Fix a clipped metric strip, unify the diagram fonts, trim validation, and document driving the model directly - #187
Merged
Conversation
… and document driving the model directly ## The metric strip was clipped, not scrolling Measured on the live page: inside a scrollytelling column the four-up vmet-strip has 562px and needs 659px, and its overflow-x is `visible`, so the fourth stat (free-running, 4.48% MAPE) was cut off with no way to reach it. The existing four-up rule is a `min-width: 720px` VIEWPORT media query, which cannot see that the column it sits in is 562px wide at a 1440px viewport. Two-up inside `.scrolly-steps`, four-up again past 1500px. Verified against the live DOM: scrollWidth 659 -> 562, no overflow. The wide tables in the same column are fine -- they are `overflow-x: auto`, so they scroll rather than clip. ## Fonts The methodology diagrams hardcoded `font-family="Roboto, sans-serif"` in 54 places and `"JetBrains Mono, monospace"` in 8. The site's body face is Inter (`--body`); Roboto is only the fifth fallback in that stack, so every label in those diagrams rendered in a different typeface from the rest of the site, and none of them would follow a token change. All 62 now use `var(--body)` / `var(--mono)`. ## Validation text Seven blocks trimmed -- the intro, the anchored lead-in and the longest configuration-table cells -- which is also what was making that table scroll furthest in the narrower column. No fact removed. ## Driving the model directly The overview described only the PolicyEngine bridge, so nothing on it told a reader the model can be driven on its own. Adds the direct route: `pe-macro obr-shock`, the equivalent `run_reform` call with no PolicyEngine dependency, the five levers, and a pointer to the caveats -- since two of them do not work. ## Paper Removed the `paper-result` summary paragraph from papers/obr-macro as requested. The .tex is NOT edited: there is no LaTeX toolchain here, so main.pdf could not be rebuilt and the published PDF would contradict its own source. Findings reported separately. Site suite 1564 passed; contract and navigation clean. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Lcj9DDqam9KmVCfhEdnJcJ
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
vahid-ahmadi
added a commit
that referenced
this pull request
Aug 26, 2026
My own fix in #187 reintroduced the bug it repaired. I added @media (min-width: 1500px) to restore the four-up strip on large screens, on the assumption that a wide viewport means a wide column. It does not: the column width is set by the scrolly grid. Measured on the live page at a 1728px viewport, that query matched while the column was still 562px, so the strip went back to four-up and clipped again -- exactly the failure mode, from the other side. The column width is a container property, so this has to be a container query. Two-up is now the floor, and four-up returns only when the column itself passes 700px. Verified against the live DOM: container queries supported, column 588px, columns resolve to 281px + 281px, scrollWidth 562 == clientWidth, no clipping.
vahid-ahmadi
added a commit
that referenced
this pull request
Aug 26, 2026
* Bump the CSS cache key, and stop flex breaking inline links Two bugs, both found by measuring the live page rather than the source. ## The CSS fix from #188 was not reaching browsers style.css is cache-busted by a hand-written `?v=N` in each page's link tag, and it had sat at v=3 since #180. #187 and #188 both changed style.css without bumping it, so any browser that had already loaded `/style.css?v=3` kept serving the old file from cache and never saw the container-query fix -- confirmed on the live page: the deployed CSS contained the rule, the browser did not, and the strip stayed clipped until the stylesheet was reloaded under a different URL. Bumped to v=4 across 62 pages. Worth noting for future CSS changes: the version is manual, so it has to move whenever style.css does, or the fix ships to the CDN and stops there. ## An inline link was breaking mid-word `.vmet-notes-line` was `display: flex`, which makes EVERY child its own flex item -- including the inline <a> in the "gated" note. It shrank to 40px and broke "working paper" across three lines, splitting the word. Replaced with a hanging indent, which reproduces the same key/body look while leaving inline content in normal flow. Only two pages use this component and only one line in it carries a link, so the blast radius is small. Site suite 1564 passed; contract clean. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Lcj9DDqam9KmVCfhEdnJcJ * Bump the CSS version in the topic-page generator too The economy topic pages are emitted by economy/topics.py, which carries its own copy of the stylesheet link. Bumping the committed HTML without bumping the template made the two disagree, and the 'Economy topic pages match committed data' gate caught it. Template updated and all 11 pages regenerated. --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
vahid-ahmadi
added a commit
that referenced
this pull request
Aug 26, 2026
