{"id":"838eebbf-4f56-4d95-8ea5-342983e7b26f","arxiv_id":"2607.22654","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"MINT-V2X is a new publicly released simulated dataset coupling vehicle trajectories with per-vehicle wireless network metrics at 10 Hz, with a case study showing trajectory data improves RSU load prediction.","lead":"This paper presents MINT-V2X, a simulated dataset that pairs the movements of 1,386 vehicles with the wireless-network quality each one experiences while driving through an urban area. It is meant to give AI researchers the joint mobility-plus-network data needed to predict roadside-network congestion a few seconds ahead, and it shows that adding trajectory data improves those load predictions.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Case study uses nearest-RSU association (Eq. 1) without showing it matches the recorded cell ID; if Simu5G used SINR/RSRP-based association, the >90% R2 claim loses its meaning.","rationale":"The reader's weakest assumption is exactly the concern I find most load-bearing. The paper is a resource-placement paper; its central scientific claim is that the dataset is uniquely useful and that the case study demonstrates this. Equation (1) is a strong, testable assumption about the ground-truth association labels. If it is wrong, the case study's headline number cannot be interpreted as evidence for the dataset's value. The other issues—validation circularity, missing URLs, arithmetic typos, missing error bars—are real but either admitted (internal consistency) or fixable without changing the core argument. The association mismatch is the one that would invalidate the central demonstration if it lands. I also agree with the reader's CONDITIONAL verdict: a conditional acceptance with a request to clarify/verify this association rule is the right level. The proposed test is cheap and decisive: compute the mismatch rate from the released data or simulator logs. If it is negligible, the concern is resolved; if not, the case study must be corrected.","tokens_in":9154,"tokens_out":4288,"duration_ms":42219,"concrete_test":"On the released dataset (or by instrumenting Simu5G), compute the fraction of vehicle–timestep records where the recorded cell ID differs from argmin_r ||p_i(t) − p_RSU_r||. If the mismatch rate is above ~1%, re-run the case study replacing Eq. (1) with the recorded cell ID (or an SINR/RSRP-based association rule matching the simulator's configuration) and check whether the >90% R2 result persists. Also report the Simu5G cell-selection configuration used to generate the dataset.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's value demonstration rests on the RSU load prediction case study (Sec. VI). Equation (1) defines the future RSU association as the nearest RSU in Euclidean distance, and Eq. (2) computes predicted load from that assignment. But the dataset's recorded 'RSU association (cell ID)' (Sec. III.B) is produced by Simu5G, which typically selects the serving cell based on received power/SINR, not proximity. The paper never states the simulator's cell-selection rule, and the only mention of association validation (Sec. III.C) says 'checks RSU associations based on distance and line-of-sight constraints' without specifying the actual rule used to generate the ground truth. With Nakagami fading and a 1,000 m RSU grid, the recorded association can differ from the nearest-RSU rule at a non-negligible fraction of records, especially near cell edges. If that is the case, then even perfect trajectory predictions cannot reproduce the true RSU load under Eq. (2), and the reported >90% R2 would be an artifact of evaluating against a self-generated label rather than the dataset's ground truth. This is load-bearing because the strongest claim explicitly includes the case study as evidence that trajectory information improves load forecasting.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"MINT-V2X is a simulated V2X dataset produced by coupling SUMO traffic simulation with OMNeT++/Simu5G network simulation. The paper reports 9.87 million synchronized records from 1,386 vehicles and 15 RSUs at 10 Hz, with per-vehicle trajectory, kinematic, physical-layer (SINR, PDR, CQI), RSU association, and performance features. The authors propose a 14-point validation framework based on 3GPP, ETSI, IEEE, and Shannon-theoretic references, and demonstrate high correlations among network metrics. A case study on RSU load prediction is used to argue that adding predicted trajectory information improves load forecasting, with the full model reportedly explaining more than 90% of RSU load fluctuations. The dataset, SUMO configuration files, and experiments are released for reproducibility.","tokens_in":9302,"tokens_out":5790,"duration_ms":60396,"significance":"If the claims are supported, MINT-V2X would be a valuable community resource: it is among the first publicly released, time-synchronized, 10 Hz trajectory-plus-physical-layer V2X datasets at moderate scale, and the release of configuration files and a generation pipeline is a concrete reproducibility contribution. The dataset artifact itself appears internally consistent, and the paper honestly acknowledges several limitations in Section V, including the absence of field-measurement