Coverage

Measured on 11 October 2026 against v2026.10.2, the release in production (main at 7117914), in the working checkout on acc at c51ec0a — a tree identical to 7117914 — after npm run deps:check reported "Dependencies are ready.", on Node 22.23.3, the version .nvmrc names. Each package was measured serially, one workspace at a time: the backend by npx jest --coverage --runInBand, the Jest command its test:serial runs, and the four Vitest packages by their own npm run test:serial. All five collect coverage, as their npm test counterparts do. Every figure on this page is read from the coverage/coverage-final.json each of those runs wrote, and checked against the run's own totals.

Read the backend from coverage-final.json, not from a stale summary

None of the five runner configs asks for istanbul's json-summary reporter, so no ordinary run writes coverage/coverage-summary.json. A copy left in packages/backend/coverage/ by an earlier run that did ask for it — the one in this checkout dates from 30 September — is not overwritten, and reads 98.90 / 95.14 / 97.98 / 99.20: the 30 September figures, not today's. The 11 October backend figures below come from the coverage-final.json and lcov.info that run wrote. To get a summary file, ask for it, as Running the tests does with --coverageReporters=json-summary.

Package	Statements	Branches	Functions	Lines	Δ branches since v2026.08.36
Backend	98.78%	94.95%	97.95%	99.10%	+4.94
Frontend	95.35%	93.18%	91.48%	95.85%	+12.85
pa-cockpit	92.61%	93.89%	89.57%	93.13%	+18.34
pa-demo	93.47%	95.65%	85.00%	92.85%	+8.70
Public site	96.68%	97.41%	95.09%	97.23%	+27.02

Against the 9 October reading of v2026.10.1, pa-cockpit, pa-demo and the public site reproduced all four figures to the decimal, on packages v2026.10.2 did not touch. The backend moved on two measures, by a hundredth or two — statements 98.77 → 98.78, branches 94.97 → 94.95, functions and lines unchanged — all of it in services/operaton.service.ts, which now picks the decision document of the path a case took and reads 98.41 / 92.78 / 96.93 / 98.80. The frontend fell by three to eight hundredths on all four — 95.42 → 95.35, 93.21 → 93.18, 91.54 → 91.48, 95.93 → 95.85 — on code whose only change was the changelog data, and in one area row only, CaseworkerDashboardV2: run-to-run noise, set out in the last two decimals warning below.

History, 9 October: against the 3 October reading of v2026.10.0, the two packages v2026.10.1 changed moved in opposite directions. The frontend rose on all four — 95.23 → 95.42, 93.02 → 93.21, 91.21 → 91.54, 95.76 → 95.93 — on new code that arrived well covered and on the DVTP sections leaving. The backend fell by a tenth or two on three — statements 98.91 → 98.77, branches 95.21 → 94.97, lines 99.18 → 99.10, functions up 97.92 → 97.95 — on the eDOCS-as-a-person work: services/edocs.service.ts took on most of the new branches and reads 92.95% of them, and the new auth/entra-token.service.ts arrived at 91.48%. pa-demo reproduced all four figures to the decimal, and pa-cockpit and the public site all four but one hundredth each — pa-cockpit functions 89.56 → 89.57, public-site branches 97.40 → 97.41 — on packages v2026.10.1 did not touch: git diff 0625d48 ebec288 -- packages/pa-cockpit packages/public-site packages/pa-demo is empty. That hundredth is the noise the last two decimals warning below describes.

The repository-wide figure, and why it is not an average of that column

Across all five workspaces together: 96.54% statements · 94.13% branches · 93.03% functions · 97.01% lines.

Those come from summing the covered and total counts across the five workspaces' coverage reports and dividing once — 14744/15271 statements, 9514/10107 branches, 3417/3673 functions, 13329/13739 lines — not from averaging the five percentages in the table. Averaging the column would weight pa-demo's 92 statements exactly as heavily as the backend's 6889 and give a repository-wide branch figure of 95.02% instead of 94.13%: nearly a point of pure arithmetic error, in the flattering direction. The denominators differ by two orders of magnitude here, so the distinction is not academic. On 11 October the counts were summed from each workspace's coverage/coverage-final.json, which every one of the five runs writes — each file's timestamp matching that day's run; each workspace's own total reproduces the table row above exactly, the four Vitest totals match the covered/total counts in the run's own Coverage summary block, and the backend's match the lcov.info its run wrote.

Against the 9 October measurement of v2026.10.1 every repository-wide figure fell by two or three hundredths — 96.57 → 96.54, 94.15 → 94.13, 93.05 → 93.03, 97.04 → 97.01 — on a body of code that grew by 10 statements and 8 branches. Most of the fall is the frontend's noise, which took one to four off each of its covered counts; the backend's new code is 10 statements, all covered, and 8 branches, 7 of them covered.

History: against the 3 October measurement of v2026.10.0 the repository-wide figures held or rose in the second decimal — statements 96.51 → 96.57, branches 94.13 → 94.15, functions 92.83 → 93.05, lines 96.97 → 97.04 — on a body of code that grew by 544 statements and 331 branches. That is the shape of a feature release: new code arriving roughly as well covered as the old.

Which frontend run the aggregate uses — like for like since 3 October

Every pass from 24 to 30 September took the frontend's figures from its default parallel run, the one CI makes. The 3 and 9 October passes ran every workspace with test:serial, the frontend included, and so did the 11 October pass, so the frontend row comes from a serial run and the 3 → 9 → 11 October comparison is like for like. (11 October also made a default parallel run of the four Vitest workspaces, to collect per-file test counts; none of this page's figures is taken from it.) Serial and parallel runs of the same code do not report identical v8 coverage.

