{"id":"69bd7789-022d-460a-9ac1-dee8c87cc739","arxiv_id":"2508.01671","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":4.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":1,"one_line_summary":"The advertised drone planning framework is absent from the supplied text, which instead contains an LLM personalization benchmark.","lead":"This submission's title and abstract propose an energy-predictive drone delivery framework using Bi-LSTM forecasts and heuristic routing. The supplied full text is an unrelated paper on LLM contextual preference inference, so the advertised drone contribution cannot be assessed from this file.","discovery_kind":"unclear","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The supplied full text is a different paper (CUPID, COLM 2025), so the EPDS drone-planning central claim has no supporting methods, experiments, or data in this submission.","rationale":"The reader's verdict is UNVERDICTED with low confidence, citing both the abstract/body mismatch and unverifiable assumptions about Bi-LSTM accuracy and dataset representativeness. I agree with the mismatch as the decisive finding, but I would frame it more strongly: the body is not a partial or incomplete version of the drone paper; it is an entirely different paper on LLM contextual preference benchmarking. The reader's 'weakest assumption' about prediction accuracy and dataset suitability is a plausible concern for the advertised EPDS paper, but it is secondary given that the submitted text contains no EPDS model at all. Therefore the load-bearing concern is not a subtle modeling assumption but the absence of the claimed contribution itself. The recommendation remains UNVERDICTED: the central claim cannot be supported or refuted from the supplied artifact. I do not recommend REJECT because the mismatch may stem from an arXiv submission/retrieval error; the correct drone paper may exist elsewhere. A concrete check of the arXiv record would settle whether this is a corrupted or mislabeled submission. The verdict should stay UNCHANGED relative to the reader's UNVERDICTED, since my analysis reinforces rather than overturns it.","tokens_in":31172,"tokens_out":1378,"duration_ms":18431,"concrete_test":"Download the arXiv source for 2508.01671 (or its abs page) and search the full text for the strings 'EPDS', 'Bi-LSTM', 'skyway', 'drone', and 'recharging'. If the paper body contains none of these terms and instead matches the CUPID/COLM 2025 LLM benchmark paper, the central claim is unsupported. If a corrected full text with the EPDS sections is found, then re-evaluate the claimed results against the actual dataset and optimizer details.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract and title promise an Energy-Predictive Drone Service (EPDS) framework: a formal EPDS model, an adaptive Bi-LSTM predictor of other drones' energy status and stochastic arrival times, a heuristic optimizer for path and recharging schedules, and evaluation on a real-world drone flight dataset. The full text provided is instead the CUPID paper on contextual preference inference for LLMs, with its own abstract, experiments, and appendices. None of the EPDS components appear anywhere in the body: there is no formal model, no Bi-LSTM architecture or training procedure, no heuristic optimization formulation, no skyway network model, and no description of the real-world drone dataset or evaluation protocol. The central claim that EPDS 'identifies the most time-efficient and energy-efficient skyway path and recharging schedule' therefore has no evidentiary support in this manuscript. This is not merely a missing baseline or an unstated assumption; the submitted artifact is internally inconsistent at the level of the paper's identity. The load-bearing condition for the central claim is that the text actually contains the EPDS framework and its validation. That condition fails. Consequently, no assessment of predictive accuracy, optimization quality, or dataset representativeness is possible from this submission, and the claim must be treated as unverified.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript, arXiv:2508.01671, presents an abstract proposing an Energy-Predictive Drone Service (EPDS) framework for drone package delivery in a skyway network. The abstract describes a formal EPDS model, an adaptive Bi-LSTM predictor of other drones' energy status and stochastic arrival times, a heuristic optimizer for path and recharging schedules, and validation on a real-world drone flight dataset. However, the supplied full text is a completely different paper: the COLM 2025 paper 'CUPID: Evaluating Personalized and Contextualized Alignment of LLMs from Interactions', which concerns benchmarking LLMs' contextual preference inference. None of the EPDS components, equations, experiments, dataset descriptions, or results described in the abstract appear anywhere in the full text. The submitted artifact is therefore internally inconsistent, and the central claim of the EPDS framework is entirely unsupported by the provided manuscript.","tokens_in":31378,"tokens_out":1452,"duration_ms":18069,"significance":"If the EPDS framework as described in the abstract were actually present and validated, its contribution could be significant for drone logistics in shared skyways, combining predictive modeling of other agents' energy states with joint path and recharging optimization. The framework's use of a real-world drone flight dataset and its promise of concrete efficiency gains would merit attention. As submitted, however, the paper contains none of its own methods, experiments, or data. The abstract alone is insufficient to establish technical validity, and the full text is devoted to an unrelated LLM benchmark. Consequently, the significance of the actual submission cannot be assessed, and it is not publishable in its current form.","major_comments":[{"comment":"The full text provided is the CUPID paper on LLM contextual preference inference (COLM 2025), not the EPDS drone-planning paper advertised in the abstract and title. The abstract promises a formal EPDS model, a Bi-LSTM predictor, a heuristic optimizer, and experiments on a real-world drone flight dataset, but