validation. However, the central 'validation' evidence is substantially circular, and the load-prediction case study may evaluate against a synthetic association rule rather than the dataset's recorded ground truth. These issues do not invalidate the dataset itself, but they weaken the paper's strongest empirical claims and require correction before the contribution can be fully assessed.","major_comments":[{"comment":"The case study defines the future RSU association as the Euclidean nearest RSU (Eq. 1) and computes RSU load from that assignment (Eq. 2). The dataset records an 'RSU association (cell ID)' feature (Section III.B) generated by Simu5G, but the paper never states the simulator's serving-cell/handover selection rule. Section III.C only says that connectivity validation 'checks RSU associations based on distance and line-of-sight constraints,' which does not establish that the logged cell ID equals the nearest-RSU rule. If Simu5G uses RSRP/SINR-based cell selection with Nakagami fading, the recorded association can differ from the Euclidean-nearest rule, especially near cell edges. In that case, Eq. (2) is not measuring prediction error against the dataset's ground-truth association, and the >90% R2 claim in Section VI.B becomes an artifact of the paper's own evaluation rule. The authors sho","section":"§VI.A, Eq. (1)–(2); §III.B; §III.C"},{"comment":"The headline validation metrics are largely internal-consistency checks, not independent validations. The CQI–SINR correlation of 0.993 is a direct consequence of the deterministic SINR-to-MCS/CQI mapping that the paper itself describes in Section II.C. The SINR–PDR correlation of 0.946 couples two outputs of the same simulated SINR/channel model. The fitted path-loss exponent n=1.991 in Section III.C.2 is obtained by regression on data generated by the same log-distance path-loss formula used to create the dataset. Several rows in Table II simply restate simulator-configured ranges (e.g., SINR range '-5 to +25 dB' is labeled 'Shannon theory' without any theoretical basis for that interval). The paper's own Section V states that the framework 'does not constitute field-measurement validation,' which is an important limitation, but this caveat is not reflected in the abstract's validation","section":"§IV.A, Table II; §II.C; §III.C.2"},{"comment":"The dataset-size statement is ambiguous. The abstract says '9.87 million synchronized data points from 1,386 vehicles during 3 hours of urban traffic simulation,' and Section III.B says 'one record per vehicle per timestep in the simulation.' If every one of the 1,386 vehicles were logged for the full 10,800 s at 10 Hz, the total would be approximately 149.7 million records, not 9.87 million. The reported total is consistent with an average per-vehicle presence of roughly 712 s, but the paper does not state the mean/median track duration or the distribution of vehicle lifetimes. Readers need this information to interpret the temporal coverage and to assess whether the dataset is truly '3 hours' of dense V2X data or a much sparser trace. Please add explicit statistics on vehicle presence duration and the exact record-count derivation.","section":"§IV.B; §III.B"}],"minor_comments":[{"comment":"The text labels the log-distance path-loss formula as 'Friis (1946)' but the Friis equation is the free-space transmission formula with exponent 2; the log-distance model with an empirical exponent n is a generalization. Consider clarifying the terminology.","section":"§III.C.2"},{"comment":"The outlier detection step says speed changes exceeding 5 m/s between consecutive timesteps are 'flagged for removal,' but it is not stated whether the released dataset already excludes these samples or whether users must apply the filter. Please specify the final dataset-handling policy.","section":"§III.C"},{"comment":"The case-study section does not specify the model architecture, input/output dimensions, hyperparameters (beyond 400 epochs), or the exact definition of R2 used. Figure 4 lacks axis labels and confidence intervals. This makes the case study difficult to reproduce and the reported >90% R2 unverifiable without the code.","section":"§VI.B, Fig. 4"},{"comment":"The '29 features' are enumerated in prose but not presented as a schema table. A table with feature names, units, and types should be included in the paper or in the repository documentation.","section":"§III.B"},{"comment":"Several 'expected ranges' appear to be self-defined rather than standard-derived. For example, the SINR–PDR correlation threshold '>0.8 (strong)' is arbitrary, and the 'SINR range -5 to +25 dB' is not a Shannon-theoretic bound. The table would be more credible if each row cited a concrete standard formula or specified the simulator's configuration.","section":"Table II"},{"comment":"The text says 'SINR–PDR curve is a sigmoid curve, as expected by Shannon theory.' Shannon theory relates SINR to capacity, not directly to PDR for a fixed MCS; the sigmoid shape comes from bit-error-rate and MCS-threshold behavior. Please rephrase to avoid a technically inaccurate attribution.","section":"§IV.A"},{"comment":"Nakagami fading is mentioned but no fading parameter (e.g., m) is given. Please report the exact Nakagami parameters used in Simu5G, or refer to the released configuration file.","section":"§III.A.3"}],"recommendation":"major_revision","confidential_remarks":"The dataset release is genuinely useful and the reproducibility effort is commendable. The two main issues are correctable within the manuscript's scope: (1) clarifying and/or fixing the RSU-association rule in the case study, and (2) reframing the validation framework as internal consistency rather than standards-based external validation. The very high correlations and the >90% R2 claim should be downgraded until those points are addressed. I would not reject the paper; the underlying artifact is likely a valuable contribution, but the current presentation overstates what has been demonstrated."