The size of that effect is on record. On 28 September, when the parallel run carried one contention-only failure, the green serial run of the same suite reported slightly lower frontend coverage — 93.16 / 89.88 / 87.84 / 93.98 against 93.24 / 89.91 / 87.98 / 94.07, four statements and two functions out of 4971 and 1423. That is v8 worker-attribution noise, the same phenomenon the last two decimals warning below describes. So part of the frontend's tenth-of-a-point fall on 3 October was the change of run, not the code, and a frontend difference of that size against 30 September should not be read as a regression. The backend's figures have a sensitivity of their own — see that warning.

Every package moved, and branches moved most

v2026.09.2 extended the backend-only coverage campaign to all five workspaces with a test runner, against a per-file 80% branch floor: 53 files were below it, none are now. That is why the branch column moved furthest and why public-site — which started lowest at 70.39% — moved most.

That was unusual and worth noting as a check that the run was sound: the pass before it had three packages reproducing their figures to the decimal, while in the campaign pass every package gained on every measure — which is what a campaign targeting all of them should look like. The table above is several releases further on. It moved only in decimals until v2026.09.14, then by whole points again on the margin work, and in v2026.10.0, v2026.10.1 and v2026.10.2 by decimals once more; what changed is set out below.

The floor is a gate, and it is per file

It used to be a convention held by review, with no threshold configured anywhere. v2026.09.6 configured it in all five runner configs, and the two runners express the same rule differently:

Workspace	Config	How the per-file floor is written
backend	jest.config.js	coverageThreshold: { './src/**/*.ts': { branches: 80 } }
frontend	vite.config.ts	coverage.thresholds: { branches: 80, perFile: true }
pa-cockpit	vitest.config.ts	coverage.thresholds: { branches: 80, perFile: true }
pa-demo	vite.config.ts	coverage.thresholds: { branches: 80, perFile: true }
public-site	vite.config.ts	coverage.thresholds: { branches: 80, perFile: true }

The Jest form is a glob key rather than global: Jest applies a glob threshold to each matching file individually, which is what makes it a per-file floor and not a package average. Vitest needs perFile: true alongside the number to say the same thing. Get either detail wrong and the threshold silently becomes an average, which is the one thing it exists not to be.

A file below the line exits the run non-zero and names that file, locally and in CI alike — npm test already collects coverage in every workspace, so nothing needed a separate coverage job. The five thresholds are unchanged at c51ec0a — the frontend's vite.config.ts changed in v2026.10.1, but only to add the tenant-pages build plugin, and no runner config changed in v2026.10.2 — and no file was named in any of the five runs of this pass; reading every file entry in the five workspaces' coverage reports on 11 October confirms it independently: zero of 337 files below 80% branches — 105 in backend, 134 in frontend, 38 in pa-cockpit, 16 in pa-demo and 44 in public-site. (Since 3 October the backend gained four source files: auth/citizen-services.ts and services/edocs-author.ts at 100 on all four measures, routes/edocs.access.ts at 98.52 / 89.28 / 100 / 100 and auth/entra-token.service.ts at 95.87 / 91.48 / 92.85 / 96.77. The frontend gained eight and lost two: LoginChoice/AccessDeniedDialog.tsx (95.45% branches) and SingleBoardLanding.tsx (100% branches, 91.66% functions), services/board-request.ts (91.66% branches) and services/landing.ts, login-choice/landing-login.ts and single-board.copy.ts, and pages/citizen/citizenServiceUi.ts and heusdenpasTestCases.ts — the last five at 100 on all four, three of them data with no branches at all — while CaseworkerDashboard/DvtpStartSection.tsx and DvtpTakenSection.tsx went. vite-plugin-tenant-pages.ts, new at the frontend's package root, is tested but outside src/, so outside the coverage report. For the 3 October additions, see the history below.)

Zero below the line — but since v2026.10.0, two frontend files within five points of it

On 30 September every file in the repository cleared 85% branches. On 3 October two did not, both new or rewritten in v2026.10.0 and both in the frontend — and restated from the 11 October coverage reports, the same two are still the only files below 85 in any workspace, at the same figures. Neither v2026.10.1 nor v2026.10.2 touched either:

File	Branches
components/CaseworkerDashboard/BesluitOverzichtSection.tsx	80.43 (37/46)	New: the Lopende and Afgeronde besluiten list. One more uncovered branch fails the gate
components/process/phaseSet.ts	83.33 (10/12)	New in place of awbStepper.ts: the phase set a process declares, or the Awb set

85 is a margin, not a floor, and v2026.10.0 was the first release since the margin was reached to arrive below it; v2026.10.1 and v2026.10.2 added nothing below it. The two are the obvious next targets; BesluitOverzichtSection.tsx is the urgent one, because the next uncovered edge added to it fails a required check. Of the files v2026.10.1 changed or added, the nearest to the margin are components/signing/SigningPanel.tsx at 85.93%, and in the backend routes/edocs.access.ts at 89.28% and routes/brp.routes.ts at 89.47%. v2026.10.2's one changed source file under coverage, services/operaton.service.ts, reads 92.78% (270/291).

History. On 28 September the margin was thinner than the headline sounded: three files sat at exactly 80.00% branches — backend/src/media-aggregator/sanitize.ts, frontend/src/components/CaseworkerDashboardV2/NoAccessPanel.tsx and pa-cockpit/src/components/PADashboardV2/dossierbeheer/DossierRow.tsx — so one uncovered edge added to any of them would have failed a required check. Two commits closed that:

• 391b1a8 (#256, v2026.09.14) covered the remaining arms of all three, taking each to 100%. Its missing-kompas test exposed a latent crash in DossierRow, fixed in the same commit.
• 73a6764 (v2026.09.15) brought 32 files that sat between 80.00 and 85 — 11 in the backend, 13 in the frontend, 6 in pa-cockpit, 2 in public-site — to 85 or above, most well past 90. services/edocs.service.ts, which that commit records at exactly 80.00% before it, read 98.67% on 30 September — and 92.95% on 9 October, after v2026.10.1 more than doubled its branch surface. It needed one production change: public-site's lib/api.ts now reads import.meta.env directly, so vi.stubEnv can reach it, and went from 83.78% branches to 97.29% (96.55% on 3 October, after v2026.10.0 taught it to read a problem's detail). The arms left there cannot occur in Node and are tracked in #294.

