implement analysis for decisions 3, 5, and 7 in dicussion #85

[jvendries]

cc: cornerstone-data/methods#85
Closes:

What changed? Why?

Adds a new d_85 analysis module under bedrock/analysis/electricity/ to support the methods discussion #85 resolution analysis covering Decisions 3, 5, and 7.

Decision 3 explores using EPA Table 8.3 IOU expense shares (Production/Transmission/Distribution) as an alternative to UGO305-A gross output weights for the Use-intersection step. Two variants are implemented: d8_mixed (Table 8.3 weights for the Use intersection only) and d8_offdiag (a full 3×3 absolute-dollar intersection matrix where generation columns are forced diagonal per Rules D1/D2).

Decision 5 introduces price-differentiated commodity row splits using EIA Table 2.4 retail electricity prices (cents/kWh) by end-use sector (Residential, Commercial, Industrial, Transportation). Each purchaser column is assigned an EPA end-use class via a rule-based NAICS chapter mapping, and weights are tilted by the ratio of that sector's price to the national average.

Decision 7 simulates differentiated UGO305 year-scaling by computing per-child GO ratios (generation/transmission/distribution) between the 2017 detail year and the IO target year, with a pure variant and an anchored variant that preserves the weighted mean against the Utilities summary ratio.

A derive_post_reallocation_checkpoint() helper is extracted into disaggregation_matrices.py so the stage-2 IO state (post-reallocation, pre-disaggregation) can be shared cleanly between the existing pipeline and the new scenario analysis. The existing _build_stage_tables function is updated to call this helper rather than duplicating the logic.

Each decision has a corresponding report builder that writes an Excel workbook to d_85/output/ (gitignored). A CLI entry point (run_resolution_analysis.py) accepts --decision 3|5|7|all and a config name argument.

Testing

Unit tests cover:

• table83_go_weights normalization and ordering
• build_ugo_col_table83_row_intersection_matrix column-sum (Rule D1) and total preservation, including a worked numeric example
• apply_use_intersection_custom total preservation and aggregate cell zeroing
• split_y_commodity_row_by_column aggregate row removal
• table_8_3_gtd_expenses_musd and table_2_4_prices_cents_kwh value extraction and missing-flow error raising
• classify_industry_end_use rule coverage and build_price_tilt_weights_by_column per-column normalization
• compute_balance_metrics shape and zero relative error on balanced inputs; va_summary_by_scenario negative VA flagging
• diff_ef_vectors skipping metrics_only scenarios; report_metadata key presence
• _run_baseline parity against the production disaggregate_electricity_make_use_va path (V, Udom, Uimp, VA, Y, q, x all match within DISAGG_BALANCE_ATOL)
• d8_mixed scenario routing Table 8.3 weights to the Use-intersection step
• p24_2017 per-column weight normalization
• build_ugo305_detail_ratios ratio computation and build_anchored_overrides weighted-mean preservation
• derive_Aq_from_scenario, scenario_vnorm, and _apply_electricity_q_overrides shape and value correctness

[jvendries]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• Implement unit conversion #492
• Implement changes from resolutions from discussion #85 #490
• implement analysis for decisions 3, 5, and 7 in dicussion #85 #481 👈 (View in Graphite)
• Implement print to excel for PR3 #463
• Implement 221100 split #462
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[jvendries]

Report from the analysis on this branch can be found in this discussion_85_analysis_report.md

[WesIngwersen]

Report from the analysis on this branch can be found in this discussion_85_analysis_report.md

These AI-generated reports are just unreadable to an extent. This needs some human interpretation.

[WesIngwersen]

I ran run_resolution_analysis successfully but I don't trust my results due to use of GHGI instead of eGRID, related to the comment i made recommending just coding for eGRID use (and the code error trying to import the flowsa package).
I would like to see the differences in use of generation after Decision 5 compared more.