test(shell): record narrow v2 acceptance and keyboard coverage (V2-ACC1b) - #1540

Dieser PR ergänzt Tastatur- und Not-Aus-Tests für zwölf erreichbare Routen der Hülle und des Leitstands.
Die lokalen Prüfungen bestehen, aber Kriterium 4 ist auf der vorgegebenen Basis nicht erfüllt, weil dort der Zustandszähler samt Obergrenze fehlt.
Die Bildabnahme, die Release-Prüfung und die persönliche Klick-Abnahme bleiben beim Nutzer.

Report

Package: V2-ACC1b · Operator: UMSETZEN · Review tier: B, as commissioned.
Candidate: 666c0a9f16c571c59d6c40ecc6234d903607d1fe.
Base: claude/v2-scope-1-sidebar / #1499 at 91cba1f3e3bbef612ad7a0098c4c330fd3c6bbca, itself stacked on #1484.
Observed: 2026-10-15, Linux. Acceptance applies to this stack, not current main.

Goal, scope and changes

Bounded worker goal: test shell + Leitstand with D1 on, document criteria 1/2-list/3/4/5/6-shell/8/9, and hand over this draft. Parent: v2.0 narrow acceptance, section 10.1 of docs/plan/v2.0/plan.md; execution owner: assigned V2-ACC1b worker; acceptance/integration owner: Root.
Native get_goal returned no goal initially; create_goal returned this bounded goal as active, and the milestone get_goal confirmed it active. The parent acceptance remains open because criterion 4 fails.

• 127 added lines in three files, all under src/shell/acceptance/; no product change.
• Twelve real Chromium Tab/Tab/Shift+Tab tests load the app with the existing Tauri browser mock, D1 on and weitere_bereiche off. They check entry into the shell and return between the two visible navigation links, at 1440×900. Vite uses an allocated local port; browser and server are awaited and closed by the test run.
• Twelve active StopBand tests render the real Shell with only the Tauri boundary mocked; they check the status band, its resume button and the persistent emergency-stop control. One additional test pins default flags and registry coverage.
• routes.ts is the unavoidable 15-line test-only route fixture, used to check all registry paths and document the extra mounted paths. No helper outside the assigned directory, dependencies, screenshots, Rust, seams, flag or route-registry changes.

Kriterium → Befehl → Exit-Code → Ergebnis

Criterion 4 uses the approved wording: „Keine zweite Definition der sechs Zustände außerhalb von src/design/, außer den gezählten Alt-Aliasen; die Obergrenze sinkt nur.“ The required #1494 change is absent from the prescribed predecessor snapshot. The existing ratchet prints ts-suppress: 0 = 0, eslint-disable: 0 = 0, rust-allow: 221 = 221, ts-any: 2 = 2. These are not state counters. Root must arrange an updated predecessor containing the state ratchet, then commission a new acceptance run; this package does not merge main or modify the predecessor.

Data-source evidence and test boundaries

• Existing real source proof: src-tauri/src/web_interface.rs::the_real_backend_puts_live_workers_on_the_board (:1851), observed PASS in the full Rust suite. It opens a real Store, inserts workers, starts the real board backend, and checks /board.json for project/worker/state data.
• App wiring: src-tauri/src/main.rs:1689 (get_board_state) reads store.list_workers and passes those workers to the status engine. src/screens/leitstand/LeitstandRoute.tsx uses useBoard and useQuestions.
• Existing frontend boundary proof: src/screens/leitstand/LeitstandRoute.test.tsx::shows the cards and the Braucht dich list from the real board and question data at /leitstand. Despite its name, this frontend test mocks Tauri invoke. It proves rendering/wiring, not a real Store-to-DOM end-to-end run; the separate native test supplies the real-store evidence.

Image inventory for V2-SHOT-BASE

Capture each path in light and dark, 1440×900: 24 images total, outside Git. Titles below are navigation/view labels, not claims that a screenshot was inspected. Include the registry paths exposed by the palette and the extra paths reachable from mounted screens.

