test(e2e): three behavioural tests for the Features & Roadmap surface (gate-19 850 → 847)

[rubenvdlinde]

What

Three behavioural e2e tests for the Features & Roadmap surface, plus the one-line allow-list entry that makes CI run them.

gate-19, full tree, canonical checker ConductionNL/.github@3c8da4c:

	covered	uncovered
development (fdb0d6ca8)	10	850
this branch	13	847

Zero new @e2e excludes.

Why these three, and why a new file

spec-coverage/feature-pages.spec.ts asserts that each page renders. These assert that it decides:

• an empty features manifest renders No features documented yet;
• the suggest-feature form stays unsubmittable until every required field is set;
• features_roadmap_enabled defaults to true, and the guard that enforces it can still refuse.

Every test carries its own control

Each of those requirements is otherwise satisfied for free by a page that failed to mount — a disabled button is disabled when it does not exist, and an empty state is empty when nothing rendered. So each test pairs its requirement with a state the same locator must distinguish:

test	requirement	control (same locator)
empty manifest	.cn-features-tab shows No features documented yet	toggle to roadmap → same <main> swaps to .cn-roadmap-tab + No roadmap items yet, then back
submit gate	Continue on <forge> disabled while invalid	four-state gradient ending in enabled; Cancel enabled throughout
default flag	key absent → guard answers not-403	key set false → 403 feature_disabled, then removed

Proven able to fail

Each requirement assertion was inverted, run, observed red naming the real observed value, and restored byte-identical (md5 963de99d… before and after):

Expected: "MUTANT-1 no features documented yet"   Received: "No features documented yet"
Expected: enabled                                 Received: disabled
Expected: 200                                     Received: 403

Negative controls (the real evidence — see scope note)

shipped                                    covered=13  uncovered=847
remove this PR's spec file                 covered=10  uncovered=850   (-3, exactly its 3 anchors)
restore                                    covered=13  uncovered=847

remove openspec/specs/features-roadmap-menu/spec.md (45 counted scenarios, 0 excluded)
                                           covered=9   uncovered=806   (-4 and -41 = -45, exact)
restore                                    covered=13  uncovered=847

⚠️ Scope. This PR changes no spec file, so gate-19's diff scope on this PR is empty and its cell will go green having inspected nothing. That green is not evidence for the numbers above. Both figures are --full on the canonical checker; the negative controls are the evidence.

Admission criteria

Checked against all four criteria in tests/e2e/ci/playwright.config.ts: hermetic (the outbound GitHub call is stubbed with page.route(), so no external network dependency is added), self-cleaning (the only mutation is the app-config key, removed in a finally), no conditional-assert guards, no test.skip().

Full CI subset on a dedicated NC 34 rig: 44 passed, 0 failed (41 before + 3).

Deliberately not anchored

#pat-not-configured-roadmap is exercised in substance but not anchored: the shipped CnRoadmapTab renders Roadmap not yet configured, while the spec's THEN clause names Roadmap currently unavailable. Crediting it would report a stated outcome as verified when it was not. Filed separately, along with the finding that the features_roadmap_enabled UI gate is unimplemented.

🤖 Generated with Claude Code

[github-actions]

Quality Report — ConductionNL/openregister @ 5285fb7

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	❌
psalm	✅
phpstan	❌
phpmetrics	✅
eslint		✅
stylelint		✅
build		✅
check-specs		✅
test-l10n		✅
composer			✅	✅ 173/173
npm			✅	✅ 546/546
PHPUnit					❌
Newman					✅
Playwright					✅
Hydra gates					✅

---

Quality workflow — 2026-08-11 17:31 UTC

Download the full PDF report from the workflow artifacts.