[vahid-ahmadi]

Stacked on #200 (vahid/redesign-r2-integrate). Round 5 UI pass on the front door: home, /models, /connect, /contact, a new /glossary, and the shared shell (footer, sub-tab icons, STYLE_VERSION 20). No numbers were added or changed. Every verdict and failure statement is still there; some moved into expanders.

What changed

• /models#choose: question router. The seven-row accordion is now a grid of model cards you can filter by question (cost a reform, forecast, shock, who gains and loses, decades, climate) and by country. Each card shows the question it answers, the model name and id, the country, and the check class. It links to the model, its evidence, its results and its paper. The old accordion text sits in the card's "What it is, strengths and limits" expander. If no model matches, an empty-state line says so (for example, forecast + US gives "No model here answers that question for that country"). The US-reform gap is now a "Not covered" card in the grid, so it is always visible.
• /connect: copy-paste onboarding.
  • The half-empty two-sided panel is gone. The page is now one column with three numbered steps.
  • Step 1 is now: copy the URL, pick a client (its instructions and terminal command show together), then copy a prompt. The terminal commands used to sit in an expander with an upper-case label.
  • The CLI commands were six nested cards. They are now copy rows (label, command, copy).
  • The 8-tab model picker is now a setup list that shows all eight models at once.
  • Limits are label/value rows. This also fixes the "Uk microdata" casing, and the tools table head is now sentence case.
  • Code no longer uses ligatures, so >= and == show as typed (the hosted.py install line used to show ≥).
• Sub-tab icons. The /models and /connect tabs fell back to a generic document icon. Each tab now has its own icon: signpost, table, link, check, book, plug, terminal, code, wrench, alert.
• /glossary (new). 27 terms from the site's own trust wording (anchor, by construction, free-running, deltas only, outturn, vintage, credible band, drifting random walk, forecast skill, incidence, SVAR/HANK/OLG/SFC, MCP, job, cold start). Each term links to where it matters. The page is linked from the footer, the home verdict list and the contact page, and is in the sitemap.
• Home. The h1 now says what the site is: "Open economic models for UK and US policy". The lede says what you can do, followed by one audience line. The start cards are now questions.
• Contact. Three self-serve routes under the email card: report a wrong number (GitHub issues), run a model, read the glossary.
• Footer. Adds Glossary and Report an issue. site_nav.py --write restamped all 66 pages, and STYLE_VERSION is now 20.

Checks

• pytest: 1180 passed, 1 failed.
• The failure is test_model_cards_quote_the_capability_registry (us-hank Status row). It is not caused by this PR: on us-hank/index.html this branch changes only the footer and stylesheet lines. It comes from the base branch's latest commit and is left for the correctness owner.
• site_contract.py, site_nav.py and every generator --check pass.
• Link check: 66 pages, 0 broken.
• No horizontal scroll at 375 px on any page I touched.

Screenshots: scratchpad/shots/r5/ui-home/ (before_* / v2_* / v3_*).

FRB/US multipliers hub row (d6eb922, macro#209)

The models/validation FRB/US purchases row now reads against CBO's converted ranges and the 2010 FRB-US lines, not "the pegged-rate cross-model range". It matches frb-us/validation (#206) and us-frb-model docs/benchmarks.md.

🤖 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 3:52am UTC

[vahid-ahmadi]

r5 shared components (commits 63c0cea, ab870b2):

• Keep reading row is one level. The where-next section on all 63 pages that carry it was a card of tiles inside a section card. It now draws as the Autumn Budget landing does: the section head (icon badge + title) and the entry tiles sit straight on the canvas, with no white card around them. CSS only, in the r5:ui-home section of style.css. The markup is unchanged, so the hand-written pages, site_nav.where_next and the generators that import it (economy/topics.py, notes/generate_release_notes.py) need no change, and every generator --check still passes.
• Papers cards on /models. The lowercase mono read → is now Read → in the body face. It is pinned to the card bottom, so the row lines up across cards of different lengths.
• The router cards' bottom-aligned action rows and the US HANK Survey link (so the row no longer wraps) came in QA's commits ac7d019 and 63cf919. The 1440 shots confirm both.
• STYLE_VERSION 20 → 21; site_nav.py --write restamped 66 pages.

Checks: pytest 1181 passed, site_contract, all site-provenance generator --checks, and the link check (0 broken) all pass. Screenshots at 1440 and 375 of /pe, /economy/topics/rates, /notes/releases, /papers/boe-svar, /forecasts and /models: no horizontal scroll at either width.

Open: two notes pages (ui-data's area) still carry a lowercase where next kicker in the head. A request to change it has been filed for #205.

🤖 Generated with Claude Code

[vahid-ahmadi]

QA on the combined preview of #204 + #206 + #205 (local branch, not pushed): 66 pages at 1440 and 375 px, axe (WCAG 2.1 AA), keyboard focus, landmarks and headings, anchors, page weight, console.

Fixed here (ac7d019, 63cf919, e1dd4bc):

• /models router: the expander and action row now sit at the bottom of every card (details.about margin was beating .mcard-more); the US HANK link reads "Survey" (screen-reader text "the HANK survey") so the row fits on one line at 1440 and 375 px.
• The pipeline on /models was role="img", which hid its two links from screen readers; it is now role="group".
• The focus ring on the sub-tab rows was clipped by the row's overflow-x; it is now drawn inside the tab.
• AA contrast: .stat-sub on tinted tiles and code comments and the copy button on code blocks were 4.4:1; they now use --text-muted. The axe re-run is clean.
• The "copy" button no longer covers long commands, and the scrolling combine command takes keyboard focus.