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the dataset is a real contribution and the paper deserves a look, but the 14-point validation should be read as internal-consistency checking, not standards validation, and the >90% R2 claim should not be quoted until the cell-association question is settled.\n\nWhat's new: the combination of 10 Hz per-vehicle kinematics plus SINR/CQI/PDR/RSU association in one time-synchronized record at 9.87M rows is not something I've seen in the public datasets the paper compares against. The record count is internally consistent once you account for mean vehicle presence (~712 s). The SUMO/Veins/Simu5G pipeline is standard but the release of configuration files is a step in the right direction, and the limitations section honestly concedes that the validation is internal consistency.\n\nThe soft spots are real but uneven. The validation tests are circular in the ways the reader flagged: CQI is a deterministic mapping from SINR, PDR is a link-layer output of the same SINR, and the path-loss exponent comes from regressing on data that the path-loss model itself produced. Table II's expected ranges are wide enough that passing them means almost nothing. There's a spawn-rate arithmetic error: 1.1 vehicles/s over 10,800 s gives ~11,880 vehicles, not 1,386; the actual average is about 0.13/s. The GitHub URL is promised but not given. These are fixable.\n\nThe bigger issue is the case study. Equation (1) assumes future RSU association is the nearest RSU, but the paper never states what rule Simu5G used to produce the recorded cell ID. With Nakagami fading and a 1 km grid, RSRP-based association will differ from the nearest-RSU rule at a nontrivial fraction of records, especially near cell edges. If the ground-truth load is built from a different association function than the prediction rule, then the >90% R2 is partly evaluating the model against a surrogate label. The case study also reports no architecture, seeds, or error bars, so it's not independently checkable. This is the one concern that could change how much weight the paper's main scientific claim deserves.\n\nWho this is for: researchers who need a public benchmark for mobility-aware V2X resource prediction and are willing to work with a simulated dataset. It deserves a serious referee, but the revision must state the simulator's association rule, fix the spawn-rate typo, provide the repository link, and soften the validation claims from \"standards-based\" to \"internal-consistency.\" If those changes land, the dataset becomes a genuinely useful resource.","headline":"A genuinely new integrated V2X dataset, but the validation is mostly internal consistency and the case study's headline R2 depends on an association rule that may not match the simulator.","tokens_in":10028,"tokens_out":2895,"would_cite":true,"duration_ms":27209,"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":"MINT-V2X claims to be the first publicly available dataset that fuses 10 Hz vehicle trajectories with per-link wireless metrics (SINR, PDR, CQI) and RSU association, and shows that adding predicted trajectories lets a model explain more tha","keywords":["V2X dataset","trajectory–network fusion","RSU load prediction","C-V2X","SUMO","Simu5G","physical-layer metrics","predictive resource management"],"falsifier":"Compute, from the released dataset, the fraction of timesteps in which the recorded RSU association (cell ID) equals the nearest-RSU assignment from vehicle coordinates. If that fraction is far below 100% in high-fading regions, then Eq. 1 does not reproduce the dataset's ground-truth associations, and the load-prediction result is an artifact of the evaluation rule rather than of trajectory information.","tokens_in":8827,"feed_emoji":"📡","tokens_out":5219,"duration_ms":45784,"temperature":0.7,"pith_summary":"This paper tries to close a data gap in V2X research: existing public datasets offer either detailed vehicle trajectories or network statistics, but not both in one time-synchronized record. The authors generate MINT-V2X by coupling a traffic simulator with a network simulator, producing 9.87 million records from 1,386 vehicles at 10 Hz, with 29 features per record spanning kinematics, physical-layer metrics, and RSU association. They argue the dataset is realistic by passing 14 validation tests tied to 3GPP, ETSI, and Shannon theory, with strong correlations such as CQI–SINR of 0.993 and SINR–PDR of 0.946. A case study on RSU load prediction reports that adding predicted trajectory features outperforms