Measured on 11 October, the lowest file in each workspace — the same five as on 3 and 9 October, at the same figures:

Workspace	Lowest file by branches	Branches
backend	src/pa-monitoring/pa-cache.ts and src/services/llm/OpenAILlmProvider.ts	86.36 (19/22 each)
frontend	src/components/CaseworkerDashboard/BesluitOverzichtSection.tsx	80.43 (37/46)
pa-cockpit	src/services/dossierbeheer.api.ts	85.10 (40/47)
pa-demo	src/demo/changelog/DemoChangelogPanel.tsx	87.50 (⅞)
public-site	src/lib/useQueryState.ts and src/pages/herkomst/HerkomstTrace.tsx	87.50 (14/16 each)

Four of the five are as on 30 September. IouFeedbackSection.tsx, which then sat exactly on the 85 line at 51 of 60 branches, has read 86.21% — 50 of 58 — since a change in v2026.10.0 that came with a test.

Per file is the whole point: against a package average, one file falling to 40% barely moves 95%, and the regression the floor exists to catch would pass. Branches only is equally deliberate. Measured the same way on 11 October, a functions floor at 80 would fail 23 files — frontend 8, pa-cockpit 7, pa-demo 5, public-site 3, backend 0 — so the symmetry is a trap for whoever adds functions: 80 on the assumption that it is free.

public-site/src/components/TopBar.tsx is the example the configs themselves reach for, and it still holds: 100% branches, 66.66% functions. The two metrics are not interchangeable, and a file can be exemplary on one while failing the other — though note that TopBar.tsx has no branches at all, so its 100% is the reporter's default for an empty denominator rather than anything a test achieved.

See Coverage Floor.

The configs' own comments say 26 — corrected on 28 September, and three high on 11 October

Through v2026.09.13 all five runner configs claimed a functions floor would fail 31 files, a figure the reports had contradicted on 20, 24, 26 and 28 September, which each measured 26. 391b1a8 (v2026.09.14) corrected the comments to 26 and dated the measurement 28 September; they are unchanged at c51ec0a. Measured since, the figure was 23 on 30 September, 24 on 3 October, 23 on 9 October and is 23 on 11 October:

Workspace	Comment until v2026.09.13	Comment since 391b1a8	Measured, 30 Sep	Measured, 3 Oct	Measured, 9 Oct	Measured, 11 Oct
frontend	11	10	8	9	8	8
pa-cockpit	10	8	7	7	7	7
pa-demo	7	5	5	5	5	5
public-site	3	3	3	3	3	3
backend	0	0	0	0	0	0
Total	31	26	23	24	23	23

The comments are not wrong so much as dated: each says when it was measured, which is the fix that matters, since a count like this moves with every release that touches a container — v2026.10.0 added one, BesluitStartSection.tsx at 2 of 3 functions, and v2026.10.1 removed one when it deleted DvtpTakenSection.tsx. Where the two disagree, re-run the suites rather than trusting either.

The 23 are concentrated where the largest containers live: in the frontend, the four board page containers (CaseworkerDashboardV2, Dashboard, InfraBoardDashboard, WooDashboard), three CaseworkerDashboard sections — BesluitStartSection, IouFeedbackSection and IouGebruiksscenarioSection — and RegelSimulatie at 54%; in pa-cockpit, PADashboardV2, Dossierbeheer, MdEditor, Vandaag, pa.data, CuratieSpecSection and ZoekcriteriaSection; in pa-demo, App.tsx and four shims, two of which — the dock stand-in and the session-expiry warning — render null by design; and in public-site, TopBar, Herkomst and Results. The backend has none.

The frontend row is not comparable to v2026.08.23

The Public Affairs cockpit was extracted into packages/pa-cockpit in this window, taking 41 test files and 368 tests with it. The frontend percentage is therefore measured over a different, smaller body of code than the 87.47% recorded last time — it did not simply improve. Read the frontend and pa-cockpit rows together.

pa-demo's jump is the vendored fork leaving, not a testing campaign. Its figures previously spanned a byte-identical copy of the cockpit with src/vendor/** excluded by hand; the fork was deleted in v2026.08.28 and the package is now a thin host adapter over @ronl/pa-cockpit. There is nothing left to exclude — see pa-demo by area below.

What moved in v2026.10.2 — measured 11 October, against the 9 October reading of v2026.10.1. git diff --name-status ebec288 c51ec0a -- 'packages/*/src/**' is the whole window: services/operaton.service.ts and its test, bpmn-swimlane.test.ts and seven of its fixtures in the backend, and pages/changelog-data.ts in the frontend.

• Backend: no new file, +5 tests, 2633 → 2638. Package figures: statements 98.77 → 98.78, branches 94.97 → 94.95, functions 97.95 and lines 99.10 unchanged. All of it is operaton.service.ts, whose getDecisionDocument now reads the instance's completed user tasks to pick the document of the path taken; the file reads 98.41 / 92.78 / 96.93 / 98.80, and the services row is the only area row that moved. See Backend by area.
• Frontend: no test change, 1540. Package figures 95.42 → 95.35, 93.21 → 93.18, 91.54 → 91.48, 95.93 → 95.85, both runs serial, on source whose only change is the changelog data. Every area row reproduced to the decimal but CaseworkerDashboardV2, in which no file has changed since 14 September — see Frontend by area and the last two decimals warning below.
• pa-cockpit, pa-demo and the public site: no source and no test change, and every package figure and every area row reproduced to the decimal against 9 October — pa-demo for the tenth release running.

