{"id":"921778b7-57d6-473c-80ab-266d3541aa6c","arxiv_id":"2602.03521","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"FeederBW provides two years of 1-minute measurements from 200 German low-voltage feeders with weather and customer-equipment metadata, filling a gap in public LV grid data.","lead":"FeederBW is a new public dataset of 200 real-world low-voltage grid feeders in Germany, with one-minute power, current, and voltage measurements over two years (2023–2025), plus weather data and detailed metadata on solar panels, heat pumps, EV chargers, and housing. It gives energy researchers a rare large-scale resource for training machine-learning models and studying how low-carbon technology changes the grid.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified","rationale":"The reader's verdict is ACCEPT with high confidence. I read the full text and found the data descriptor to be transparent and internally consistent. The metadata accuracy issue is the most plausible weak point, but it is already thoroughly acknowledged; the paper explicitly invites users to work on metadata-error correction. No formal verification exists, but the dataset itself is the falsifiable artifact. A direct download-and-validate test would settle the central claim. Therefore I recommend keeping the original verdict.","tokens_in":15537,"tokens_out":9780,"duration_ms":99917,"concrete_test":"Download the three data records from the Zenodo DOI 10.5281/zenodo.17831177; verify that the feeder measurement parquet files sum to 209,189,649 rows and 20 columns, the metadata CSV has 1,608 rows and 22 columns, and the weather parquet has 3,508,800 rows and 12 columns, and that the column names and units match Tables 3-5. If any count mismatches by more than a few rows, the central existence/description claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper is a data descriptor whose central claim is that a specific dataset exists, is publicly accessible, and has the described properties. The weakest link in that claim is the accuracy of the feeder metadata, which the reader correctly identifies. However, the paper itself explicitly and in detail discloses the metadata limitations (assignment inaccuracies, unregistered equipment, registration/operation date mismatches, undocumented switching events) in the 'Limitations in the feeder metadata' section. These are honest caveats rather than hidden flaws; they reduce interpretability for some uses but do not falsify the existence claim or the core description. The measurement data limitations (single-phase voltage, 3% accuracy, approximate power computation) are likewise disclosed. I find no internal inconsistency or unacknowledged unsupported assumption that would threaten the central claim. The remaining verification step is purely empirical: confirming the Zenodo deposit matches the stated row counts and schemas.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper describes FeederBW, a public dataset of real-world measurements from 200 low-voltage feeders in Baden-Württemberg, Germany, over two years (April 2023–March 2025). The dataset comprises 209,189,649 rows of one-minute feeder measurements (voltage, phase currents, active/reactive/apparent power, power factors, PEN current), event-driven feeder metadata (housing units, installed capacities of PV, batteries, heat pumps, EV chargers, etc., and aggregated industry/commerce consumption), and hourly weather data from the ICON-D2 NWP model. The authors describe the measurement hardware and accuracy, selection criteria, data records, technical validation using quantile profiles, and detailed limitations. The dataset is available on Zenodo with a DOI, and the paper recommends train-test splits for ML applications.","tokens_in":15731,"tokens_out":13481,"duration_ms":121310,"significance":"If the dataset exists as described, FeederBW fills a clear gap in publicly available low-voltage grid data: it combines a relatively large number of feeders (200), one-minute resolution, weather data, and detailed consumer/producer metadata, which is rare in existing datasets. The paper is commendably transparent about hardware limitations (3% accuracy, single-phase voltage, PEN measurement issues), metadata error sources, and selection biases. The public Zenodo release with a DOI and the explicit train-test split recommendations are practical strengths that support reproducibility and comparability of ML studies. No code is provided, and the technical validation is qualitative, but the availability of the data itself makes this a useful community resource.","major_comments":[],"minor_comments":[{"comment":"The definition of λ_tot is garbled in the sentence: 'filtered based on the total power factor λ_tot = P λL1 +λL2 +λL3'. It should presumably read λ_tot = λ_L1 + λ_L2 + λ_L3. In the same paragraph, the active currents for feeder F127 are listed as I_L1,act = −43 A, I_L2,act = −48 A and I_L1,act = 84 A; the third current should be I_L3,act. Please correct these typographical errors.","section":"Limitations in the feeder measurement data"},{"comment":"The text contains several wording errors: 'this is not necessary the case' should be 'not necessarily the case'; 'for shops and hairdresser' should be 'shops and hairdressers'; 'F181 has only around two third' should be 'two thirds'. These are minor but should be fixed.","section":"Limitations in the feeder metadata"},{"comment":"The sentence 'Null values only occur in columns now height m' contains a typo: 'now height' should be 'snow height'. Also consider rephrasing the following sentence 'In applications and models you can assume 0 m' to a more formal style, e.g., 'A value of 0 m can be assumed'.","section":"Limitations in the weather data"},{"comment":"The validation is largely qualitative, based on four weekly quantile plots and two time-series plots. The statement 'we conducted extensive checks