the manuscript body contains no such content: there are no equations, no Bi-LSTM architecture or training details, no optimization formulation, no skyway network model, and no evaluation results. This is a load-bearing failure: the central claim that EPDS 'identifies the most time-efficient and energy-efficient skyway path and recharging schedule' has no evidentiary support in the submitted text.","section":"Full text (entire manuscript body)"},{"comment":"The abstract states that 'extensive experiments using a real-world drone flight dataset' were conducted, yet no experimental setup, baselines, metrics, or quantitative results are presented anywhere in the manuscript. Even if the full-text mismatch were overlooked, the abstract alone cannot support the claimed performance of the proposed framework, and there is no way for a reader to assess predictive accuracy, optimization quality, or dataset representativeness.","section":"Abstract, claims of validation"},{"comment":"The manuscript's internal inconsistency—an abstract and title about drone service planning followed by a full text about LLM personalization—makes it impossible to review in the standard manner. This is not a local flaw or missing comparison; it is a mismatch at the level of the paper's identity. The issue cannot be fixed within the scope of the submitted text without replacing essentially the entire manuscript with the EPDS content.","section":"Paper identity and scope"}],"minor_comments":[{"comment":"If the authors intend to resubmit the EPDS paper, the abstract should state the specific evaluation baselines and the size and source of the real-world drone flight dataset, as these details are currently absent.","section":"Abstract"}],"recommendation":"reject","confidential_remarks":"This appears to be a submission error in which the full text of a different paper (CUPID, COLM 2025) was uploaded under the arXiv ID for the drone-planning paper. The editor may wish to verify the submission files with the authors. Because the provided manuscript is internally inconsistent and the advertised EPDS framework is entirely absent, rejection is the appropriate recommendation; if a corrected full text becomes available, a fresh review would be required."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this one is not reviewable as submitted. The title and abstract describe EPDS, a drone delivery planning framework with Bi-LSTM energy prediction and heuristic path/recharge optimization, evaluated on a real-world flight dataset. The body of the manuscript, however, is a completely different paper: CUPID, a COLM 2025 benchmark on contextual preference inference for LLMs, by different authors. None of the EPDS components—the formal model, the Bi-LSTM architecture, the optimizer formulation, the skyway network abstraction, or the dataset—appear anywhere in the supplied text. The drone claim is not merely under-supported; it is absent.\n\nWhat the paper does well: nothing to do with the abstract. If you take the CUPID text on its own terms, it is a solid piece of benchmark work: 756 human-curated interaction histories, a data-generation pipeline with human validation, careful LLM-as-judge meta-evaluation with Krippendorff's alpha, and public release of data and code. But those authors and that result are not the submission we were asked to assess. The mismatch is at the level of paper identity, not a missing appendix or a weak baseline.\n\nSoft spots: the decisive one is the empty central claim. There is no way to check the EPDS model's soundness, novelty, or experimental support because none of it is here. Even the abstract's \"novel\" framing can't be tested—energy-aware drone routing with ML-based prediction is an established area, but we can't compare this contribution to prior work since its methods are missing. There is also no information on the real-world dataset, so any assessment of generalizability or representativeness is impossible.\n\nWho this is for: nobody, until the authors resubmit with the correct full text. The CUPID paper itself would be of interest to people working on LLM personalization, but it's a different paper for a different venue. As it stands, this submission is a file/metadata error in search of a desk reject.\n\nRecommendation: do not send to peer review. Return it to the authors so they can fix the PDF or the arXiv listing, and ask them to confirm which paper they intended to submit. If the drone paper actually exists in proper form, it could be a legitimate within-subfield contribution worth a serious referee—but that version is not what's in front of us.","headline":"The submission's abstract promises an energy-predictive drone routing framework, but the full text is an unrelated LLM benchmark paper, so the central claim has no supporting content at all.","tokens_in":31870,"tokens_out":2683,"would_cite":false,"duration_ms":29324,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that predicting other drones' energy status and arrival times with a Bi-LSTM, then planning each drone's path and recharging schedule from those predictions, makes shared-skyway drone delivery more time-efficient and…","keywords":["drone delivery","skyway network","energy prediction","Bi-LSTM","recharging schedule","path optimization","composite drone services","stochastic arrival times"],"falsifier":"Hold the optimizer fixed and run the same skyway fleet twice, once planning from predicted energy status and arrival times and once from the drones' currently reported battery levels and positions: if total delivery time and total energy are not reliably lower in the predicted case, the forecasting component is not carrying the gain. A second check is to compare each drone's predicted and realized energy at each leg and see whether the forecast error is smaller than the scheduling slack.","tokens_in":30991,"feed_emoji":"📦","tokens_out":9366,"duration_ms":96903,"temperature":0.7,"pith_summary":"The paper proposes the Energy-Predictive Drone Service (EPDS) framework, which claims that package deliveries through a shared skyway network become more time-efficient and energy-efficient when the planner does not merely react to the current positions and battery