Algorithm filing in Chinese public procurement decision chains: replication dataset, 2023 to 2025

The corpus holds 322 procurement instruments resolving to about 208 projects, built on a Zhejiang provincial spine with targeted recovery from national and other provincial portals, together with 38 non-AI control instruments and 327 screened complaint-handling decisions. The archive contains the retrieval log with per-attempt HTTP status, the coded corpus with source addresses, two versions of the search batteries with their disambiguation rules, the recall measurement against 21 declared control clauses, every battery hit with hand adjudication, the second coder's packet and completed sheet, reliability statistics with contingency tables for 349 double-coded records, three-level filing matches for the winning suppliers, the six-case locus, slot, and register coding, the 30-instrument deep dive with evidence files and qualification classification, the clause-independence computation, and a corrections log.

Source documents (PDF, DOCX, DOC, HTML, TXT) are deposited as served by the issuing portals; documents whose redistribution a portal restricts are identified by source address and retrieval date so that a reader can retrieve them directly. Tabular files are UTF-8 CSV with a data dictionary and SHA-256 checksums; scripts run on Python 3 standard library. Data, code, and documentation are released under CC BY 4.0; rights in the public records remain with the issuing bodies.