Synthetic Intermittent Demand Benchmark

Synthetic Spare-Parts Demand Benchmark (v1) A fidelity-certified synthetic benchmark for decision-aware forecasting of intermittent spare-parts demand, accompanying the paper "When Forecast Accuracy Hurts Service: A Cross-Family Benchmark and Bias-Direction Diagnostic for Intermittent Spare-Parts Demand." There is no public, decision-aware benchmark for genuinely intermittent spare-parts demand — existing forecasting benchmarks (M4/M5) are structurally smooth, and the industrial panels where intermittency bites are confidential. This dataset removes that barrier: it is a masked synthetic surrogate of a confidential industrial spare-parts panel, generated so that each material is regenerated until it passes a per-material fidelity audit (Syntetos–Boylan class, zero fraction, ADI, CV², zero-run distribution), and certified against the real data two levels deeper than summary moments — at the level of fitted marginal distributions (family-mix total-variation distance 0.044) and cross-material copula dependence (copula-family TVD 0.081). Contents: 5,000 synthetic materials × 60 monthly periods, plus a multi-item order log and MAP unit prices; per-material demand statistics and fidelity-audit records; the generator and a reproduction guide. Preserves ~70% monthly zeros, ~90% intermittent/lumpy, right-skewed nonzero sizes, multi-item orders, and ~20% cold-start. Intended uses: decision-aware forecasting evaluation (accuracy and order-level service under a policy), foundation-model adaptation on sparse demand, intermittent-demand inventory-policy design under realistic (light) cross-material dependence, and bias-direction diagnostics. Limitations: one industrial domain, monthly granularity, 60-month window, light dependence regime; tail-dependence estimates noisy at 60 periods. No real identifiers, prices, or schema are released. License CC-BY-4.0.