What moved in v2026.10.1 — measured 9 October, against the 3 October reading of v2026.10.0. git diff --name-status 0625d48 ebec288 -- 'packages/*/src/**' is the whole window, plus the frontend's vite-plugin-tenant-pages.ts at its package root.

• Backend: +5 files, +176 tests, 2457 → 2633, and four new source files. Package figures: statements 98.91 → 98.77, branches 95.21 → 94.97, functions 97.92 → 97.95, lines 99.18 → 99.10. The new code is the eDOCS connection acting as the person: auth/ grew from two files to four and from 100 on all four measures to 98.21 / 96.07 / 97.72 / 98.56, on entra-token.service.ts; routes/ gained edocs.access.ts; and services/edocs.service.ts went from 98.67% branches on 30 September to 92.95% (132 of 142). citizen-services.ts and edocs-author.ts arrived at 100 on all four. See Backend by area.
• Frontend: +6 files (eight new, two removed), +162 tests, 1378 → 1540. Package figures: 95.23 → 95.42, 93.02 → 93.21, 91.21 → 91.54, 95.76 → 95.93, both runs serial. The organisation landing pages and the citizen services arrived well covered — components/LoginChoice at 97.82% branches over four files, pages/login-choice and the new pages/citizen at 100 — and src/pages rose on all four, functions 78.68 → 81.2, on the citizen Dashboard and LoginChoice. The one row that fell is CaseworkerDashboardV2, functions 84.24 → 82.92.
• pa-cockpit, pa-demo and the public site: no source and no test change. pa-demo reproduced every figure to the decimal, for the ninth release running; pa-cockpit and the public site moved by one hundredth in one package figure each, and pa-cockpit's one-file src/pages row by more — see pa-cockpit by area.

What moved in v2026.10.0 — measured 3 October, against the 30 September reading of v2026.09.15. git diff --name-status ae06c9e 0625d48 -- 'packages/*/src/**' is the whole window.

• Backend: +3 files, +87 tests, 2370 → 2457. Package figures: statements 98.90 → 98.91, branches 95.14 → 95.21, functions 97.98 → 97.92, lines 99.20 → 99.18. The three new source files — middleware/error.middleware.ts, utils/problem.ts and routes/besluitvorming.routes.ts — arrived at 100 on all four measures; operaton.service.ts, validsignCompletion.service.ts, validsign.routes.ts, m2m.routes.ts and bpmn-swimlane.ts each rose a little. The one file that fell is routes/public.routes.ts, from 100% functions to 96.66% and 98.59 → 98.14 statements, after the problem-details change. See Backend by area.
• Frontend: +4 files, +60 tests, 1318 → 1378, and four new source files. Package figures: 95.36 → 95.23, 93.09 → 93.02, 91.41 → 91.21, 95.89 → 95.76 — measured serially this time; see Which frontend run the aggregate uses above. The new code is uneven: utils/problem.ts and useTaskSignature.ts at 100, BesluitStartSection.tsx at 66.66% functions, and BesluitOverzichtSection.tsx at 80.43% branches. components/process/ fell from 91.96 to 91.18 branches on the declared-phases work, with ProcessWhere.tsx 91.66 → 85.71 and the new phaseSet.ts at 83.33; and TakenInbox.tsx, which now hosts the signing panel, went from 96.02 to 94.9 statements while its branches rose, 93.24 → 94.11.
• Public site: +1 file, +10 tests, 265 → 275 — lib/problem.test.ts (9) and one more case in lib/api.test.ts. Package figures: 96.60 → 96.68, 97.34 → 97.40, 95.07 → 95.09, 97.17 → 97.23, all of the movement in src/lib.
• pa-cockpit and pa-demo reproduced all four package figures and every per-area row to the decimal, on source that did not change — pa-demo for the eighth release running.

What moved in v2026.09.14 and v2026.09.15 — measured 30 September, against the 28 September reading of v2026.09.13; v2026.09.14 was not measured on its own. git diff --name-status 963fe24 ae06c9e -- 'packages/*/src/**' is the whole window.

• Backend: +1 file, +168 tests, 2202 → 2370. Package figures: statements 98.49 → 98.90, branches 92.57 → 95.14, functions 97.51 → 97.98, lines 98.89 → 99.20. The new file is the conformance helper's own test; the branch gain is mostly the margin work in routes/, services/, auth/ and mcp-servers/lde — see Backend by area.
• Frontend: +10 files, +196 tests, 1122 → 1318, and twelve new source files, nine of them the caseworker's process view. Package figures: 93.24 → 95.36, 89.91 → 93.09, 87.98 → 91.41, 94.07 → 95.89. The new code arrived well covered — components/process/ at 99.56% statements and 91.96% branches — and 73a6764 lifted CaseworkerDashboard/ and src/pages by several points each.
• pa-cockpit: no new file, +39 tests, 476 → 515, all from the two margin commits. Package figures: 90.07 → 92.61, 88.52 → 93.89, 86.39 → 89.56, 91.33 → 93.13 — the largest branch gain of the five.
• Public site: +1 file, +30 tests, 235 → 265 — the new indexHtml.test.ts for the link-preview tags (23) and the rest in prerender.test.ts, lib/api.test.ts and Footer.test.tsx. Package figures: 95.92 → 96.60, 96.31 → 97.34, functions unchanged at 95.07, 96.41 → 97.17, all of the movement in src/lib.
• pa-demo reproduced all four package figures and every per-area row to the decimal, for the seventh release running.

