SMT2020 Full-Fab Benchmark — 30-Replication Simulation Output (P-30)

Event-level output of 30 independent replications of a four-year (1460-day, 2018-01-01 to 2021-12-31, UTC) discrete-event simulation of the SMT2020 full-fab semiconductor manufacturing benchmark model. Each replication is an independent random realization of the same fab under the same demand plan, differing only in the random seed driving stochastic equipment events; the preventive-maintenance schedule is deterministic while breakdown realizations vary by seed. No single replication is privileged — all reported KPIs are means (and standard deviations) across the 30 replications. Each replication is provided as a ZIP archive (rep_01.zip … rep_30.zip, ~650 MB each) containing two Apache Parquet files: lot_log.parquet (per-lot, per-step processing intervals; ~38 million rows per replication) and equipment_stops.parquet (equipment down intervals tagged preventive maintenance (PM) or breakdown (BD)). Two summary files derived from the logs are also included — kpis_by_replication.csv (per-replication throughput, cycle time, WIP, and per-product cycle time) and wip_daily.csv (daily in-system WIP for every replication) — so the entire confidence band can be reconstructed without re-running the engine. README.md and DATA_DICTIONARY.md provide the full column-level schema, and replications.csv maps each replication to its random seed and event counts. This is the companion dataset to "Independent Reproduction of the SMT2020 Full-Fab Benchmark: Capacity Equivalence and Cycle-Time Sensitivity to an Underspecified Batch-Composition Policy" (SimTrek Inc., 2026). The input master data is the publicly available SMT2020 testbed (FernUniversität in Hagen). The simulation engine itself is not part of this release.