load-only and communication-only baselines, reaching an R² above 0.9. If these claims hold, the dataset would support predictive resource management that anticipates handovers and load changes from mobility.","feed_headline":"Trajectory data lifts V2X load prediction past 90 percent","feed_subtitle":"New public dataset pairs vehicle kinematics with SINR, PDR, and CQI at 10 Hz, enabling mobility-aware network forecasting.","key_machinery":"The central mechanism is a three-layer co-simulation chain: a traffic simulator produces vehicle motion, middleware bridges it through a control interface, and a network simulator computes wireless metrics per 100 ms timestep; every timestep yields a 29-feature record aligned by vehicle ID and timestamp. The load-prediction case study uses a nearest-RSU association rule (Eq. 1) to convert predicted positions into future RSU assignments, then aggregates per-vehicle demand into predicted RSU load (Eq. 2), in a joint loss that combines trajectory, per-vehicle, and RSU-level objectives.","core_discovery":"The dataset MINT-V2X combines, for the first time at this scale, per-vehicle trajectory kinematics and physical-layer V2X metrics (SINR, CQI, PDR, received power, channel busy ratio) together with the serving RSU identity, sampled synchronously at 10 Hz over three hours of urban traffic in a 61.19 km² area with 15 RSUs. The authors show internal consistency between metrics and compliance with 14 validation checks. In the load-prediction case study, a model that takes predicted trajectories as features explains over 90% of RSU load variation on a held-out temporal split, outperforming models that use only historical load or link-quality metrics.","pith_inferences":["The strong CQI–SINR and SINR–PDR correlations may reflect the same simulation code generating both sides, so they certify algorithmic consistency rather than real-world fidelity; a field test would be needed to anchor the dataset to reality.","Because the simulator's RSU association rule is not specified, the case study's nearest-RSU assumption may not match the recorded cell IDs; a check of how often recorded association equals nearest-RSU would determine whether the 90% R² is a property of the dataset or of the evaluation rule.","The dataset could be extended with building shadows, multipath, and non-LOS links to move from the idealized path-loss exponent (n≈2.0) toward realistic urban values (n≈2.7–3.0), which might change the load predictions.","If association is signal-strength-based rather than distance-based, then trajectory data alone may not be sufficient to forecast load; one would also need to model fading and received power."],"forward_implications":["MINT-V2X would let researchers train models that map predicted vehicle positions directly to future RSU load and handover events.","It would provide a reproducible baseline for benchmarking predictive resource management algorithms on synthetic urban C-V2X scenarios.","It suggests that trajectory features, not just historical load, carry predictive signal for network congestion, motivating mobility-aware scheduling.","The 14-point validation framework could serve as a template for auditing other simulation-generated V2X datasets.","The scale of 9.87 million records supports deep-learning models that need large synchronized trajectory–network samples."],"fun_headline_variants":["First V2X dataset pairing vehicle trajectories with network metrics","MINT-V2X: 9.87M synced points for predictive V2X resource management","Trajectory features beat network-history baselines in V2X load prediction","Integrated V2X dataset: kinematics + SINR, CQI, PDR at 10 Hz","MINT-V2X: trajectories beat network-only baselines in V2X load forecast"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The paper's case study assumes the RSU a vehicle connects to is simply the nearest RSU, but the paper never states that the simulator's recorded serving-cell selection follows this rule.","fun_headline_variants_meta":{"raw":{"variants":["First V2X dataset pairing vehicle trajectories with network metrics","MINT-V2X: 9.87M synced points for predictive V2X resource management","Trajectory features beat network-history baselines in V2X load prediction","Integrated V2X dataset: kinematics + SINR, CQI, PDR at 10 Hz","MINT-V2X: trajectories beat network-only baselines in V2X load forecast"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000608,"raw_usage":{"total_tokens":2695,"prompt_tokens":796,"completion_tokens":1899,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":540,"completion_tokens_details":{"reasoning_tokens":1788}},"tokens_in":540,"tokens_out":1899,"duration_ms":14321,"temperature":1.0,"reasoning_tokens":1788,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T09:48:19.159327+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute, from the released dataset, the fraction of timesteps in which the recorded RSU association (cell ID) equals the nearest-RSU assignment from vehicle coordinates. If that fraction is far below 100% in high-fading regions, then Eq. 1 does not reproduce the dataset's ground-truth associations, and the load-prediction result is an artifact of the evaluation rule rather than of trajectory information.","supporting_citations":[],"review_version":1}