and validation to all of them' is not backed by concrete numbers. Consider adding a small table of summary statistics (e.g., per-feeder ranges of active power, missing-data rates) or a short quantitative description of the checks performed on current, voltage, and power-factor plausibility. This would strengthen the data quality assessment without requiring exhaustive validation.","section":"Technical Validation"},{"comment":"The weather data are outputs of the numerical weather prediction model ICON-D2, not ground-based station measurements. This is stated in the Methods, but readers may overlook it in Table 1 and the Data Records. Please explicitly note in the weather data section or as a footnote to Table 1 that these are NWP model outputs, not direct observations.","section":"Methods / Data Records"},{"comment":"There are several language issues in this section: 'a obligation' should be 'an obligation', 'prized' should be 'priced', and 'shortly summarize' should be 'briefly summarize'. Also, the sentence 'The data is collected after accounting for grid losses in the LV grid' is ambiguous; if the measurements are at the feeder, they include network losses, so consider rephrasing to clarify the relationship between measured values and customer consumption.","section":"Further background information"}],"recommendation":"minor_revision","confidential_remarks":"The central claim of this data descriptor is the existence and description of the FeederBW dataset. I could not independently verify that the Zenodo deposit matches the reported row counts and schemas; this remains an empirical check for the editors. The paper's internal consistency and transparency are good, and the disclosed limitations are appropriate for a data descriptor. The 'unique' phrasing in the abstract is somewhat stronger than the comparative evidence provided, but it is not a blocking issue. I recommend minor revision to address the typographical and clarification points listed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a data paper worth taking seriously. The authors release a real, substantial dataset — 200 LV feeders in Baden-Württemberg, 1-minute measurements over two years, weather data, and event-driven feeder metadata — on Zenodo with a DOI. That is a real contribution. The field has few public grid-measured feeder datasets this size, and none with this combination of temporal resolution and metadata depth. I believe the existence claim is credible and the paper honestly describes how the data was collected.\n\nWhat the paper does well: the methods are concrete. They name the SMIGHT sensors, give the 3% accuracy, explain the energy-harvesting constraint (minimum 32 A), and state the PEN replacement policy. They report completeness (99.36% rows, 25.6% NULL PEN values explained by uninstrumented PEN). They discuss the single-phase voltage limitation and why S != sqrt(P^2+Q^2) occurs. The metadata limitations — unregistered equipment, assignment errors, registration vs operation dates, undocumented switching — are explicitly described. The recommended train-test splits are a sensible touch. The selection criteria are clear and the bias they introduce is acknowledged.\n\nNow the soft spots, in proportion. The technical validation is mostly qualitative. Weekly quantile plots of active power for four feeders show plausible patterns, but there is no systematic error analysis against, say, a calibrated reference measurement. That is not a fatal flaw for a data descriptor — the reader can verify the data themselves — but it does mean the 3% accuracy claim is taken on trust. The metadata is the weakest link. The authors say it comes from the DSO's customer management system, has event-driven entries, and can contain errors. If those errors are systematic, some of the feeder-level interpretability claims weaken. The authors do not quantify metadata accuracy. A direct comparison of registered PV capacity against measured feed-in for a subset would have strengthened the paper. Also, the 'individual data check' blacklist is opaque — we don't know what was excluded and why. That is acceptable but limits reproducibility of the selection process.\n\nI do not see a load-bearing flaw. The citation pattern is fine; self-citations to the data platform and a prior pseudo-measurement paper are relevant, not circular. The central claim — the dataset exists and is described accurately — is testable by downloading the Zenodo deposit.\n\nWho is this for: anyone doing ML on distribution grids — forecasting, NILM at the feeder level, synthetic data generation, or studies of PV/heat pump impact. Serious referee time is warranted. I would accept this into the record, with a request for more systematic validation of the power measurements and metadata if the journal wants it.","headline":"A genuinely useful public dataset of 200 German LV feeders at 1-minute resolution over two years, described by a careful, transparent data descriptor; the main risk is metadata accuracy, which the authors openly acknowledge.","tokens_in":16143,"tokens_out":2388,"would_cite":true,"duration_ms":23765,"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":"A new public dataset records 200 German low-voltage feeders at one-minute resolution over two years, with detailed metadata on solar panels, heat pumps, and EV chargers.","keywords":["low-voltage grid","feeder measurements","public dataset","photovoltaic","heat pumps","EV chargers","load forecasting","weather data"],"falsifier":"Compare the recorded installed PV capacity against the measured midday feed-in peaks for a random sample of feeders; if many feeders show persistent unexplained shortfalls or surpluses, the metadata's trustworthiness—and with it the dataset's flagship interpretability—is called into question.","tokens_in":15486,"feed_emoji":"⚡","tokens_out":2319,"duration_ms":26521,"temperature":0.7,"pith_summary":"The