This reference describes registered tools, not an automatic unit-price selection or scientific-validation pipeline. Run from the repository root using the extraction skill and model guide.
Each tool creates its own results/<project_id>/<run_id>/<stage>/<tool_id>/:
| Tool ID | Actual files |
|---|---|
gcam_extract_market_sector |
market_sector_rows.csv, market_sector_summary.csv |
gcam_extract_technology_components |
technology_component_rows.csv, technology_component_summary.csv |
gcam_aggregate_provinces |
province_aggregate.csv |
gcam_plot_sector_change (run-tool, R) |
sector_change_2021_2035.png, sector_change_weighted_mean.csv |
The extractors take comma-separated scenario/region/sector/year strings; aggregation takes the existing market-sector CSV as input_csv. See the skill for parameter JSON and run-tool commands. Pass the successful market task ID explicitly; there is no automatic chaining.
Market rows distinguish marketplace from sector_cost. Preserve source, year, unit and scenario columns. Confirm provincial coverage; China-only rows do not support province aggregation. AH/BJ examples are subsets, not national coverage.
The aggregator excludes China/Global, groups by scenario/source/sector/year, and writes weighted_price, quantity_sum, province_count. Nonnegative quantities weight available prices; it falls back to an ordinary mean when valid weights sum to zero. Units and coverage are not automatically validated. Inspect original rows and these assumptions before interpreting results.
The technology tool extracts available costs, output, coefficients, demand, component prices and CO2 fields, then forms its existing summary. These are not a validated sector-cost decomposition or universal price proxy. Review sample coverage and denominators before comparing technology-weighted and sector-weighted values.
No current tool creates selected_unit_price, selected_quantity, selected_aggregation_scope or an automatic explainability-quality grade. Selection among marketplace, sector-cost, provincial and China representations requires a separately reviewed method and implementation. Do not silently apply fixed residual thresholds or assume a conceptual hierarchy has been accepted.
Plotting uses run-tool with an explicit input_csv and fixed 2021/2035 years; the runner injects its output directory and writes a sanitized receipt. Check packages and coverage first; use the gcam-plot.json example after replacing its source run ID and scenario.
Document unit compatibility, model version, missing rows, zero-weight behavior and coverage locally. For a justified conversion or override, preserve original units and method evidence; successful extraction alone does not authorize a revenue or national-price claim.