Reconciling data years

[WesIngwersen]

cc: @MoLi7
Closes: http://github.com/cornerstone-data/team-in/issues/43

Reconciling data years analysis (bedrock/analysis/reconciling_data_years/)

Four model specifications

Model	A matrix	B matrix / x	q
1	2017 USEEIO (fixed)	2017 E and V-sum x (fixed)	2017 V column sums
2	2017 USEEIO (fixed)	Annual E with BEA time-series x	Authoritative x at model_base_year=year
3a	2017 USEEIO (fixed)	Annual E with GHG-year x deflated to 2017	GHG-year x deflated to 2017
3b	Commodity-PI inflated to year	Annual E with BEA time-series x	Authoritative x at model_base_year=year
4	Summary-table scaled + commodity-PI inflated	Annual E; B scaled 2017→year via summary tables	Summary-scaled q

analysis.py — main() loop

For each model × year:

• Clears all function caches and resets config per iteration.
• Derives B, A, q, e_c, L, d, and n in current-year dollars.
• Deflates d and n to LATEST_TARGET_YEAR (2024) USD for cross-year comparison.

Output statistics

• n_stats.csv — mean and median of n across sectors, per model per year.
• q_ec_correlation.csv — YoY Pearson r (Δq vs Δe_c) and pct_same_sign (fraction of sectors where q and e_c move in the same direction) per model per year-transition.
• q_ec_correlation_summary.csv — wide-format pivot of the above with per-model averages across all transitions.
• efs.csv — long-format record of d, n, q, e_c per model × year × sector.

Figures

• trends_d_q_ec.png — time-series of e_c, q, d, and n indexed to first year = 100, one row per model (including model 1), y-axis shared within each column for direct cross-model comparison.
• n_yoy_distribution.png — signed YoY violin plot of n across models.

Additional code changes

B-matrix scaling guard (derive_cornerstone_B_non_finetuned)

scale_cornerstone_B is now only invoked when use_scaled_x_and_scaled_Vnorm_for_B=True, preventing unintended B scaling in configs that do not opt into it.