What moved in v2026.09.13, measured 28 September against the 26 September reading of v2026.09.12, for the record: the backend +4 tests in existing files, the frontend +5 and one new source file, pa-demo and the public site unchanged to the decimal, and pa-cockpit moving in the second decimal on statements and functions with no src/ change and no test change at all — the nondeterminism this page warns about below, caught in the cleanest possible conditions.

Two windows further back, for the record: 24 → 26 September was the backend alone, 89 → 96 files and 2072 → 2198 tests, the seven new files covering six new source modules all at 100 on all four measures. 20 → 24 September took the backend 86 → 89 files and 2028 → 2072 tests on root.routes, build-info and cors-origin; the public site gained four tests in Detail.test.tsx, the frontend three in infra-board.data.test.ts.

The last two decimals are noise

Frontend coverage is not deterministic. Six runs at the same commit, with no source change between them, produced 87.34% twice and 87.47% four times — the same 6789-statement denominator, nine statements apart. Clearing the Vite cache changed nothing. Treat a mismatch in the second decimal as expected rather than as something to chase; a difference of a whole point is worth investigating.

The backend has a different sensitivity: its figures depend on the invocation. Through 30 September they came from npm test --workspace=@ronl/backend; the 3, 9 and 11 October figures come from the same Jest command with --runInBand, so a second-decimal difference against earlier passes may be the invocation rather than the code.

9 October caught it again on unchanged packages. pa-cockpit's one-file src/pages row read 86.87% statements and 79.03% functions on 3 October and 86.95% and 79.36% on 9 October, on byte-identical source — enough to move the package's functions figure by one hundredth, 89.56 → 89.57. The public site's branches did the same, 97.40 → 97.41.

11 October caught it in the frontend, and larger. Between 9 and 11 October no file under components/CaseworkerDashboardV2/ changed — none has since 14 September — yet its row went from 91.98 / 93.22 / 82.92 / 92.35 to 91.22 / 92.97 / 82.31 / 91.46, serial run against serial run. RegelSimulatie.tsx is in it: 54% functions (27 of 50), where 9 October recorded 56% (28 of 50) and 3 October 60%. That one row moved the package by three to eight hundredths on every measure. A move that size, on code nobody touched, is noise to note rather than a regression to chase; a fall of a whole point would not be. pa-cockpit's and the public site's rows, by contrast, reproduced their 9 October readings exactly.

All five configure collectCoverageFrom / coverage.include to span the whole src tree rather than only the files a test happens to import, so an untested file shows as 0% instead of silently disappearing from the report. pa-demo no longer needs a src/vendor/** exclusion, because there is no vendored tree left to exclude — see pa-demo by area.

What the headline numbers mean here. Backend, frontend and public site have been through a dedicated coverage campaign that closed breadth gaps deliberately — every backend feature area, every frontend component and page, and every public-site module now has at least a test file. What remains there is depth. pa-cockpit inherits the frontend's profile, since it is the code that used to be measured there. pa-demo is the outlier in the other direction (see pa-demo by area):

Package	Statements → branches	Gap	Was, v2026.08.36
Backend	98.78 → 94.95	3.8	7.5
Frontend	95.35 → 93.18	2.2	8.0
pa-cockpit	92.61 → 93.89	−1.3	10.6
pa-demo	93.47 → 95.65	−2.2	4.4
Public site	96.68 → 97.41	−0.7	16.4

Every gap narrowed, and three have gone negative — branches now sit above statements, pa-cockpit joining pa-demo and the public site on 30 September, and all three still there on 11 October. The backend's gap widened by a tenth in v2026.10.1, on the new eDOCS branches. That is what a campaign aimed at branch edges produces once it reaches the guards inside files whose plain statements nobody had a reason to execute. The public site, whose 16.4-point gap was the widest in the repository at v2026.08.36, is one of the three. pa-demo's position is a property of what it became rather than of effort spent: a thin host adapter over a package that carries its own tests. What is left elsewhere is the same kind of thing: defensive if (!req.user) guards behind real middleware, ?? null fallbacks, catch blocks unreachable through a legal input, and deliberately-scoped "critical interactions only" passes on the largest components — documented per-file rather than silently absent.

---

Backend by area

Sub-directories report separately rather than rolling up into their parent, and istanbul truncates to two decimals rather than rounding. Match these against npm test --workspace=@ronl/backend -- --coverageReporters=text.

Re-derived on 11 October 2026 from the run's own coverage/coverage-final.json, grouped per directory the way istanbul's text reporter groups them and truncated to two decimals the same way. The rows reconcile to the package total in the table at the top of this page, and that total to the lcov.info the same run wrote, which is the check that they are current.

Area	Files	Statements	Branches	Functions	Lines
mcp-servers/edocs	1	100	100	100	100
openapi	1	100	100	100	100
openapi/testing	2	100	100	100	100
services/document	4	100	100	100	100
rip-swimlane	2	100	99.26	100	100
middleware	4	100	96.66	94.11	100
mcp-servers/triplydb	1	100	90.9	100	100
routes	20	99.16	97.41	99.42	99.25
services/llm	4	99.02	92.3	100	98.95
utils	13	98.96	98.95	100	98.84
pa-monitoring	10	98.95	89.35	95.93	98.99
services	16	98.71	94.98	98.69	99.09
auth	4	98.21	96.07	97.72	98.56
pa-monitoring/sources	6	98.13	93.05	94.52	99.13
mcp-servers/lde	1	98.03	100	100	97.95
media-aggregator	10	97.58	93.37	100	98.61
services/mcp	6	96.35	100	93.9	97.86

One row moved in v2026.10.2, and sixteen are unchanged to the decimal. services went 98.7 → 98.71 statements, 95.05 → 94.98 branches and 98.68 → 98.69 functions, lines unchanged, on operaton.service.ts at 98.41 / 92.78 / 96.93 / 98.80 (558/567, 270/291, 95/98, 495/501): the decision-document path brought 10 statements and 8 branches, and the five new tests reach all the statements and 7 of the branches.

