on-demand economy has profoundly reshaped urban mobility and logistics, yet high-resolution trajectory data on delivery riders' consistent movements remains scarce. Here, we present a city-scale, high-resolution spatiotemporal trajectory dataset of on-demand instant delivery riders in Beijing. This dataset was produced through a path-reconstruction methodology applied to an open dataset containing delivery order information. Subsequently, detailed and continuous trajectories were reconstructed by simulating cycling routes via a major online map service to ensure they were realistically aligned. For validation, the reconstructed paths were compared against ground-truth travel metrics, revealing a strong correlation with actual travel patterns. The analysis yielded Pearson correlation coefficients of 0.92 for route distance and 0.79 for route duration. This high fidelity ensures the dataset's utility for describing delivery riders' mobility. This publicly available resource offers unprecedented opportunities for researchers in urban planning, transportation studies, logistics optimization, and computational social science to investigate rider behavior, model urban freight systems, and develop more efficient and sustainable city-wide logistics solutions.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Involution game with migration and spatial heterogeneity of social resources

    physics.soc-ph 2026-03 unverdicted novelty 5.0 of 10

    Equal regional resources suppress involution while resource disparity and higher total resources promote it; migration probability barely changes the evolutionary outcome.

Reference graph

Works this paper leans on

23 extracted references · 21 canonical work pages · cited by 1 Pith paper

  1. [1]

    & Haworth, N

    Oviedo-Trespalacios, O., Rubie, E. & Haworth, N. Risky business: Comparing the riding behaviours of food delivery and private bicycle riders. Accident Analysis & Prevention 177, 106820(2022)

    work page 2022

  2. [2]

    & Rallet, A

    Aguilera, A., Dablanc, L. & Rallet, A. Digital Work and Urban Delivery: Profile, Activity and Mobility Practices of On-Demand Food Delivery Couriers in Paris (France). Information 13, 433(2022)

    work page 2022

  3. [3]

    & Louvet, N

    Krier, C., Dablanc, L., Aguiléra, A. & Louvet, N. Sharing within the gig economy: The use of shared e-bikes by on-demand platform-based instant meal delivery workers in Paris. Case Studies on Transport Policy 10,2280–2289(2022)

    work page 2022

  4. [4]

    How Many Uber Drivers Are There In 2025 (Stats)

    Kumar, N. How Many Uber Drivers Are There In 2025 (Stats). DemandSage https://www.demandsage.com/uber-statistics/(2025)

    work page 2025

  5. [5]

    Backlinko https://backlinko.com/doordash-users(2021)

    How Many People Use DoorDash in 2025? [New Data]. Backlinko https://backlinko.com/doordash-users(2021)

    work page 2021

  6. [6]

    TechNode 16 http://technode.com/2024/09/19/meituan-paid-11-3-billion-to-millions-of-delivery-riders-i n-2023-averaging-less-than-0-52-per-order/(2024)

    Dong,C.Meituanpaid$11.3billiontomillionsofdeliveryridersin2023,averaginglessthan $0.52 per order · TechNode. TechNode 16 http://technode.com/2024/09/19/meituan-paid-11-3-billion-to-millions-of-delivery-riders-i n-2023-averaging-less-than-0-52-per-order/(2024)

    work page 2024

  7. [7]

    https://www.grandviewresearch.com/industry-analysis/online-food-delivery-services-market

    Online Food Delivery Services Market | Industry Report, 2030. https://www.grandviewresearch.com/industry-analysis/online-food-delivery-services-market

    work page

  8. [8]

    et al.CarbonemissionsfromurbantakeawaydeliveryinChina

    Zhong,Y. et al.CarbonemissionsfromurbantakeawaydeliveryinChina. npj Urban Sustain 4,39(2024)

    work page 2024

Show all 23 references
  1. [9]

    Lord, C. et al. The sustainability of the gig economy food delivery system (Deliveroo, UberEATS and Just-Eat): Histories and futures of rebound, lock-in and path dependency. International Journal of Sustainable Transportation 17,490–502(2023)

