Reproducibility package for "Detection without discrimination: a blinded, pre-registered test of pre-fault signals in public wind-turbine SCADA data"

Registration documents, hash records, change logs, prediction hashes, analysis code, sealed label and window maps, and result tables for a four-stage blinded and pre-registered study of pre-fault signals in public 10-minute wind-turbine SCADA data (CARE to Compare; Kelmarsh and Penmanshiel; Hill of Towie; Fuhrlander FL2500). The study measures how often faults go undetected by five detectors, how well detectors separate faults from normal periods, and whether two models aimed at the cause of false alarms meet a pre-registered bar for replacing existing detectors on held-out data. Raw data are not redistributed; input checksums and download sources are listed. Registration documents and change logs are in Korean; the accompanying manuscript gives the rules and deviations in English.