History: five rows moved in v2026.10.1, and twelve were unchanged to the decimal — mcp-servers/edocs among them, still at 100 on all four although its server now refuses a person's call in Dutch when their eDOCS session cannot be used. auth moved most: two new files, citizen-services.ts at 100 and entra-token.service.ts at 95.87 / 91.48 / 92.85 / 96.77, took the row from 100 on all four to 98.21 / 96.07 / 97.72 / 98.56. routes gained edocs.access.ts (89.28% branches) and fell on branches, 98.3 → 97.41, while its functions rose, 99.39 → 99.42. services gained edocs-author.ts at 100, and fell on branches, 95.54 → 95.05, almost all of it edocs.service.ts at 92.95%. utils rose a tenth on config.ts — 96.29% statements and 98.52% branches, up from 95.12 and 98.31, on the Entra settings it now validates — and services/mcp by two hundredths on EdocsMcpProvider.ts.

History: seven rows moved in v2026.10.0, and ten were unchanged to the decimal. Three gained a file at 100 on all four — middleware (error.middleware.ts; branches 95.83 → 96.66, functions 92.85 → 94.11), utils (problem.ts; 98.82 → 98.89 statements) and routes (besluitvorming.routes.ts). routes is also the one row that fell on a measure: functions 100 → 99.39, every bit of it public.routes.ts at 96.66%. rip-swimlane rose to 99.26 branches with the declared-phases parser, services moved in the second decimal (operaton.service.ts and validsignCompletion.service.ts up, branches 95.61 → 95.54 on the larger denominator), and pa-monitoring lost a little on lines, 99.16 → 98.99, in pa.routes.ts and pa-dossiers.routes.ts after their errors became problems. media-aggregator moved by one hundredth on statements.

History: nine rows moved between v2026.09.13 and v2026.09.15, and eight were unchanged to the decimal. The biggest moves are the margin work: auth from 90.74 / 88.63 / 90.9 / 91.91 to 100 on all four, jwt.middleware.ts reaching 100 alongside tenant-access.ts; mcp-servers/lde from 82.14% branches to 100; routes 94.6 → 98.26 branches; services 91.79 → 95.61 branches, edocs.service.ts, lde.service.ts and search.service.ts above all. rip-swimlane reached 100 statements, functions and lines, and 99.12 branches, while its parser grew to read any laned process. openapi/testing gained conformance.ts at 100 on all four. pa-monitoring, pa-monitoring/sources and media-aggregator each moved by about a point on branches. media-aggregator/sanitize.ts, one of the three files at exactly 80.00% on 28 September, is at 100% branches.

The v2026.09.13 window, before that: two rows moved in the second decimal — rip-swimlane on bpmn-swimlane.ts reading a list of document references instead of one, and services/ on operaton.service.ts, where three caches keyed by process-definition key became one keyed by definition id.

The v2026.09.12 window, which this table also absorbed: openapi and openapi/testing — document.ts and routeOperations.ts — arrived at 100 across the board; auth/ gained tenant-access.ts at 100 and rose on every measure; middleware/ gained version.middleware.ts at 100, its branch and function figures moving a fraction (95.91 → 95.83, 92.3 → 92.85) on the larger denominator; routes/ gained openapi.routes.ts and registry.ts, both at 100.

utils/ is no longer the exception

Through v2026.08.20 this table's lowest row by a wide margin was utils/ at 43.47% statements and 6.54% branches, documented as an accepted artifact: config.ts sat at 0% because it self-runs dotenv and validateConfig on import, and logger.ts was mocked in every test that touched it.

Both are closed. Of the thirteen source files in utils/ on 11 October, twelve report 100 / 100 / 100 / 100 — altcha, build-info, client-ip, cors-origin, dutch-datetime, env, errors, logger, operaton-variables, problem (new in v2026.10.0), slug, tls-bootstrap. The thirteenth is config.ts at 96.29% statements and 98.52% branches, which is what pulls the area row off 100 and is comfortably the largest branch surface in the area. tls-bootstrap.ts was also at 0% and is now fully covered. The area that was the standing excuse is now the joint-best large area in the package, and the one that grew most in v2026.09.11.

auth/ was the lowest row on statements, functions and lines through the 28 September pass. Through v2026.09.11 it was jwt.middleware.ts alone, at 88.23% statements and 82.75% branches; v2026.09.12 added tenant-access.ts at 100 on all four, lifting the row to 90.74% statements and 88.63% branches; and 73a6764 took jwt.middleware.ts itself to 100, so the row read 100 on all four from 30 September to 3 October, until v2026.10.1's entra-token.service.ts arrived. The lowest rows on 11 October, as on 9 October, are services/mcp on statements, functions and lines (96.35%, 93.9% and 97.86%) — middleware, lowest on functions through 30 September, rose past it with error.middleware.ts — and pa-monitoring on branches at 89.35%, more than nine points clear of the floor. At file level the closest are pa-monitoring/pa-cache.ts and services/llm/OpenAILlmProvider.ts, both at 86.36% branches.

---

Frontend by area

Re-derived on 11 October 2026, from the serial run, for the reason given in Which frontend run the aggregate uses at the top of this page. The rows reconcile to the 95.35 / 93.18 / 91.48 / 95.85 package total above. Vitest's text table omits fully covered files, so the rows are grouped from the run's coverage-final.json.

In v2026.10.2 one row moved, on unchanged code, and twenty-one reproduced to the decimal. CaseworkerDashboardV2 went 91.98 → 91.22 statements, 93.22 → 92.97 branches, 82.92 → 82.31 functions and 92.35 → 91.46 lines, although no file in it has changed since 14 September; RegelSimulatie.tsx reads 54% functions (27 of 50), against 56% on 9 October. That is run-to-run noise — see the last two decimals warning. src/pages, where the changelog data did change, reproduced to the decimal.

