Public site suite¶

packages/public-site, Vitest with jsdom. 34 files · 275 tests · all passing · 71.32s without file parallelism.

Re-measured at v2026.10.1: nothing changed, and the figures held

Re-run on 9 October 2026 for v2026.10.1 (main at ebec288), in the working checkout on acc at 0ea4985, a tree identical to ebec288, with npm run test:serial --workspace=packages/public-site on Node 22.23.3: 34 files · 275 tests · all passing · Duration 71.32s, coverage 96.68 / 97.41 / 95.09 / 97.23. v2026.10.1 did not touch this package — git diff 0625d48 ebec288 -- packages/public-site is empty — and every count reproduced. Branches read 97.41 against 97.40 on 3 October: one hundredth of run-to-run noise on identical code, in src/pages (97.29 → 97.31), which Coverage explains.

History: re-measured at v2026.10.0, one new file, ten new tests

Re-run on 3 October 2026 for v2026.10.0 (main at 0625d48), in the working checkout on acc at 0e3eed8, a tree identical to 0625d48, with npm run test:serial --workspace=packages/public-site on Node 22.23.2: 34 files · 275 tests · all passing · Duration 76.58s, coverage 96.68 / 97.40 / 95.09 / 97.23, against 33 · 265 and 96.60 / 97.34 / 95.07 / 97.17 on 30 September. The duration is a serial run's and not comparable with the 24.09s parallel figure of 30 September.

One change accounts for all of it: v2026.10.0's problem details (#216). The backend now answers every error as RFC 9457 application/problem+json, and the site reads the message from a problem's detail. src/lib/problem.test.ts is new, 9 tests — problemMessage returning a problem's detail, still reading a legacy envelope's error.message, and falling back for anything else — and lib/api.test.ts gained one, 13 → 14. lib/problem.ts is at 100 on all four measures; lib/api.ts reads 96.55% branches, against 97.29% on 30 September.

History: re-measured at v2026.09.15, one new file, thirty new tests

Measured on 30 September 2026 for v2026.09.15: 33 files · 265 tests, against 32 · 235 on 28 September.

Two changes accounted for all of it. v2026.09.14's link previews (6c3fe10) added src/indexHtml.test.ts — the Open Graph and robots tags in index.html, filled per build mode from the committed .env files, 23 tests — and three cases to scripts/prerender.test.ts, which now reads the site's origin, API and indexability through Vite's loadEnv, builds a robots.txt that shuts crawlers out on ACC, and replaces the shell's canonical link rather than adding a second. v2026.09.15's branch-margin commit (73a6764) added three cases to lib/api.test.ts and one to Footer.test.tsx, and made the package's only source change: lib/api.ts now reads import.meta.env directly so its base-URL fallbacks can be stubbed, taking that file from 83.78% branches to 97.29%.

The public site is the auth-free search and rule-catalogue package, measured with npm run test:serial --workspace=packages/public-site, coverage included, on Node 22.23.3, the version .nvmrc names. The suite grew in every window up to v2026.10.0 — 30 files and 204 tests at v2026.09.5, 31 and 225 at v2026.09.7, 32 and 231 at v2026.09.9, 32 and 235 at v2026.09.11 through v2026.09.13, 33 and 265 at v2026.09.15, 34 and 275 at v2026.10.0 — and held at that in v2026.10.1.

Coverage is 96.68% statements · 97.41% branches · 95.09% functions · 97.23% lines, and the package passes the per-file 80% branch floor that its vite.config.ts enforces rather than merely records: no file in it is below the line, and since v2026.09.15 none is below 85 either. The lowest by branches are lib/useQueryState.ts and pages/herkomst/HerkomstTrace.tsx, both at 87.50%.

History: branches fractionally down at v2026.09.11, and why

Against v2026.09.9 — 95.82 / 96.45 / 94.89 / 96.33 — v2026.09.11 gained on statements, functions and lines while branches gave up 0.14 of a point. Detail.tsx was at 100% statements and 97.84% branches after its change, so nothing regressed; what moved the package figure was lib/api.ts, which gained branch surface it was not yet fully exercising (83.78% branches). That is the ordinary shape of adding a guarded code path and testing its main case first — and v2026.09.15 tested the rest, taking the file to 97.29%.

Inventory¶

Re-derived on 9 October 2026 at ebec288, identical to 3 October at 0625d48. The Files column accounts for all 34 files and the Tests column sums to 275, the runner's total. The per-area test counts are taken from the source — each test file parsed, with it.each tables and the describe.each in indexHtml.test.ts expanded — because this pass used Vitest's console reporter, which prints only the package total; the same method reproduces the 24 September runner-derived table exactly at 963fe24.

