Synthetic Intermittent Demand Benchmark

Synthetic Spare-Parts Demand Benchmark (v2) A fully synthetic, masked replication benchmark for intermittent spare-parts demand forecasting and inventory evaluation. It reproduces the structure of a confidential industrial panel without disclosing any real data: all identifiers and categorical fields are tokenized to neutral labels, MAP prices are drawn from a calibrated distribution (not real prices), and demand values are synthetic surrogates. SeePRIVACY.md. 5,000 synthetic materials × 60 monthly periods (2018-09 … 2023-08). v2 note: The paper was submitted with the final name: "Accuracy Is Not Service: A Decision-Aware Benchmark for Intermittent-Demand Forecasting". The generation technique received a small patch to better reproduce the real data: a fully parametric occurrence×size generator (each material drawn from a real SKU's fitted count-family by inverse-CDF, with a per-platform latent factor for cross-material tail dependence and a per-SKU non-stationarity). The volume and schema are unchanged; only the generation method and the fidelity scores differ.See DATASHEET.md. Files File Contents synthetic_demand_matrix_5y.csv monthly demand panel (material_id × month) synthetic_material_metadata_5y.csv per-material demand statistics (Syntetos–Boylan class, ADI, CV², run-lengths, …) synthetic_material_catalog_5y.csv selection-ready catalogue (statistics, activity bands, fidelity pass) synthetic_context_materials_5y.csv per-material categorical context (neutral category_* labels) synthetic_full_data_generic_5y.csv order-line event log with neutral column names synthetic_map_prices.csv synthetic MAP prices (material_id, map_price) synthetic_inventory_planning_params_5y.csv derived inventory planning parameters synthetic_material_fidelity_5y.csv per-material fidelity audit (real source IDs withheld) period_lookup_5y.csv period index → calendar month generate_synthetic_replicable_dataset_v2.py, _inverse_cdf_families.py the generator (procedural transparency; needs the confidential catalogue to run) DATASHEET.md, REPRODUCE.md, PRIVACY.md, CITATION.cff, .zenodo.json, checksums.sha256 metadata + integrity All column names are generic or neutral; none reproduce the original system's schema. License: CC-BY-4.0. Accompanies the paper "Accuracy Is Not Service: A Decision-Aware Benchmark for Intermittent-Demand Forecasting".