History: in v2026.10.1 six rows moved, one was new and fifteen reproduced to the decimal. src/pages/citizen is new, at 100 — citizenServiceUi.ts and heusdenpasTestCases.ts, both data. src/components/LoginChoice went from two files at 100 to four (98.38 / 97.82 / 96 / 98.21), with AccessDeniedDialog.tsx and SingleBoardLanding.tsx; src/pages/login-choice from one file to three, still at 100, with landing-login.ts and single-board.copy.ts; and src/services from eight to ten (96.9 → 97.25 statements, 91.07 → 91.69 branches) with board-request.ts and landing.ts, tenant.ts reading 97.72% branches. CaseworkerDashboard lost the two DVTP sections (28 → 26 files) and rose a little on every measure; src/pages rose on all four, functions 78.68 → 81.2; and CaseworkerDashboardV2 fell on statements, functions and lines — functions 84.24 → 82.92 — not only in the files v2026.10.1 changed: RegelSimulatie.tsx, untouched, reads 56% functions (28 of 50) where 3 October recorded 60%.

History: in v2026.10.0 five rows moved, one was new and fifteen reproduced to the decimal — src/utils among them, at 100 with one more file. src/components/signing is new — SigningPanel.tsx and resolveSigningUrl.ts moved in from InfraBoardDashboard, which drops from 12 files to 10, and useTaskSignature.ts is new. CaseworkerDashboard gains BesluitOverzichtSection.tsx and BesluitStartSection.tsx (26 → 28) and src/utils gains problem.ts at 100 (2 → 3). The moves are small and mostly downward on new or reworked code: CaseworkerDashboard 94.93 → 94.45 branches, components/process 91.96 → 91.18 branches, CaseworkerDashboardV2 84.66 → 84.24 functions, and src/services 97.81 → 96.9 statements and 97.32 → 95.65 functions on the new response interceptor in api.ts. InfraBoardDashboard rose on branches, 89.37 → 89.91, once the signing panel left it.

Area	Files	Statements	Branches	Functions	Lines
src/ (root files)	1	100	100	100	100
src/components/…/regelsimulatie/__helpers__	1	100	100	100	100
src/components/PADashboardV2	2	100	100	100	100
src/hooks	1	100	100	100	100
src/pages/caseworker-v2	1	100	100	100	100
src/pages/citizen	2	100	100	100	100
src/pages/login-choice	3	100	100	100	100
src/types	1	100	100	100	100
src/utils	3	100	100	100	100
src/pages/infra-board	6	100	98.47	100	100
src/components/process	10	99.16	91.18	100	99.74
src/pages/woo	2	99.13	100	100	99
src/components/LoginChoice	4	98.38	97.82	96	98.21
src/components/…/regelsimulatie	8	98.33	88.59	96.38	98.89
src/components/WooDashboard	12	98.26	94.78	98.52	98.57
src/components	7	97.65	93.47	98.3	100
src/services	10	97.25	91.69	95.93	97.33
src/components/InfraBoardDashboard	10	95.34	89.91	92.59	95.85
src/components/CaseworkerDashboard	26	94.2	94.51	90.43	95.12
src/components/signing	3	93.1	88.75	95.65	96.11
src/components/CaseworkerDashboardV2	10	91.22	92.97	82.31	91.46
src/pages	11	89.03	95.73	81.2	88.97

This table no longer carries the cockpit's rows

Earlier versions listed src/pages/public-affairs-v2, src/components/PADashboardV2/dossierbeheer and a large src/components/PADashboardV2 here. Those directories live in packages/pa-cockpit since the extraction and are measured by that package's own suite — see pa-cockpit by area below. The two-file src/components/PADashboardV2 that remains in the frontend is what stayed behind, and it is at 100.

The other reason the shape changed: several areas that were middling are now at 100, and the old src/utils row of 66.66% statements / 50% branches is gone entirely. Nothing in the frontend is below the 80% branch floor, though since v2026.10.0 two files are below 85 — see the floor note at the top of this page.

src/pages is the lowest of the top-level areas because it is where the largest containers live — and at 81.2% functions it holds four of the eight frontend files that a functions floor would fail. Its branches, at 95.73%, are well clear of the floor; this is a functions gap, not a branch gap. Per-board detail is on the board pages: Caseworker, PA cockpit, Infra-board, Woo-dashboard.

---

pa-cockpit by area

Re-derived on 11 October 2026, on a package none of v2026.10.0, v2026.10.1 and v2026.10.2 touched. These rows reconcile to the 92.61 / 93.89 / 89.57 / 93.13 package total above, and all seven are identical to 9 October. Six of them are also identical to 30 September and 3 October; the seventh, the one-file src/pages row, moved on identical code between 3 and 9 October — 86.87 → 86.95 statements, 79.03 → 79.36 functions, 86.76 → 86.86 lines — the run-to-run noise the last two decimals warning describes, enough to move the package's functions figure by one hundredth, and has held there since. The account below is of the 28 → 30 September window. The package's only changes in it were tests — 391b1a8's four in DossierRow.test.tsx and 73a6764's 35 across six files — plus the small DossierRow.tsx fix the first of them exposed, so nearly everything that moved below moved on new assertions against unchanged code.

