Reproducibility data and evaluation artifacts for "Hierarchical document ranking of AAER-linked reporting risk under firm-holdout evaluation"

This dataset record accompanies the manuscript “Hierarchical document ranking of AAER-linked reporting risk under firm-holdout evaluation”.The archive contains reproducibility data and derived evaluation artifacts used to verify the reported document-ranking and financial-ratio analyses. The released HiDoc-IPCW holdout score file contains 9,643 held-out filing-level observations, including 228 AAER-linked observations, with one label and one model score per row. The archive also contains aggregate inventory information for the Bao 2020 firm-year panel and the frozen summary values used for the financial-ratio comparator, including issuer-level holdout counts, AUROC values, censoring-survival estimates, and inverse-probability weights.