| English | 中文 |
These operations apply across research stages and correspond to explanations in methods/shared. A particular study references the capability instead of copying it under each stage.
| Capability | Skill | Method | Execution status |
|---|---|---|---|
| Spatial analysis and mapping | spatial_analysis | Spatial relationships, CRS and support | R and Python direct helpers with synthetic checks; geocoding/raster guidance; no run-tool or MCP adapter |
| Machine-learning models, contribution and causal analysis | machine_learning_model | Regression targets, DAGs and interpretation | R: ml_regression_workflow; Python: ml_python_regression_workflow; causal PLR remains R-only |
| Reproduction and comparison | model_reproduction | Comparison and reliability | Reviewed direct comparison/recovery helpers; no new run-tool or MCP execution adapter |
machine_learning_model/ is the shared machine-learning home. Its models/ groups linear, tree-based and sequence references; evaluation/ supplies the common baseline comparison; frameworks/r/mlr3/ contains R framework guidance. Shared calculation code retains scripts/r/ and scripts/python/. Framework and dedicated sequence entries keep their own contracts and reference status. The general regression workflow documents data/skill combinations, while projects define particular scientific experiments.
For air-pollution tasks, read process and impact methods, then the regression skill’s operating contract and domain cases. Scientific examples do not add an executable model or causal estimand.
Stage-specific operations remain under their primary stage. Stage 02 receives conceptual method references only. Tool IDs and discovery depend on the catalog and explicit mappings, not folder names. See classified usage and the asset contract. New generic regression results use machine_learning_model/results/, ignored by Git; prior local results remain in place without automatic migration.
Control costs, distributional scenarios and GCAM output analysis are maintained in stage 04. Other stages may reference those same skills without a duplicate in shared.