Merge note: all three r5 style.css sections appended at the same EOF point. #206 moved its block, and #205 has been asked to move its block, so 204 → 206 → 205 merges without conflicts. Full list: r5/qa.md.

🤖 Generated with Claude Code

[vahid-ahmadi]

Neutrality copy (6928e71), from the r5 neutrality review, every claim re-checked against its source page:

• Home: the record now says CPI landed inside its 68% band and unemployment landed outside its 90% band (it had only mentioned CPI). PE, OBR, FRB/US, US HANK, OG-UK and DEFINE-UK verdicts now state their limits in words.
• /models#validation is the one place the verdict figures live (vphi −17.3%, DEFINE −0.30pp/0.01pp gate, −23.3% emissions, 0.30% MAPE, 64 tests, 49 origins). Home and the router cards use words and link to it.
• 'Every matched year' → the one matched year (2026–27). The US-static line is kept once, in the router gap card (#us-reform-gap). OG-UK 'Scores reforms: Yes' is qualified. 'Nothing quotable' is replaced.
• Connect: the forecast prompt carries a no-skill note. The copied text is unchanged. The HANK prompt now matches its output (liquid-wealth quartiles).
• Glossary: new entries for add-factor, design-gated, paper-anchored, frozen edge, noise floor and release-to-release gap.

🤖 Generated with Claude Code

[vahid-ahmadi]

r5 fresh-eyes pass (merged preview of #204+#206+#205+#219, all 71 pages shot at 1440, 15 at 375). Full ranking: scratchpad r5/fresh-eyes.md. No numbers changed, no failure statement removed; pytest, site_contract, linkcheck and generator --check pass on this branch and on the four-way merge, which merges cleanly.

Pushed 83ab86a on this PR:

• /glossary term index: column link list in a card in place of 36 pills.
• hosted / local tags: neutral outline (orange read as a warning grade).
• /connect step badges: tinted like .meth-step-num.
  Needs a STYLE_VERSION bump at the end (not done here). Not fixed: long inline Python on /connect, 7-line home lede at 375, two adjacent card rows with Open vs Explore on /models (models/index.html had someone else's uncommitted edit).

[vahid-ahmadi]

Phone length (r5): the home lede is now 3 lines at 375. It keeps its first clause; "Read forecasts scored against the outturns" and the engine sentence move into the audience line, wording unchanged. STYLE_VERSION goes 22 → 23 and site_nav.py --write has been run. The six methodology pages keep v=22 here because #206 carries their v=23 line (it edits the lines beside it), so the two merge cleanly. models/index.html: only its stylesheet line is committed; another agent's uncommitted edit is left in the worktree.

Merged base+204+206+205+219: no conflicts, 1248 tests pass. One fails: test_page_is_current, because #219's compare page embeds v=22. Run python3 models/compare/build.py once both are in. Shots: scratchpad shots/r5/plen/home375_ba.png.

🤖 Generated with Claude Code

[vahid-ahmadi]

Glossary, round 5 (8a1bbc4 + cc61422)

• /glossary gets 23 plain-language entries in two new bands, Estimation and scoring (#estimation) and Solving and simulating (#solving): prior and posterior, Minnesota prior, sign and zero restrictions, effective sample size, IRF, FEVD, historical decomposition, band coverage, RMSE, percentage point and pp-quarter, exogenous/endogenous (exogenising), closure, Gauss–Seidel and Newton solvers, unit and explosive roots, output gap, Taylor rule (inertial), VAR and model-consistent expectations, rstar (the HANK code's policy-rate shock, not the neutral rate), MPC, hand-to-mouth, sequence-space Jacobian, Euler equation, Okun relation. Each is one or two sentences with no numbers and no new claims. Each links to the page where the term matters.
• The first use of each term on /connect, /models, /models/side-by-side and /models/validation now links to its entry. Not linked: the /models/papers card, because the whole card is already a link, and the /models/combine routes text, because economy/topics.py (Round 5 ui-data: forecast-evaluation scoreboard with full-height rail, flat archive, chart units and sources #205) generates it.
• I wrote the first-use link list for Model pages: one-column methodology, one overview order, names not ids #206 (81 pages/terms), Round 5 ui-data: forecast-evaluation scoreboard with full-height rail, flat archive, chart units and sources #205 (incl. the generator-owned combine text) and Models: 'Same question, every model' comparison view (/models/compare) #219 (8 terms) into r5/ui_requests.md. I left out "coverage" where it means a data date range and "prior" where it means "previous".
• Also on this branch: dd3294f restamps the six methodology pages to style.css?v=23. That fixes the stale-nav CI failure. Model pages: one-column methodology, one overview order, names not ids #206 merged it, keeping its own head lines (7f2ce5c), and Models: 'Same question, every model' comparison view (/models/compare) #219 merged it too (5b0ce02).
• Checks: base + r5 UI: question router on /models, copy-paste /connect, glossary, home and contact #204 + Model pages: one-column methodology, one overview order, names not ids #206 + Round 5 ui-data: forecast-evaluation scoreboard with full-height rail, flat archive, chart units and sources #205 + Models: 'Same question, every model' comparison view (/models/compare) #219 merge with no conflicts. On that merge, pytest (PYTHONPATH=integration/src) gives 1250 passed. site_contract, site_nav, topics.py --check, compare/build.py --check and the link check (0 broken) all pass. Nothing merged.

🤖 Generated with Claude Code