AMIGA agricultural market-outlook runs — results bundle (11 commodities, 2026-Q1 vintage)

This dataset is the results bundle for the paper "AI-enabled agriculturalmarket outlook models: agentic workflows, illustrative applications, and futureoutlook" (Scherer & van Wassenaer). It contains the complete run artifacts behindboth result sets reported in the paper, for eleven agricultural commodities at the2026-Q1 data vintage, forecasting twelve months ahead. AMIGA (Agentic Market Intelligence for Global Agriculture) is an agentic workflowcoupling statistical and machine-learning forecasting with role-specific large-language-model agents under explicit human governance (variable-approval, causal-graph, and interpretation-review gates). Contents — 16 run folders + MANIFEST.csv + README.txt: * Pre-modification automated BASELINE runs — the basis of Table 3 and Figure 3 (full-LLM variable scoping, no expert edits). * Expert-modified runs — the basis of Table 4 (post human-in-the-loop review; the ablation).Six commodities were not modified(one folder serves both tables); five weremodified (EU soft wheat, Global milk, Rice, EU raw milk, Fluid milk) and haveseparate baseline and deployed folders. MANIFEST.csv maps every folder to itscommodity, paper table, and CRPS skill score. Each run folder contains the assembled modelling dataset (data_*.parquet: targetprice series and driver features at publication-date vintage), tier-1/tier-2 andfinal forecasts (CSV), metric/model/backtest artifacts (JSON), figures (PNG), theHTML report, and the interpretation markdown. The AMIGA pipeline code is archived separately (see Related identifiers). License note: CC BY 4.0 applies to the authors' own artifacts. The data_*.parquetmatrices embed values from third-party sources, which retain their original terms;see README.txt for provenance and redistribution details.