Site currency fixes, UI defects, and the corrected GDP claim#87
Merged
Conversation
Content corrections, each verified against the repos: - frb-us page claimed "no stochastic simulation ... confidence bands are not available". stochsim_bands() has existed since us-frb-model #5 and is exported from frbus/__init__.py; the validation page already said the opposite, so the site contradicted itself. Split the row: mcontrol is still genuinely absent, and the honest limit is that the hosted MCP path returns a deterministic projection, not that the engine is missing. - frb-us page said "two invariants" and described a single cross-validated shock; VALIDATION.md gates four like-for-like scenarios. - svar page attributed 6,000 draws / 430 accepted / ESS 197.9 to the hosted forecast. Those belong to the Section-5 revision run; the hosted default is 2,000 draws, 135 accepted, ESS 65.3 -- the misattribution overstated posterior precision roughly threefold. - svar page said replication claims are "scored on the paper's original sample"; they are scored on the extended 1992Q1-2025Q1 window. - capabilities.py still described frb-us fidelity as "a reference shock". - oguk version in core.py was 0.3.0; OG-UK's pyproject.toml says 0.3.2. - 5.3e-12 -> 5.23e-12; and the "7-8 points" summary heading was left behind when #85 updated the body and table beneath it to 6-10. UI defects: - --brand-wash and --radius-card were referenced 5+ times but never defined, so those backgrounds resolved to nothing -- the PUBLISHED ANCHOR chips rendered as bare text while STATUTE/CALIBRATED rendered as pills, breaking the site's own three-class visual claim. - --brand and --brand-strong both resolve to --primary (a deliberate AA fix), which silently made .vc-s1 and .vc-s2 the same colour in every chart while .vc-lab/.vc-lab2 styled the legend as two colours. Series 2 now uses a dark slate that survives greyscale. - .vc-b2/.vc-b3 measured ~2.3:1 and ~1.3:1, under the WCAG 1.4.11 3:1 floor, on the comparison bars our numbers are judged against. - .vc-warn used the same teal as links and series 1, so "Weak" rendered in the affirmative accent; now --brand-warning. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adding Diebold-Mariano tests to boe-var-model's rolling evaluation shows only 3 of 64 (variable, horizon) cells differ significantly from the benchmark at 5%. The site said "CPI beats no change; GDP does not in rolling tests", which reads as an established loss. It is not: UK GDP's ratio of 1.06-1.09 carries p = 0.38 to 0.67, so the model is statistically indistinguishable from a random walk for GDP rather than worse than it. The CPI win survives at short horizons (0.63, p < 0.005 at h=1) but not at h=8 (p = 0.11). The one significant adverse result is the exchange rate at h=8 (1.41, p = 0.03). Updates the audit cell, the chart description and the capabilities registry, and syncs the regenerated artifact carrying dm_stat, dm_pvalue and per-origin errors. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
style.css documented at length that field.js was "now deleted along with its markup" and replaced by a static lattice, because animated ornament behind continuous technical prose is a reading hazard. The deletion was never carried out in the HTML: the canvas was still loaded on all 12 pages and still drifting behind the body copy, and the stylesheet's own comments contradicted the rendered page. Removes the script and canvas from all 12 pages, deletes field.js, and drops the now-dead #field rules including the dark-mode override. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Three commits from a four-agent review (content currency, UI/accessibility, comparable-site research, chart design).
1. Contradicted claims
stochsim_bands()has existed since us-frb-model Site: rename to PolicyEngine Macro + P0 accuracy/honesty pass #5 and is exported fromfrbus/__init__.py; the validation page already said the opposite, so the site contradicted itself. Row split:mcontrolis genuinely absent, and the honest limit is that the hosted MCP path returns a deterministic projection.capabilities.pystill described frb-us fidelity as "a reference shock".ogukincore.pywas 0.3.0; OG-UK'spyproject.tomlsays 0.3.2.2. The GDP claim was wrong
Adding Diebold–Mariano tests (boe-var-model #5) shows UK GDP's ratio of 1.06–1.09 carries p = 0.38 to 0.67. The site said "CPI beats no change; GDP does not in rolling tests", which reads as an established loss. It is not — the model is statistically indistinguishable from a random walk for GDP. Only 3 of 64 cells are significant at 5%.
3. UI defects
--brand-washand--radius-cardwere referenced 5+ times but never defined, so those backgrounds resolved to nothing — the PUBLISHED ANCHOR chips rendered as bare text while STATUTE/CALIBRATED rendered as pills, breaking the site's own three-class visual claim.--brandand--brand-strongboth resolve to--primary(a deliberate AA fix), which silently made.vc-s1and.vc-s2the same colour in every chart while the legend styled them as two..vc-b2/.vc-b3measured ~2.3:1 and ~1.3:1, under the WCAG 1.4.11 3:1 floor..vc-warnused the same teal as links, so "Weak" rendered in the affirmative accent.field.jsremoved.style.cssdocumented that it was "now deleted along with its markup" because animated ornament behind technical prose is a reading hazard — but it was still loaded on all 12 pages and still drifting behind the body copy.183 integration tests pass.
Known follow-ups, deliberately not in this PR
The benchmark itself is questionable: against a random walk with drift (the correct naive for a trending log level) the CPI win largely evaporates (0.63 → 0.83 at h=1, 0.67 → 1.03 at h=8), while Bank Rate improves to 0.79–0.85. And excluding 6 Covid quarters, UK GDP flips to 0.77. Both verified. Fixing that is model-repo work and will change published conclusions, so it is sequenced separately.
🤖 Generated with Claude Code