    2023

  2. [10]

    Matsuyuki, M. et al. Assessment of the sustainability of online food delivery from the perspectiveofCO2emissionsandtransportintensity:AcasestudyinJakarta. Transportation Research Interdisciplinary Perspectives 27,101200(2024)

    2024

  3. [11]

    Zhang, Y. et al. Urban food delivery services as extreme heat adaptation. Nat Cities 1–10 (2025)doi:10.1038/s44284-024-00172-z

    2025 doi

  4. [12]

    &Tironi, M

    Vecchio,G., Tiznado-Aitken,I., Albornoz,C. &Tironi, M. Delivery workersandtheinterplay ofdigitalandmobility(in)justice. Digital Geography and Society 3,100036(2022)

    2022

  5. [13]

    & Yun, G

    Yu,D., Zhang, J. & Yun, G. Delivery riders’safetyanddelivery efficiency in on-demand food deliveryindustry:Themoderatingroleofmonitoringalgorithms. Research in Transportation Business & Management 55,101143(2024)

    2024

  6. [14]

    Yabe, T. et al. YJMob100K: City-scale and longitudinal dataset of anonymized human mobilitytrajectories. Scientific Data 11,397(2024)

    2024

  7. [15]

    Comput.Urban Sci

    Qiu,Y.,Ding,J.,Wang,M.,Hu,L.&Zhang,F.Understandingtheurbanlifepatternofyoung peoplefromdeliverydata. Comput.Urban Sci. 1,28(2021)

    2021

  8. [16]

    Zhu, L. et al. Order Fulfillment Cycle Time Estimation for On-Demand Food Delivery. in Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining2571–2580(AssociationforComputingMachinery,NewYork,NY,USA,2020). doi:10.1145/3394486.3403307

    2020

  9. [17]

    Computers, Environment and Urban Systems 108,102059(2024)

    Zare,P.,Leao,S.,Gudes,O.&Pettit,C.Asimpleagent-basedmodelforplanningforbicycling: Simulation of bicyclists’ movements in urban environments. Computers, Environment and Urban Systems 108,102059(2024)

    2024

  10. [18]

    Transportation Research Part D: Transport and Environment 122,103859(2023)

    Cai,Y.,Ong,G.P.&Meng,Q.Understandingbike-sharingasacommutemodeinSingapore: An agent-based simulation approach. Transportation Research Part D: Transport and Environment 122,103859(2023)

    2023

  11. [19]

    Kashiyama,T.,Pang,Y.,Sekimoto,Y.&Yabe,T.Pseudo-PFLOW:Developmentofnationwide synthetic open dataset for people movement based on limited travel survey and open statisticaldata.Preprintathttps://doi.org/10.48550/arXiv.2205.00657(2022). 17

    2022 doi arXiv

  12. [20]

    Toledo, T. et al. Intercity truck route choices incorporating toll road alternatives using enhancedGPSdata. Transportmetrica A: Transport Science 16,654–675(2020)

    2020

  13. [21]

    Bandeira, J. M. et al. Exploring multiple eco-routing guidance strategies in a commuting corridor. International Journal of Sustainable Transportation 12,53–65(2018)

    2018

  14. [22]

    Transportation Research Part D: Transport and Environment 101,103093(2021)

    Sampaio,C.,Coelho,M.C.,Macedo,E.&Bandeira,J.M.Emissionsbasedtolls–Impactson thetotalemissionsofanintercitycorridor. Transportation Research Part D: Transport and Environment 101,103093(2021)

    2021

  15. [23]

    figshare https://doi.org/10.6084/m9.figshare.29314796.v1(2025)

    City-Scale, Path-Reconstructed Trajectory Dataset of Instant Delivery Riders. figshare https://doi.org/10.6084/m9.figshare.29314796.v1(2025)

    2025 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.