Six models described in six paragraphs made the reader do the comparison themselves: to find out which model answers over which horizon, you had to read all six and hold them in your head. The information is two-dimensional, so it is now drawn that way. The map places each model on one shared time axis -- recent past, coming quarters, three to five years, decades -- and splits UK from US into lanes. Where a model sits IS the answer to "which one do I want": the recent past is boe-svar, a UK reform three to five years out is obr-macro, decades is psl-og, climate scenarios is define-uk, and the US pair sits in coming quarters. Evidence strength is in the shape rather than a label: a solid outline means the model is checked against an external published anchor, a dashed one means no independent outcome benchmark exists. That puts psl-og and define-uk visibly apart from the other four, which is the honest reading and the thing a table buried in a column. The six evidence statements are kept in full, in the key/value grid fixed in #192, so nothing that made this section worth reading is lost -- only the arrangement changed. Details: every pill is a link to its model page, with hover and focus-visible states. Fonts come from --body/--mono rather than a hardcoded family (the #187 lesson). Colours are all tokens, so it follows the theme. Below 720px the map is hidden and the evidence lines carry the section alone, because a two-dimensional layout stops being readable at that width. Verified geometrically: no pill overlaps another, all six sit inside the viewBox, every label falls within its own pill, and no font-family is hardcoded. The Vercel preview is SSO-protected so it could not be eyeballed before merge. Cache key v=6 -> v=7, template included. Site suite 1564 passed; contract and generator clean. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Lcj9DDqam9KmVCfhEdnJcJ
vahid-ahmadi
added a commit
that referenced
this pull request
Aug 26, 2026
Six models described in six paragraphs made the reader do the comparison themselves: to find out which model answers over which horizon, you had to read all six and hold them in your head. The information is two-dimensional, so it is now drawn that way. The map places each model on one shared time axis -- recent past, coming quarters, three to five years, decades -- and splits UK from US into lanes. Where a model sits IS the answer to "which one do I want": the recent past is boe-svar, a UK reform three to five years out is obr-macro, decades is psl-og, climate scenarios is define-uk, and the US pair sits in coming quarters. Evidence strength is in the shape rather than a label: a solid outline means the model is checked against an external published anchor, a dashed one means no independent outcome benchmark exists. That puts psl-og and define-uk visibly apart from the other four, which is the honest reading and the thing a table buried in a column. The six evidence statements are kept in full, in the key/value grid fixed in #192, so nothing that made this section worth reading is lost -- only the arrangement changed. Details: every pill is a link to its model page, with hover and focus-visible states. Fonts come from --body/--mono rather than a hardcoded family (the #187 lesson). Colours are all tokens, so it follows the theme. Below 720px the map is hidden and the evidence lines carry the section alone, because a two-dimensional layout stops being readable at that width. Verified geometrically: no pill overlaps another, all six sit inside the viewBox, every label falls within its own pill, and no font-family is hardcoded. The Vercel preview is SSO-protected so it could not be eyeballed before merge. Cache key v=6 -> v=7, template included. Site suite 1564 passed; contract and generator clean. Claude-Session: https://claude.ai/code/session_01Lcj9DDqam9KmVCfhEdnJcJ Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
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.
The metric strip was clipped, not scrolling
Measured on the live page: inside a scrollytelling column the four-up
vmet-striphas 562px and needs 659px, and itsoverflow-xisvisible— so the fourth stat (free-running, 4.48% MAPE) was cut off with no way to reach it.The cause: the existing four-up rule is a
min-width: 720pxviewport media query, which can't see that the column it sits in is 562px wide at a 1440px viewport.Two-up inside
.scrolly-steps, four-up again past 1500px. Verified against the live DOM:The wide tables in the same column are fine — they're
overflow-x: auto, so they scroll rather than clip.Fonts
The methodology diagrams hardcoded
font-family="Roboto, sans-serif"in 54 places and"JetBrains Mono, monospace"in 8. The site's body face is Inter (--body) — Roboto is only the fifth fallback in that stack — so every label in those diagrams rendered in a visibly different typeface from the rest of the site, and none would follow a token change. All 62 now usevar(--body)/var(--mono).Validation text
Seven blocks trimmed — the intro, the anchored lead-in, and the longest configuration-table cells, which were also what made that table scroll furthest in the narrower column. No fact removed.
Driving the model directly
The overview described only the PolicyEngine bridge, so nothing on it told a reader the model can be driven on its own. Adds the direct route:
pe-macro obr-shock, the equivalentrun_reformcall with no PolicyEngine dependency, the five levers, and a pointer to the caveats — since two of them don't work.Paper
Removed the
paper-resultsummary paragraph frompapers/obr-macroas requested.The
.texis deliberately not edited. There's no LaTeX toolchain on this machine, somain.pdfcouldn't be rebuilt and the published PDF would contradict its own source. Review findings are reported separately — several are material.Testing
Site suite 1564 passed;
site_contract.pyandsite_nav.pyclean.🤖 Generated with Claude Code
https://claude.ai/code/session_01Lcj9DDqam9KmVCfhEdnJcJ