[vahid-ahmadi]

Merge only if boe-var-model #24 merges. This is option (b); option (a), #231, pins the figures to the paper vintage instead. Stacked on #231.

It updates the site's BoE SVAR replication and evaluation figures to #24's committed outputs at d4ec3e7. Every value below was read from #24's results/ files or recomputed from them. Nothing was typed in by hand.

Number changes (old -> new, source)

Item	Old	New	Source (#24)
Accepted draws / rate	744 / 7.4%	765 / 7.6%	results/summary.md
ESS	356.1	368.0	summary.md
Global FEVD share, UK GDP (mean, 68% band)	37.4 [23.4, 52.2]	38.7 [24.0, 53.1]	summary.md
Global FEVD share, UK CPI (mean, 68% band)	42.3 [25.5, 60.3]	41.5 [25.6, 56.3]	summary.md
CPI shortfall vs ~50 / band half-width	~8pp / ±14pp	~8.5pp / ±15pp	derived
Identified total, GDP / CPI	76.3 / 78.9	77.3 / 78.3	global + domestic means
Renormalised, superseded	42.8 / 49.5	43.2 / 48.0	summary.md
UK supply / demand / mon. pol. share of CPI	13.9 / 11.4 / 8.7	14.6 / 10.5 / 8.9 (same ranking)	summary.md per-shock table
Rolling origins (last origin)	49 (2024Q1)	50 (2024Q2); data through 2026Q2	rolling_evaluation.json
Smallest adjusted q, headline / published spec	0.36 / 0.40	0.39 / 0.42	rolling_evaluation.json multiplicity
Mean 68% coverage at h=1 / h=8	71% / 57%	71% / 58%	coverage_evaluation.json
Worst-origin share, UK GDP h=1	63%	64%	rolling_evaluation.json
Fast CI config, GDP / CPI	42.0 / 37.6	40.1 / 40.1	make_comparison.py re-run on the #24 checkout (under the heavy-job lock). A run on 845e2a4 first reproduced 41.95 / 37.57 exactly.
Forecast-revision adding-up error	5.23e-12	3.64e-12	forecast_summary.md

These values do not change at two decimals: Bank Rate 0.79/0.85, UK CPI 0.83/1.03 and 0.63/0.67, UK GDP 1.06/1.12 and the published spec 0.99/0.95, 16 of 64 cells, AR(1) 94.5%, and the satellite 0.82–0.99.

The inline charts were regenerated by make_charts.py. Its coverage title now reads the origin count from the data rather than the hard-coded "49". fig_comparison.pdf and comparison_numbers.json come from the re-run. The paper was rebuilt with tectonic: 33 pages, with no new overfull boxes. The results/ links go back to blob/main, and the vintage line now cites #24, with 845e2a4 kept for the earlier values.

Provenance change: coverage_evaluation.json is now #24's repo artifact. The previous file came from this repo's own make_coverage.py and differed slightly from 845e2a4's.

Not updated: needs Vahid or a heavy run

• svar/validation "99.2% of draws explosive, median root 1.004, 4 of the 744 accepted draws stable". Rename technical identifiers: macromod → policyengine-macro #24 has no committed stability diagnostic, so this needs a re-run.
• "p = 0.33–0.43" (published-spec UK GDP). I could not reproduce it from either JSON, so I left it.
• The 90% recalibration figures (77%, 82.2%, 82.4%, 80.5%, 86.5%; boe Round-4 precision fixes: hosted claim carves out score_reform; .venv-og gets the CLI #19–SVAR is the baseline/conditioning member — retire 'reform scoring: planned' #21) and the 0.40% monetary IRF. They are not in Rename technical identifiers: macromod → policyengine-macro #24's outputs.
• The paper's frozen-2024Q2 figures (fig_irf/fevd/hd/fan.pdf, figure_numbers.json, and the 3,000-draw 38.1 / 42.3) and the forecast-revision row 487 / ESS 207.5 still use the pre-Rename technical identifiers: macromod → policyengine-macro #24 dataset. Under (b) the paper therefore mixes vintages. A make_figures.py re-run is heavy and is left as a follow-up. 487 / 207.5 was already a mismatch before this PR.
• The "data through 2026Q1" wording for the hosted forecast (main.tex:35, svar/index, core.py) changes with the Modal redeploy, not here.
• Rename technical identifiers: macromod → policyengine-macro #24 holds out five quarters (2025Q2–2026Q2), not "the last year". The paper now says five; methodology.md in boe-var-model still says one year.
• models/validation had a one-word edit ("8pp" → "8.5pp"). That file is outside ui-models.

Checks: pytest (PYTHONPATH=integration/src) 1222 passed; site_contract OK; make_charts.py --check OK; linkcheck 0 broken.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
macromod	Ready	Preview	Nov 6, 2026 4:04am UTC