Area	Files	Statements	Branches	Functions	Lines
src/ (root files)	2	100	100	100	100
src/services	3	95.97	93.99	98.19	95.28
src/modes	1	95.65	100	100	95
src/pages/public-affairs-v2	13	94.73	95.04	92.55	95.53
src/components/PADashboardV2	10	92.12	91.78	86.98	93.73
src/pages	1	86.95	95.57	79.36	86.86
src/components/PADashboardV2/dossierbeheer	8	85.47	92.88	81.35	86.29

In the 28 → 30 September window four rows moved and three were identical to 28 September. src/pages/public-affairs-v2 moved most — 88.2 → 95.04 branches, 86.4 → 92.55 functions — on the 16 new Monitoring.test.tsx cases; src/services 84.98 → 93.99 branches on 13 new cases in pa.api.test.ts; dossierbeheer 89.12 → 92.88 branches, DossierRow.tsx now at 100; and src/components/PADashboardV2 89.44 → 91.78. The 28 September reading had the second-decimal wobble this page warns about — statements and functions alternating on identical code — which is why a move that small would not be worth a sentence, and why these are.

dossierbeheer is still the lowest area on statements and functions, but at 92.88% branches it clears the floor by nearly thirteen points — the same pattern as the frontend's src/pages. This package holds 7 of the 23 files that an 80% functions floor would fail.

---

Public site by area

Re-derived on 11 October 2026, on a package neither v2026.10.1 nor v2026.10.2 touched. These rows reconcile to the 96.68 / 97.41 / 95.09 / 97.23 package total above, and all six are identical to 9 October. Five of them are also identical to 3 October; the sixth, src/pages, read 97.29% branches then and 97.31% on 9 and 11 October, on identical code — the same run-to-run noise as pa-cockpit's. On 3 October src/lib was the row that moved, on v2026.10.0's problem details: lib/problem.ts is new, at 100 on all four, and lib/api.ts now reads a problem's detail through it.

Area	Files	Statements	Branches	Functions	Lines
src/ (root files)	1	100	100	100	100
src/i18n	3	100	100	100	100
src/pages/herkomst	8	100	97.05	100	100
src/components	13	97.14	100	96.29	97.05
src/lib	9	96.8	97.05	97.43	98.03
src/pages	10	95.45	97.31	92	95.83

src/lib went 96.49 → 96.8 statements, 96.66 → 97.05 branches, 97.36 → 97.43 functions and 97.87 → 98.03 lines. Before that, the v2026.09.15 window took it from 93.85 / 91.11 to 96.49 / 96.66, on the one production change 73a6764 made anywhere: lib/api.ts reads import.meta.env directly, so its base-URL fallbacks can be tested with vi.stubEnv. lib/api.ts itself went from 83.78% branches — the lowest file in the package from v2026.09.11 to v2026.09.13 — to 97.29% then, and has read 96.55% since 3 October; the arms left cannot occur in Node and are tracked in #294. The lowest files by branches are lib/useQueryState.ts and pages/herkomst/HerkomstTrace.tsx, both at 87.50%.

The new src/indexHtml.test.ts adds no row: it tests index.html's link-preview tags, filled from each mode's .env file, and index.html is not a source file under coverage.

The pre-campaign rows are gone, and the difference is the point

This table used to carry three rows derived on 22 August 2026 — src/components 96.77, src/lib 84.9 / 73.17 branches, src/pages 82 / 61.51 branches — which sat visibly below a package total of 96.41% and carried a warning saying so. They are now re-measured: src/lib branches have gone 73.17 → 97.05, and src/pages branches 61.51 → 97.31. Nothing in the package is below the 80% branch floor, and src/components is at 100% branches outright.

The whole-tree src/pages row covers the ten files directly in that directory; the eight-file herkomst/ provenance explorer reports as its own row, at 100% statements.

See Public site suite for what these files are and what is deliberately not covered.

---

pa-demo by area

Re-derived on 11 October 2026, and every row below reconciles to the All files row rather than predating it. All five rows reproduced to the decimal against 24, 26, 28 and 30 September and 3 and 9 October, on a package whose src/ tree has not changed since v2026.09.5.

Area	Files	Statements	Branches	Functions	Lines
All files	16	93.47	95.65	85.00	92.85
src/demo	8	100	100	100	100
src/demo/changelog	2	100	87.50	100	100
src/ (root)	2	75.00	100	50.00	75.00
src/demo/shims	4	75.00	100	44.44	75.00

src/demo reached 100 across the board

The rows that used to sit here were derived on 30 August and did not sum to the total — v2026.09.2's campaign had added tests across the package after they were taken. Re-measured, src/demo is at 100 / 100 / 100 / 100 where it previously read 95.91 / 84.61, and the package's remaining gaps are entirely in the two rows at 75%, both of which are at 100% branches. This package contributes 5 of the 23 files an 80% functions floor would fail, and none at all to a branch floor.

The src/vendor/** exclusion is gone, because the vendored tree is. Until v2026.08.28 this package held a byte-identical copy of the cockpit — 39 files kept honest by a manifest, a sync script and a drift checker — which had to be excluded from these figures to avoid double-counting code the frontend suite already exercised. The extraction into @ronl/pa-cockpit deleted the fork and all of that machinery, so every figure above is now pa-demo's own demo-owned surface and nothing else. That is the whole reason the package total jumped from 73.94% to 91.30% without a single test being written for that purpose.

What remains uncovered is almost entirely shims that deliberately return nothing: the dock stand-in and the session-expiry warning both render null by design, because the real components pull in chat machinery and session handling that a public, unauthenticated page must not have. They depress the function percentage without representing a gap.

Two more files are excluded from coverage entirely, by config rather than by this table's rounding: src/main.tsx (it calls createRoot; its one meaningful line, forceMockMode(), is tested separately via the extracted, fully-covered src/main-helpers.ts) and src/vite-env.d.ts (a type-only ambient declaration file, nothing to execute). The src/ row above is just these two remaining root files, App.tsx and main-helpers.ts — 100% on main-helpers.ts (3/3 statements) pulled down by 0% on App.tsx (0/1) gives the row's 75%. App.tsx is the lowest-covered file in the package — its one statement, the route shell itself, has no test rendering it.