Schools without KS2/KS4 results (special post-16 institutions, sixth-form
centres, PRUs, new schools) come back from the marts LEFT JOIN with NaN in
every numeric column. school_info passed those raw pandas values straight
into JSONResponse, which renders with allow_nan=False, so the detail
endpoint 500d and the frontend turned that into a 404 on every such SEO
landing page.
Run school_info values through convert_to_native (the same treatment
yearly_data already gets), add backend unit tests plus a pytest step in PR
checks, and an e2e journey that finds a results-less school via the search
API and asserts its page renders.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The geolocation shortcut lived in the discovery strip below the results,
away from the search. Move it directly under the hero search input, paired
with the postcode hint, so the two ways to find nearby schools ("type a
postcode" / "use my location") read as one idea and are visible at first
glance.
- FilterBar gains optional onNearMe/geoState/geoError props and renders the
teal "Use my location" pill (with spinner + error) in hero mode; the
geolocation flow itself still lives in HomeView.
- Remove the now-duplicate near-me button and its dead CSS from the
discovery section.
- Refresh the search hint copy to pair with the button.
- e2e: assert the "use my location" shortcut renders in the hero.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The results-view map's fullscreen button called requestFullscreen(),
which iOS Safari doesn't implement (fullscreen is video-only there), so
tapping it did nothing on iPhones — the same gap already fixed for the
school hero map.
When the Fullscreen API is missing or its promise rejects, fall back to
a fixed-position overlay (.fsFallback, z-index 5000) driven by state,
locking body scroll while open. Leaflet re-measures on window resize, so
dispatch a resize when fullscreen toggles (the CSS overlay fires none) or
the map would fill only part of the screen. Native fullscreen is
unchanged.
New e2e journey deletes Element.requestFullscreen on a mobile viewport,
opens the results map fullscreen, and asserts the exit control appears
then releases; it fails against current production, reproducing the bug.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
On a location search the results header shows the view toggle and the
sort <select> side by side. The select sizes to its widest option
('Highest Reading, Writing & Maths %', ~273px), so on a phone its right
edge ran ~46px past the viewport and was clipped off-screen.
On mobile let the select flex into the remaining space with min-width:0
so its label truncates instead of overflowing, and keep the view toggle
from shrinking. Verified live at 390px: the select now sits fully within
the viewport.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The mobile chart legend stacked one school chip per line, so up to five
schools pushed the chart down and left the plot cramped. Switch the chip
row to a two-column grid; each chip fills its column and truncates its
name with an ellipsis (full names remain on the school cards and in the
tooltip). Five schools now take three rows instead of five, giving the
chart noticeably more height.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The test picked the first two /school/ links from a 'primary' search and
asserted exactly two mobile chips. But a 'primary' search can return
all-through schools (e.g. 'Hessle High School and Penshurst Primary')
that classify as secondary, so the two picks can split across phases —
the active phase then holds one school and the chips are correctly gated
out (they need ≥2 in the active phase), while the canvas still shows one
line. That's a test artefact, not a bug.
Pick three schools instead: across two phases the auto-selected majority
phase always holds ≥2, so the chip legend is guaranteed. Assert ≥2 chips
(the majority may be 2 or 3). Verified against staging.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
next.config.js rewrites() bakes its destination into the build
(routes-manifest.json), capturing FASTAPI_URL at build time. Because one
frontend image is promoted staging->prod, the baked backend host forced
every environment to name the backend service identically; staging names
it 'backend_stg', so the browser's /api/* calls proxied to the baked
'http://backend' and failed with getaddrinfo ENOTFOUND backend. (SSR was
unaffected because lib/api.ts reads FASTAPI_URL at runtime.)
Replace the rewrites with route handlers that read FASTAPI_URL per
request:
- app/api/[...path]/route.ts — transparent proxy for all methods, streams
the response, strips hop-by-hop headers, and returns 502 on upstream
failure instead of crashing.
- app/sitemap.xml/route.ts — proxies the backend sitemap (robots.ts points
crawlers here).
The same promoted image now adapts to whatever the backend is called in
each environment. Verified: production build succeeds with /api/[...path]
and /sitemap.xml as dynamic routes and an empty rewrites manifest.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Removes the 'What Parents Say' section and all supporting elements:
Frontend:
- Drop the OfstedParentView type, the parent_view field, the survey
section and the 'X% would recommend' callouts in the primary and
secondary detail views, the Parents nav item, and the parent-view CSS.
Backend:
- Remove the FactParentView model, its loading in data_loader, and
parent_view from the school-details API response.