paper presents FeederBW, a real-world dataset of 200 low-voltage feeders in Germany measured every minute for two years (April 2023–March 2025). It combines feeder-level electrical measurements—per-phase currents, active and reactive power, voltage, and power factor—with hourly weather data and event-driven feeder metadata that tracks the installed power of low-carbon technologies. The authors argue that this combination fills a scarcity of public, grid-measured low-voltage data, which is essential for reproducible research in load forecasting, non-intrusive load monitoring, synthetic data generation, and understanding how weather and decentralized generation interact. A sympathetic reader would see this as a substantial, curated contribution to open energy data, useful for both machine learning and grid operation studies.","feed_headline":"Minute-level data from 200 German low-voltage feeders goes public","feed_subtitle":"Two years of real feeder measurements with solar, battery, heat pump, and EV metadata support load forecasting and grid studies.","key_machinery":"The dataset itself is the central object, produced by a measurement chain: clip-on current sensors on each feeder phase measure RMS currents, a gateway measures a single-phase voltage and derives active/reactive power and power factor, and a customer management system supplies equipment-level metadata, matched via zip code to hourly weather from a numerical weather prediction model. A set of selection filters (radial grid, ≥90% data completeness, privacy thresholds, capacity utilization) ensures high-quality, interpretable feeders.","core_discovery":"FeederBW is a dataset of 200 low-voltage feeders in Baden-Württemberg, Germany, containing 209,189,649 rows of one-minute feeder measurements plus hourly weather data and 1,608 rows of event-driven metadata. The measurements include voltage (one phase), active/reactive/apparent power, phase currents, and power factors, with about 99.36% completeness. The metadata records housing units, industry/commerce counts, installed power of PV, batteries, EV chargers, heat pumps, storage heaters, and other equipment, updated when customer connections change. The paper demonstrates plausible patterns—weekly load shapes, seasonal feed-in, and a clear two-year increase in installed PV and battery capacity","pith_inferences":["The metadata accuracy is the key risk: unregistered equipment, registration-to-operation delays, and topology changes not documented in the data could systematically bias any model trained on FeederBW, especially for feeders where such events are common.","Because only one phase voltage is measured and the formula S = sqrt(P² + Q²) fails under opposing phase flows, users should treat apparent power as the safer thermal indicator, a nuance the paper explicitly notes but which might be overlooked in practice.","The recommended locational and temporal splits could become a de facto benchmark, but the paper does not validate them across many algorithms; users may need to test sensitivity to split choice.","Combining FeederBW with publicly available synthetic grid topologies (not included) could enable full power-flow studies, but such integration is left for future work."],"forward_implications":["Researchers can train and benchmark load forecasting models on real, minute-level feeder data with weather context, improving reproducibility in low-voltage research.","Feeder-level non-intrusive load monitoring becomes feasible on aggregated residential/commercial loads, potentially identifying the operation of heat pumps, EV chargers, and PV systems.","The combination of weather and metadata allows studying how solar irradiance, temperature, and installed low-carbon capacity drive feeder load and feed-in patterns.","The two-year span with recorded PV installation events provides a natural experiment for estimating the grid impact of rooftop solar and battery adoption.","The dataset can serve as a realistic basis for generating synthetic low-voltage grid scenarios, supporting power flow simulations without exposing topology."],"fun_headline_variants":["FeederBW: 200 German low-voltage feeders, minute data, 2 years","Two years of minute-level grid data from 200 German feeders","Open dataset tracks low-carbon tech on 200 German low-voltage grids","200 feeders, 2 years, 1-minute resolution: German grid dataset","Minute-level measurements from 200 German low-voltage feeders"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The feeder metadata from the utility's customer management system accurately reflects the actual equipment, housing units, and industrial loads connected to each feeder, despite acknowledged unregistered devices and registration delays.","fun_headline_variants_meta":{"raw":{"variants":["FeederBW: 200 German low-voltage feeders, minute data, 2 years","Two years of minute-level grid data from 200 German feeders","Open dataset tracks low-carbon tech on 200 German low-voltage grids","200 feeders, 2 years, 1-minute resolution: German grid dataset","Minute-level measurements from 200 German low-voltage feeders"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000242,"raw_usage":{"total_tokens":1356,"prompt_tokens":731,"completion_tokens":625,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":475,"completion_tokens_details":{"reasoning_tokens":531}},"tokens_in":475,"tokens_out":625,"duration_ms":6393,"temperature":1.0,"reasoning_tokens":531,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T04:55:28.379503+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare the recorded installed PV capacity against the measured midday feed-in peaks for a random sample of feeders; if many feeders show persistent unexplained shortfalls or surpluses, the metadata's trustworthiness—and with it the dataset's flagship interpretability—is called into question.","supporting_citations":[],"review_version":1}