Dataset for "Lost in summarisation" (bigbio.ai/articles/lost-in-summarisation/), built 2026-10-04T00:40:28.451Z.
English-language coverage of China's Pharmaceutical Industry Development 15th Five-Year Plan
(MIIT and nine bodies, 工信部联规〔2026〕210号), collected 2026-09-17 to 2026-10-03.

corpus_index.csv   128 articles: URL, outlet, type, date, title (article text not redistributed)
fetch_log.csv      publisher HTML fetch outcome per article (ok / shell / failed)
links.csv          outbound links classified; links_original = links a Chinese government page for the plan
cluster_raw.csv    MinHash copy family (5-shingles, 256 perms, Jaccard >= 0.5) and HDBSCAN groups
pairs.csv          near-copy pairs with Jaccard and embedding cosine
fidelity.csv       sentence matches to the plan's Chinese text and to Xinhua's English feature
claims.csv         594 Box 1 claims; verdict (final), judge_verdict (DeepSeek, t=0, blind)
truth_box1.csv     the plan's ten Box 1 targets and their conditions
*_meta.json        corpus sha256, model (BAAI/bge-m3 rev 5617a9f6), thresholds, calibration
pipeline/          the scripts that produced every file (fetch_html, links, cluster, fidelity, claims)
collection_log.md  queries run, tools, rejected items

Corpus sha256 518f7b1eae533b051f277194d93c4a2e2202675396c01d92a7af07f6d4f40301.
Plan PDF sha256 e4ba3a4c737f9ed1deafde79fb0d7cf63b4bd47aac93f68502aadaa95a25da8e.