fix(scripts): resolve the test's script path natively, and record a drifted example folder (iou-architectuur#105)

[sgort]

Items 5 and 6 of sgort/iou-architectuur#105.

5 — test:scripts failed on Windows, and took the second suite with it

scripts/promotion-targets.test.mjs:8 resolved the script under test with new URL(...).pathname. On Windows that is /C:/Users/.../promotion-targets.mjs — the leading slash makes it not a native path, so execFileSync(process.execPath, [SCRIPT, …]) could not resolve it and every command-line check failed.

Demonstrated rather than assumed, from this Linux host:

.pathname                          /C:/Users/.../promotion-targets.mjs
fileURLToPath(url, {windows:true})  C:\Users\...\promotion-targets.mjs

The script itself was never at fault — run by hand it answered correctly and exited 0, and Linux CI passed throughout, so only Windows contributors saw it. Knock-on: test:scripts joins the two suites with &&, so dso-dossier.test.mjs never ran there either.

Now fileURLToPath(new URL(...)), which gives a native path on every platform.

A guard goes with it. The test exits 1 if SCRIPT is not an existing path — portable, and exactly how the bug hid: false on Windows, true everywhere else. The same mistake cannot return silently.

The other scripts were checked, as the issue asked. dso-dossier.test.mjs uses .href for a dynamic import, which is correct; this file's other two uses (line 27 .href, line 155 a URL passed to readFileSync) are both fine. One line needed changing.

Both suites pass: 24 checks each, and test:scripts now runs both.

6 — four of the five were already fixed; the fifth is a folder the repo has written off

The issue found camunda:formKey="embedded:deployment:awb-missing-info-form.html" in five files, naming a form that exists nowhere. Four are already fixed on acc: the seeded and fixture copies of AwbShellProcess and AwbZorgtoeslagProcess now carry camunda:formRef to kapvergunning-aanvullende-gegevens and zorgtoeslag-aanvullende-gegevens, both of which exist — so Gateway_StillIncomplete's ${supplementReceived} branch has something that can set it.

The fifth is examples/organizations/flevoland/thuisbatterij/, which this repository has already written off in two places:

• e2e-fixtures/manifest.json — the fixture came from the deployed bundle "not from examples/organizations/flevoland/thuisbatterij/ — that folder's -main/-subprocess pair and its two forms have drifted behind what is actually deployed"
• the 2026-09-24 swimlanes plan — "Source of truth: packages/frontend/public/examples/flevoland/. examples/organizations/flevoland/thuisbatterij/ is not touched."

Reading it against the tree shows the drift is wider than the item. -main.bpmn references three forms that exist nowhere, not one:

reference	exists
thuisbatterij-subsidie-start	✗
awb-missing-info-form.html	✗ — the one item 6 names
thuisbatterij-notify-applicant	✗

The maintained copy's three all resolve.

So the BPMN is left as it is and a README records what the folder is. Repairing one of three dangling references would fix a third of the problem while making a written-off file look maintained; repairing all three would mean re-deriving a definition the repository has already replaced. Neither existing statement about this folder is visible from inside it, which is the gap the README closes — and it says to delete itself if the folder is ever made current.

If you would rather have the BPMN repaired outright, say so and I will — it is a different call, not a harder one.

Checks

test:scripts 24 + 24 · check-rip-bpmn-copies green (12 fingerprinted, 2 fixture copies identical) · prettier --check clean on both changed files.