Area	Files	Tests	Covers
`src/pages`	16	137	Includes a full `herkomst/` provenance-explorer sub-area (`HerkomstExplorer`, `HerkomstTrace`, `HerkomstChip`, `HerkomstBackground`, `herkomstConcepts`, `herkomstData`, `herkomstScroll`, `herkomstTrail` — 8 files) plus the generic `SectionIndex` / `Regelcatalogus` / `Results` / `Detail` (25) / `Woordenboek` / static pages
`src/lib`	8	59	`slug` (kept identical to the backend's slugifier by design), `useQueryState` (URL-backed filters), `search` (`highlight()`), `api` (the typed `/v1/public/` client, 14 — three new in v2026.09.15 for its base-URL fallbacks, one in v2026.10.0), `problem` (9, new in v2026.10.0* — reading the message from an RFC 9457 problem's `detail`, or a legacy envelope's `error.message`), `sectionHits` (`mapToHits()`), `prerenderedData` (the seeded-render reader), `buildInfo`
`scripts/`	2	24	`prerender.test.ts` (14 — `escapeHtml`, `readSiteEnv` per build mode, `buildRobots`, `buildSitemap`, `injectIntoShell`), `check-bundle.test.ts` (10 — the build-time gate that fails if any auth or telemetry string ships in the bundle)
`src/components`	4	23	`chrome` (7), `Footer` (7), `StatusTag` (5, new in v2026.09.9), `TechDetails` (4)
`src/indexHtml.test.ts`	1	23	New in v2026.09.14. The link-preview and robots tags in `index.html` as `vite build --mode <mode>` writes them, for production, acceptance, development and test (five tests each), plus the fixed Dutch copy, the title and the card image's size
`src/App.test.tsx`	1	5	Routing shell — every route registered, `<html lang>` synced to the language switch
`src/i18n`	1	3	NL/EN dictionary key parity
`src/staticwebapp-csp.test.ts`	1	1	Guards the shipped CSP header — a regression here silently breaks the org-logo host

Both columns are current, and they sum

Earlier versions of this page carried a Files column from one date and a Tests column from 19 August that summed to 134 against a measured 225, with a warning attached. Re-derived, the columns agree: 34 files, 275 tests. Against 30 September, only src/lib moved — 7 files and 49 tests to 8 and 59, problem.test.ts and one case in api.test.ts. Between 24 and 30 September, src/lib gained 3, scripts/ 3 and src/components 1, and indexHtml.test.ts arrived with 23.

The prerender test runs in the node environment rather than jsdom since v2026.09.14: prerender.ts now reads the .env files through Vite's loadEnv, and the esbuild that Vite loads refuses to start under jsdom. Both it and indexHtml.test.ts set the inherited VITE_* variables aside first, because Vitest has already put the test mode's values into process.env, which loadEnv lets override the file — without that, every mode would read the test file.

StatusTag is the file v2026.09.9 added

src/components/StatusTag.test.tsx arrived with the StatusTag.tsx it covers, and both are at 100% statements, branches, functions and lines — 3/3 statements, 2/2 branches, 1/1 functions, 3/3 lines. Its five tests run in 91ms:

renders the known status example with its own class
renders the known status wip with its own class
renders the known status e2e with its own class
falls back to a neutral class for a status label it does not recognise
always shows the label as text, never colour alone

The last is an accessibility property rather than a rendering detail: a status conveyed only by colour fails for anyone who cannot distinguish the palette, and it is the kind of regression a snapshot test would happily wave through.

Per-area coverage was also re-derived on 9 October and reconciles to the package total — see Coverage.

The statement-to-branch gap this package was once known for is gone. It was the widest of the five at 16.4 points, 86.82% statements against 70.39% branches. Today branches sit above statements — 97.41% against 96.68% — which is what a campaign against a per-file branch floor looks like once it lands. The margin had narrowed from 0.63 of a point to 0.39 by 28 September; the lib/api.ts work widened it again, to 0.74 on 30 September; it was 0.72 on 3 October and is 0.73 now.

Playwright suite¶

packages/public-site/e2e/publiek.spec.ts (npm run test:e2e --workspace=@ronl/public-site) against real /v1/public/* data with no mocks — search → filter → detail → back with URL preservation, a deep link with pre-applied filters, keyboard-only navigation, and three axe-core accessibility scans (home, results, a detail page) asserting no critical or serious violations.