levels of other drones. An adaptive bidirectional Long Short-Term Memory (Bi-LSTM) model forecasts the energy status and stochastic arrival times of every drone sharing the network, and a heuristic optimizer chooses each drone's skyway path and recharging schedule from those forecasts. If the framework works as claimed, fleet operators could pack more deliveries into the same airspace while spending less energy per package, and recharges could be scheduled before battery emergencies rather than after them. The paper reports validating the framework with extensive experiments on a real-world drone flight dataset.","feed_headline":"Predicting drones' energy use picks faster, cheaper skyway routes","feed_subtitle":"A forecast-driven planner routes delivery drones and schedules recharges to save time and energy.","key_machinery":"The load-bearing mechanism is a two-stage pipeline. Stage one is an adaptive bidirectional Long Short-Term Memory (Bi-LSTM) model, a recurrent neural network that reads a drone's flight and energy history in both temporal directions and outputs predictions of that drone's future energy status and stochastic arrival times, with 'adaptive' indicating that the model tracks changing network conditions. Stage two is a heuristic optimizer for composite drone services that consumes these predictions and emits, for each drone, a skyway path and a recharging schedule. The conceptual move that carries the argument is treating other drones' future states as quantities to be predicted rather than as noise to be robust against, so that path and recharge decisions are made around forecast contention and forecast low-battery events.","core_discovery":"On the paper's own terms, the central claim is that efficient operation of a shared skyway network is a joint predictive-planning problem, not a per-drone reaction problem. An adaptive Bi-LSTM model learns the energy status and stochastic arrival times of the other drones in the network, and a heuristic optimization approach for composite drone services converts those predictions into the most time-efficient and energy-efficient skyway path and recharging schedule for each drone. The real-world drone flight dataset is offered as evidence that the framework's plans hold outside synthetic settings. Stated sympathetically: a drone fleet operates best when each drone's plan is chosen in view of where the other drones' batteries and arrivals are forecast to be, not where they were at the last report.","pith_inferences":["Note on provenance: the full text supplied with this submission is a different paper, a benchmark for contextual user-preference inference in LLMs, and contains none of the EPDS framework, its dataset, or its experiments; everything above therefore rests on the abstract and metadata alone, and the experimental claims could not be checked against the body.","The same predict-then-plan pattern should transfer to other shared-resource fleets, such as electric truck fleets contending for charging stations or warehouse robots sharing a floor, wherever one agent's future resource state is readable in its history.","An ablation that swaps the Bi-LSTM for a naive predictor, for example assuming each drone's current battery level simply persists, would isolate whether the learning or the optimizer carries the gain.","If the forecasts are good enough, the practical upside is not just lower cost per delivery but higher skyway throughput, allowing operators to accept more delivery commitments per day."],"forward_implications":["Planners could reroute drones before mid-air battery emergencies develop, because forecasts flag which parts of the skyway will be short on energy.","Recharging becomes a scheduled, cost-aware decision across the fleet instead of an emergency response, which is where the energy savings come from.","Because path and recharge choices are optimized jointly, multi-leg composite deliveries are planned as a whole rather than leg by leg.","Successful validation on a real-world flight dataset is the paper's evidence that these gains carry over from the model to actual skyway operations."],"supporting_citations":[],"fun_headline_variants":["Forecast drone energy to choose smarter skyway routes","Drone delivery planner uses AI to predict energy, pick fast routes","Bi-LSTM forecasts drone energy for optimal skyway scheduling","Predict drone energy to plan faster, cheaper delivery routes","Energy-aware drone routing with arrival-time prediction"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The plan helps only if the forecast of what other drones will do is accurate enough to beat planning from current, known information, and if the real flight data used for testing represents actual skyway operating conditions — the abstract gives no prediction-error numbers and no dataset description to confirm either.","fun_headline_variants_meta":{"raw":{"variants":["Forecast drone energy to choose smarter skyway routes","Drone delivery planner uses AI to predict energy, pick fast routes","Bi-LSTM forecasts drone energy for optimal skyway scheduling","Predict drone energy to plan faster, cheaper delivery routes","Energy-aware drone routing with arrival-time prediction"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000688,"raw_usage":{"total_tokens":3042,"prompt_tokens":795,"completion_tokens":2247,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":411,"completion_tokens_details":{"reasoning_tokens":2169}},"tokens_in":411,"tokens_out":2247,"duration_ms":17866,"temperature":1.0,"reasoning_tokens":2169,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T05:26:06.056726+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Hold the optimizer fixed and run the same skyway fleet twice, once planning from predicted energy status and arrival times and once from the drones' currently reported battery levels and positions: if total delivery time and total energy are not reliably lower in the predicted case, the forecasting component is not carrying the gain. A second check is to compare each drone's predicted and realized energy at each leg and see whether the forecast error is smaller than the scheduling slack.","supporting_citations":[],"review_version":1}