OBS-T: CDSL Correction-Event Corpus — A Twelve-Year Longitudinal Error-Typology Dataset for Sanskrit Historical Dictionaries

OBS-T (Observatory Typing) is a corpus of 52,498 typed correction events covering 43 Sanskrit historical dictionaries in the Cologne Digital Sanskrit Dictionaries (CDSL), spanning 2014–2026. Key statistics: 52,498 correction events (24,441 form layer + 28,057 git layer); 43 dictionaries; 208 corrector aliases; date range 2014-03-18 to 2026; 9 microstructure location labels (error_component); 9 edit-type labels (edit_type); evidence labels derived (form layer) and inferred (git layer); 64.3% derived by evidence label. Contents: correction_events_release.csv — 52,498 rows × 31 columns. One row = one old→new correction event. Licensed CC-BY-4.0. observatory/site/src/data/obs_t_* — aggregate tables (component distribution, timelines, corrector summaries, crosswalk labels). docs/DATASHEET.md — full Gebru et al. (2021) datasheet. data/schema/correction-event.schema.json — JSON schema for the release file. The repository code is GPL-3.0; the data files are CC-BY-4.0 (see DATA_LICENSE.md).