- Bump SCHEMA_VERSION to 6 and add an idempotent drop step
(DROP TABLE IF EXISTS marts.fact_parent_view) to the CLI migration;
add scripts/sql/drop_fact_parent_view.sql to apply directly to the
dbt-owned marts DBs on staging and prod.
Pipeline:
- Delete the stg_parent_view + fact_parent_view dbt models and their
source/schema entries, the tap-uk-parent-view Meltano extractor, and
the monthly Parent View DAG; drop it from the Dockerfile and the
staging bootstrap docs.
The rest of dbt (which builds every mart the app reads) is untouched.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The compare view refetches /api/compare on the client after SSR; on any
failure the catch nulled comparisonData, destroying the working
SSR-provided chart. A transient error (or staging's broken external /api
proxy) should not blank a comparison the user is already viewing — keep
the existing data instead.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The compare chart squashed clustered schools into a thin band (y pinned
0-100) under an in-chart title + per-school legend that ate ~40% of a
300px card, leaving converging lines indistinguishable on phones.
- Auto-fit the y-axis to the data on all viewports (computeYBounds in
lib/utils: padded + min-span for percentages, symmetric around 0 for
progress, fitted for scores; negative pct-named trend metrics are not
zero-clamped).
- Distinct point style per school (circle/triangle/rect/rectRot/star)
as secondary encoding for convergence and colour-blindness.
- Mobile: drop in-chart title/legend/axis titles; add a chip row (colour
dot + name) that doubles as tap-to-focus — highlights one school's
line and dims the rest. Chart card 300px -> 340px, nearly all plot.
- Fix a latent colour mismatch: datasets were built from Object.entries
whose integer-like URN keys enumerate in ascending numeric order,
desyncing line colours from card colours; the chart now receives the
ordered school list.
- Union years across schools instead of taking the first school's.
- Extract PerformanceChart's matchMedia pattern into hooks/useIsMobile.
Unit tests for metricKind/computeYBounds; e2e journey covers the mobile
chips and focus toggle.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
iOS Safari has no Element.requestFullscreen (fullscreen is video-only),
so tapping the map band or 'View on map' silently did nothing on
iPhones. Fall back to a fixed-position CSS overlay driven by state when
the Fullscreen API is missing or its promise rejects, locking body
scroll while open. The Leaflet map already re-measures via the shared
isFullscreen flag. New e2e journey simulates the iOS condition by
deleting the API and asserts the overlay opens and closes; it fails
against the current production build.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Staging doesn't always carry the full data history, so selecting the
oldest year legitimately returns no rows and fails the promotion gate.
Select the most recent explicit year instead: the default view already
proved it has rows, so an empty table after selecting it can only mean
the year query param was rejected — the regression this test guards.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The rankings endpoint validated year with le=2100, but the database
stores academic-year codes like 201819, so any explicit year selection
returned a 422 and the rankings page rendered its empty state. Widen
the bound to cover the codes and extend the e2e journey to pick a
specific year and assert the table stays populated.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The admissions card stacks year/trend views in one grid cell and keeps the
inactive view visibility:hidden; its canvas is first in the DOM. Use
canvas:visible instead of scrolling. Verified: full suite passes against prod.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PqGhF93UrpDNvXBLMjJENL
The same frontend image runs on staging and prod (build once, promote), so a
build-time env var can't tell them apart. Umami's data-domains attribute
scopes the tracker client-side: events only fire when location.hostname is a
production domain, so staging traffic and the E2E journeys never pollute the
dashboards.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PqGhF93UrpDNvXBLMjJENL
- ai_review.py now pipes the diff through headless Claude Code (claude -p,
--output-format json) authenticated with CLAUDE_CODE_OAUTH_TOKEN from
'claude setup-token' — subscription auth, no Anthropic API billing
- stdlib-only script (urllib instead of requests/anthropic)
- PR comments posted with the existing REGISTRY_TOKEN secret; the separate
GITEA_TOKEN secret is no longer needed
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PqGhF93UrpDNvXBLMjJENL
- pr-checks.yml: PR gate — frontend typecheck+jest, backend import smoke,
image builds (no push), Claude AI review posted as PR comment (severe
findings block merge)
- deploy.yml (replaces build-and-push.yml): merge to main builds+pushes
images tagged sha-<sha>/staging, deploys the staging Portainer stack via
webhook, runs Playwright E2E journeys against staging, then retags the
verified images :prod (previous kept as :prod-previous) and deploys prod
- docker-compose.portainer.staging.yml: second Portainer stack — :staging
images, sc_staging_* names, own macvlan IPs, Airflow on 8081; data
