temsa/govie-office-holder-regression-bilingual-v3

--- license: mit task_categories: - text-retrieval - question-answering language: - en - ga pretty_name: GOV.IE Office-Holder Regression Bilingual v3 size_categories: - n<1K --- # GOV.IE Office-Holder Regression Bilingual v3 This dataset is a tenure-aware regression truth set for GOV.IE office-holder search and chatbot checks. It is derived from `temsa/govie-office-holder-reranker-bilingual-v2` and adds: - explicit `effective_from` / `effective_to` intervals - stable role-page canonical targets for role queries - recent-history backfill for a small set of high-value offices - Education portfolio coverage across the 2025 holder change ## Intended use The dataset is intended for: - post-deploy office-holder regression checks - snapshot-aware validation with `--as-of-date` - comparing retrieval behavior against a stable indexed environment ## Core fields - `query_id`, `query`, `query_type`, `language`, `split` - `holder`, `holder_ascii`, `role` - `official_profile_url` - `effective_from`, `effective_to` - `candidates` - `source_urls`, `source_notes` `candidates` encodes the accepted canonical targets for a query. For role queries this usually includes the current stable role page plus, where useful, the associated biography path.