Re-run on 9 October 2026: 6/6 serially, 2 timeouts in the parallel run — as on 3 October

Measured 9 October 2026 for v2026.10.1, against the developer's already-running backend: default run, six parallel workers, 4 passed, 2 failed, 16.8s — the same two search-journey tests as on 3 October, each the same TimeoutError after 10s waiting for the Regel filter checkbox — and 6 passed in 6.4s with --workers=1 against the same backend. Recorded the same way: 6/6 serially, 2 timeouts in the parallel run, not a defect. Two passes a week apart failing the same wait under six workers is a pattern, and the thing to fix at the spec's own boundary before this suite goes into CI.

Measured 3 October 2026 for v2026.10.0, against the developer's already-running backend, the first run since 30 August:

Default run, six parallel workers: 4 passed, 2 failed, 17.2s. Both failures were the search-journey tests — publiek.spec.ts:6 (search → filter → detail → back preserves the filtered URL) and :27 (a deep link with filters pre-applied renders those filters checked) — and both the same TimeoutError: 10s waiting for getByRole('checkbox', { name: /Regel/ }).first() to be visible.
Re-run with --workers=1: 6 passed, 7.8s, against the same backend.

Recorded that way — 6/6 serially, 2 timeouts in the parallel run — and not as a defect: nothing failed on its own, and the detail-page axe scan, which needs search results as much as the two that failed, passed in the parallel run. A missing backend fails three, serially as well (see below). The config sets fullyParallel: true and no workers, so a default run takes Playwright's default worker count — six on that machine — and retries is 2 in CI and 0 locally. The spec is unchanged since 2443adc and still runs in no workflow.

Before this, the figures were 19 August 2026, against v2026.08.19: 6 tests, 6 passed, 0 failed, 0 flaky, 0 skipped, 8.9s, and a 30 August inventory count of six.

Unlike the frontend suite, Playwright starts its own dev server for this package — playwright.config.ts declares a webServer running npm run dev on :5175, with reuseExistingServer on outside CI. Only the backend needs to already be running, on whatever VITE_API_URL points at: these specs hit real /v1/public/* search results rather than mocks, which is why three of the six fail on timeouts without it.

Setting E2E_BASE_URL skips starting the local server entirely and points baseURL at an already-deployed site instead, which is how the suite is used for post-deploy verification against ACC. Re-read from the config at 86af73e on 24 September 2026 and unchanged at ae06c9e, 0625d48 and ebec288: webServer on :5175, no globalSetup of any kind, and the backend dependency stated in the config's own header comment rather than checked before the run.

CI¶

Public-site is the package that had a real CI test gate before the others did. Both azure-publicsite-acc.yml and -prod.yml run npm run lint, npm run type-check, then npm test before building, and the build itself gates on a prerender step and a bundle-cleanliness check. A failing test blocks the deploy.

Since v2026.09.14 the build step also runs node scripts/check-og.mjs acceptance (or production) against the built dist/, before the upload. The unit tests prove the template, the .env files and the prerender agree; this proves the files actually shipped are the ones for their environment — every prerendered page, not only the home page, since each is a copy of the shell. An ACC build carrying the production card, or letting crawlers in, fails the deploy rather than reaching every link unfurler's cache or a search index.

Since 19 September 2026 it blocks the merge as well. The acc ruleset (acc supply-chain gate) lists Build and Deploy ACC Public Site among its required status checks, alongside audit, scan, build, Build and Deploy ACC Frontend, Build and Deploy ACC PA Demo and, since v2026.10.1, lockfile-review — so a red suite here now stops the pull request rather than merely reporting against it. Read from gh api repos/sgort/ronl-business-api/rulesets on 20 September 2026, unchanged when re-read on 30 September, and re-read as the rules in force on acc on 9 October. Since v2026.10.1 azure-publicsite-acc.yml also runs on a change to package-lock.json or package.json, so a lockfile-only pull request — Renovate's above all — is built and tested here too.

This page said the opposite until 20 September 2026

It previously stated that the only required check on acc and main was audit, so "a red suite here has to be read rather than relied on to stop a pull request". That was true when written and is now false for acc. It remains true for main, deliberately — main requires audit and, since the end of September, scan, but no suite: no production workflow has a pull_request trigger, so no production check could be required of a promotion. See Overview → What actually gates a merge.

Since the per-file branch floor lives in vite.config.ts, this same npm test step is also where coverage is enforced — the required check and the coverage gate are one command.