Skip to content

Source Pages — In-package Docs Index

Eight source pages derived from the in-package documentation files under market_insights_models/src/commodity_hindcast/.

Pages

  • README.md — Primary user guide: pipeline diagram, CLI reference table, INPUT_DATA_DIR/runs/… data layout (authoritative), Make targets, MLflow wiring, project layout
  • DESIGN.md — EARS-format pipeline contract: 35 design decisions covering config loading, INPUT_DATA_DIR as sole resolver, stage artefact paths, S3-safe paths, plot separation, unit system, area imputation, and forecast isolation
  • TODO.md — Backlog of open and completed refactoring items: cross-pipeline dependency violations (verbatim file-and-line citations), naming cleanup, area-imputation consolidation, and structural debt from the large agentic refactoring session
  • experiments.md — Five-item model experiment roadmap: first-differencing, ARIMA/SARIMAX, Nixtla, phenoweighting, irrigation feature for Texas
  • features_README.md — Feature-assembly orchestrator: build_featuresassemble contract, fit/pred parquet split on harvest date, metadata.json column registry
  • features_builders_README.md — Builder protocol: accept (path, cfg, years), return DataFrame keyed by (year, geo_identifier, init_date), no duplicate keys, registry dispatch
  • CLAUDE.md — Single-line enforcement directive: "Ensure compliance with DESIGN.md" — delegates all constraints to the design document
  • in_package_DOMAIN_MODEL.md — Combined coverage of DOMAIN_MODEL.md, DOMAIN_MODEL2.md, schema.yaml, gen_linkml_schema.py: ubiquitous language (30+ terms), 7 bounded contexts, entity catalogue (9 entities + 7 value objects), 13 invariants, AbstractSlice protocol, 8-layer import DAG, auto-generated LinkML schema (42 classes, 21 enums)