bootstrapped from source via the staging Airflow DAGs
- prod compose now pins :prod instead of :latest (only the promotion step
moves it; :latest is no longer published)
- e2e/: 6 Playwright journeys (search, postcode, detail, compare, rankings)
driven by BASE_URL — the promotion gate
- scripts/ci/ai_review.py: Claude review with structured JSON findings
- docs/DEPLOY.md: full SDLC doc incl. one-time setup checklist and rollback
- replaced removed 'next lint' with tsc typecheck; fixed stale jest tests
(slug URLs, N/A formatting, stable trend, fake-timer setup)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PqGhF93UrpDNvXBLMjJENL
Phones showed only the poetic h1 and a bare search box — no coverage,
scope, or freshness statement above the fold for the 63%-of-entries,
56%-mobile audience. One compact line ('24,000+ English schools — SATs,
GCSEs, Ofsted & admissions, side by side. Updated for 2026/27') now
replaces the hidden eyebrow + full paragraph at ≤640px, costing ~2 short
lines. Desktop copy unchanged.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Table auto-layout sizes columns by unwrapped header text, so labels like
'Reading, Writing & Maths Combined Higher %' pushed the value column —
and the table — past the viewport. The label now renders inside a block
span capped at 110px (84px on phones), forcing multiline and keeping the
table fully in view at any metric.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Rankings mobile: the metric value — the point of the page — sat behind
a sideways swipe at 390px. The Area column now folds into a subline
under the school name, leaving Rank | School | Value to fit the
viewport with no horizontal scroll.
Rankings default metric becomes 'expected standard' (rwm_expected_pct);
'higher standard' stays available but no longer frames every school's
headline number in the terms parents least understand.
MetricTooltip was hover-only and display:none on phones — the mobile-
primary audience had zero access to the Attainment 8 / Progress 8 /
EBacc definitions. It is now a real button: tap/click/keyboard
toggleable with outside-click and Escape dismissal, 24px target,
shown at all viewports.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The seed effect only adopted the URL's schools when localStorage was
empty, so a recipient who had ever used compare silently saw their own
old shortlist instead of the shared one. Explicit URL urns now replace
the stored selection on load (then persist as usual); bare /compare
still restores the visitor's own selection. Adds replaceSchools() to
the comparison context. Card values also switch to CHART_TEXT_COLORS.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Live axe-core inventory found 13 failing fg/bg pairs (3.12:1 coral text/
fills, 4.0-4.5 teal near-misses, 2.7 gold badge, 1.9 chart-colour text).
All replacement values validated ≥4.5:1 against every background they
sit on:
- --accent-coral-dark deepened to #b04a2e and used for coral text roles
(nav active, back/map links, kickers, chips) and coral fills under
white text (btn-primary, phase tabs, section-nav compare, chart chips);
new --accent-coral-darker #9c3f26 for their hovers. Decorative coral
(pins, bars, borders, focus ring) keeps the brand #e07256.
- --accent-teal darkened to #296f6f (rankings values, metric rows,
ofsted grades, eyebrows all pass).
- Ofsted green #3c8c3c→#2f7a2f; off-token gold #b8920e→--accent-gold-text.
- Gender-split pink text #b45778→#a04a68; admissions tip numerals to a
3.8:1 muted brown (28px display text).
- New CHART_TEXT_COLORS: AA-dark counterparts of the Chart.js series
palette for compare-card values (swatch dots keep true series colour).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
On ≤640px the floating Compare button becomes a 40px round glyph (+ / ✓,
aria-labelled) like the section-nav compare icon, instead of a full-width
text pill over the map. The OSM attribution moves to the map band's
top-left in preview so it no longer collides with the school name sliding
up under the fade; fullscreen keeps Leaflet's default bottom-right since
the zoom control occupies the top-left there.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Regression from d4d9ae5 (map-blended hero): .headerHasMap .actions was
absolutely positioned intending to float over the map, but its containing
block was .headerContent — made position:relative in the same commit — so
top:14px anchored it to the title block below the map. On mobile the H1
spans the full width and the legacy '.actions { width: 100% }' rule still
applied, stretching the glassy button across the school name and leaving
a ~6px legible sliver between it and the map fade.
Anchor .actions to .header by moving position:relative/z-index:3 from
.headerContent to .titleSection (the element that actually needs to sit
above the fade), and set width:auto on the floating variant so the mobile
full-width rule can't reach it. Applied to both detail views.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Recheck with 'B91 3DL' via the search box shows the geocoded radius
search works (distances, nearest-first, radius selector, map toggle).
Original P0.3 tested only a partial postcode, which silently falls back
to text search — refiled as P2.0 (silent mode switch on partial input).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
tudor