CodeSoulco/TextInsightBench
TextInsightBench English | 简体中文 A natural-language data-mining benchmark for agents: 50 tasks, 435,000 task documents and 944,468 unlabeled learning documents. Each task provides 5,000 or 10,000 texts and a research objective. Agents choose the patterns, populations and comparisons to investigate, then submit up to three findings with complete document assignments, exact quotations, statistics, counterexamples and limitations. Any analysis method is allowed. Contents… See the full description on the dataset page: https://huggingface.co/datasets/CodeSoulco/TextInsightBench.
0976
1# Verification scope2 3The release was checked with local automated tests and an all-50-task abstaining4agent smoke run. This verifies loading, corpus hashes, task dispatch, output5contracts and reporting, not real-agent mining quality or empirical difficulty.6 7Synthetic tests exercise agent-selected group/date comparisons, forbidden8filters, overlapping groups, minimum population sizes, exact partitions,9Simpson-style reversals, missing metadata, counterexamples, quotation offsets,10score bindings, null reference coverage and bounded reproducible semantic packets.11 12```bash13python -m unittest discover -s tests -v14tib verify --data data/participant --with-learning15```16 17The data builder deterministically selects disjoint IDs from an already curated18pool, preserves original text, enriches released metadata and filters selected19documents out of the remaining pool. It records exact input shard hashes.20Current-snapshot disjointness does not erase historical public exposure.21