feat(analysis): step 6 phase 2/3 — EF diagnostics compile + plot scripts

[MoLi7]

cc:
Closes:

What changed? Why?

Step 6 / Phase 2 + Phase 3 of epic #337. Two new scripts under bedrock/analysis/a_matrix_time_series/:

• compile_ef_diagnostics.py — reads output/results/ef_run_index.csv for the (approach, baseline, sheet_id) triples produced by Phase 1 diagnostics runs. Pulls N_and_diffs and D_and_diffs from each Sheet via read_sheet_tab, joins them on the sector index, and writes:
  • output/results/ef_comparison.xlsx — one tab per (approach, baseline) plus summary_vs_ceda / summary_vs_useeio (per-approach p50 / p95 / max of |N_perc_diff| and |D_perc_diff| plus n_significant at a 10% threshold).
  • output/results/ef_scatter_coords.parquet — long-format (approach, baseline, ef_kind, sector, x_baseline, y_approach) for Phase 3.
  • Appends ef_summary_vs_* tabs to the run-report Sheet (sheet ID from last_run_sheet_id.txt); skipped with a warning if missing.
• plot_ef_diagnostics.py — reads the parquet from above and produces, for each (baseline, ef_kind) pair, two 2×2 figures: scatter (ef_scatter_{baseline}_{ef_kind}.png) and per-sector %-diff histogram (ef_pct_hist_{baseline}_{ef_kind}.png). Histograms mirror the baseline_snapshot_comparison/compare_B_Adom convention: ±100% clip, 60 bins, vertical zero line, consistent xlim across panels and figures.

Stop tracking the two embedded-headline PNGs that earlier PRs (#364, #365) force-added past the *.png gitignore (keysector_impact_heatmap.png, summary_a_rmse_ranking.png). They were tracked because of the embedded-PR-body pattern but kept dirtying git status after every Step 4/5 re-run. With them removed, every plot in the analysis dir is now treated as output.

Testing

• Ran both scripts end-to-end against 7 real diagnostics Sheets in the v0.3 Diagnostics Drive folder; outputs match expected schema.
• black --check, ruff check ., mypy bedrock (384 files) all clean.

[MoLi7]

• Add LMDI signal-noise diagnosis for A-matrix physical residual #405
• feat(analysis): N time-series stability comparison across A-matrix methods #381
• feat(analysis): step 7 dispatch — workflow input + EF time-series dispatcher #380
• feat(analysis): step 6 phase 2/3 — EF diagnostics compile + plot scripts #379 👈 (View in Graphite)
• main

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

[WesIngwersen]

I can't find the script that compiles ef_run_index.csv which is read from. It's giving me an error suggesting I prepare one manually but that doesn't make a lot of sense that it would have to be prepared manually and it does not give enough guidance on how to do so.

FileNotFoundError: {my local path}bedrock\bedrock\analysis\a_matrix_time_series\output\results\ef_run_index.csv not found — populate it by hand from the Phase 1 diagnostics runs (one row per (approach, baseline) Sheet).

@MoLi7

[MoLi7]

I can't find the script that compiles ef_run_index.csv which is read from. It's giving me an error suggesting I prepare one manually but that doesn't make a lot of sense that it would have to be prepared manually and it does not give enough guidance on how to do so.

FileNotFoundError: {my local path}bedrock\bedrock\analysis\a_matrix_time_series\output\results\ef_run_index.csv not found — populate it by hand from the Phase 1 diagnostics runs (one row per (approach, baseline) Sheet).

@MoLi7

@WesIngwersen I pushed rebuild_run_index_from_drive.py in 81eed45 that lists the GSheets in our diagnostics Drive folder, parses each title, and writes ef_run_index.csv. You can run it without re-dispatching and re-generating all EF diagnostics:
python -m bedrock.analysis.a_matrix_time_series.rebuild_run_index_from_drive --folder-id <DRIVE_FOLDER_ID>

The folder ID is the part after /folders/ in your Drive URL.