Building the market before adoption: administrative market formation in China's humanoid robot sector

The data are a bounded census of the Chinese public record on humanoid robot and embodied intelligence policy and procurement, covering 13 of 31 province-level units, three sub-provincial cities, and the central level, January 2021 to August 2026. The corpus holds 2,686 collected rows: 335 policy instruments, 444 procurement notices, 574 scenario and demand-release items, 165 disconfirming records, 954 firm events for an 89-firm roster, and 125 calibration evidence rows for non-humanoid comparison deployments.Contents: harmonized CSV files (canonical/), the exhibit inputs computed from them (analysis/), per-jurisdiction codings with verbatim evidence (jurisdiction_summaries/), the source runs with full query logs including empty queries (raw_runs/), a codebook defining every variable and coding rule, a corrections log recording all 33 post-audit corrections, a second-coding report with its adjudication, a checksummed manifest, and a license (CC BY 4.0 data, MIT code).

All files are UTF-8 CSV, JSON, or plain-text Markdown; no proprietary formats. Chinese source titles and quoted passages are verbatim with English translations.Reproduction: python build_exhibits.py from the package root regenerates the exhibit tables and exits nonzero on any schema violation; python validation/validation_sample.py regenerates the 56-row extraction-validation sample. Python 3.10 or later, standard library only.