Coded evidence map of measured farm outcomes in AI-in-agriculture research (520 studies)

Data, coding scripts and raw API responses underlying the manuscript: Alkhouli, A. 'Prediction is not control: a deployment-readiness map of measured farm outcomes in AI-in-agriculture research', submitted to Computers and Electronics in Agriculture, September 2026. The deposit contains a coded database of 520 AI-in-agriculture studies (17 columns, one record per study), a codebook stating the verbatim classification rule for the core outcome variable, a re-verification of every record coded as reporting a measured farm outcome, deployment-readiness ladder codes and aggregate statistics, 11 Python scripts covering retrieval through analysis, and the unmodified OpenAlex API responses retrieved on 2026-09-23 so that every coded row can be traced to its source. Important caveats: coding is from titles and abstracts, not full texts, so studies reporting farm outcomes are likely under-counted; coding was AI-assisted (Claude, Anthropic) and reviewed by the author, with no second independent human coder and therefore no inter-coder reliability statistic; the 'no' codes were not individually re-verified; ladder levels L1-L3 are keyword-assigned and noisier than the measured-outcome variable; country is missing for 95 of 520 records; and counts are reported by stratum, the pooled figure being descriptive only. See README.md and codebook.md before reuse.