Verlauf is exposed with the scope flag off: src/shell/routes.ts:26 registers it under Leitstand, :51 keeps that entire area in scope, and src/shell/header/CommandPalette.tsx:15 includes every tab of each visible area. It is not a tab in the new Leitstand's own tab row (src/screens/leitstand/LeitstandRoute.tsx:23). Its direct path renders the legacy panel according to src/shell/mount/ShellMount.tsx:61. The distinction between registry labels and the mounted legacy content needs visual inspection; this PR claims neither a new Verlauf screen nor a visual match.

Regression probe and additional evidence

The uncommitted probe removed the first sidebar link from keyboard navigation with first.evaluate(element => element.setAttribute("tabindex", "-1")) immediately before the first Tab. Product files were never modified. The probe was restored before committing.

The early trailer preflight rejected generated test titles because it requires literal names in the source. Tests were given explicit names before committing, and the final preflight and full red-first run passed. No pushed history was rewritten.

NICHT ABGEDECKT

Verbatim block from the successful prepush run:

--- NICHT ABGEDECKT von diesem Lauf ---
  Plattform Linux:
    - die #[cfg(windows)]-Tests (run_in_pty, npm_shim, ConPTY, DPAPI)
    - die Windows-Arme von clippy
  Dieser Lauf belegt die Linux-Haelfte, nicht die Windows-Haelfte.
  Windows-Haelfte: derselbe Befehl in Git-Bash auf dem Zielrechner.
  Bahn 'prepush' ist die schnelle Schleife: Browser-Smoke, Frontend-Build
  und die Workflow-Gates laufen erst in der Bahn 'linux'.
  Der Zustand externer Dienste (Updater-Endpoint, OmniRoute, Mirror)
  wird von keinem Gate geprueft — dafuer ist das Release-Verify da.
---------------------------------------

Windows runs in the merge queue. The separate build and twelve browser keyboard cases above were run, but they do not prove every browser-smoke scenario or every focus trap inside dialogs/content. No screenshot comparison, live desktop launch, real worker dispatch, server stop, release/update drill, key backup, or full Store-to-DOM test was performed. Criteria 2 (images), 11 and 13 remain with the user. Provider quota/model observation and the lage-jetzt mirror were unavailable; the read-only start check reported the missing quota input. RAM/build limits were observed and the worker waited for a free build slot.

Reviews, disposition and handover

Review: ausstehend (Stufe B). Author family: OpenAI; Root commissions one reviewer outside OpenAI. This worker did not commission a review, mark ready, set review-ok, or enqueue.

Delivery state: implemented, locally tested and pushed; draft for review, not accepted, merged or released. The acceptance table deliberately remains FAIL for criterion 4. Merge order remains #1484 → #1499 → this PR; Root retargets after predecessor merges. Fabrik remains a separate package. No architecture allowlist changes or new licenses/dependencies.

Review disposition (Root, round 1)

Tier B (tests only, author family OpenAI / Codex gpt-6-astra). Reviewer google/gemini-3.8-flash (Google) at head 666c0a9f, diff against the stacked base claude/v2-scope-1-sidebar: REQUEST_CHANGES with 4 findings. Root adjudication at file:line:

No accepted finding and no open high. Review gate tier B passed at 666c0a9f.

Criterion 4 on the integrated result (Root, BELEGEN)

The FAIL above is a property of the stacked snapshot, not of the product: #1494 (state ratchet) is on main, not on the stack base. A local scratch merge of origin/main (9938420 at the time) into 666c0a9f (merge commit 7f3df745, clean, never pushed), then node scripts/ci/ratchet.mjs check → exit 0, state-token: 191 = 191 (ceiling scripts/ci/ratchets.json:6). The merge queue re-runs the gates on top of main for each PR of the stack, so criterion 4 is checked again at merge time. Root accepts criterion 4 on that basis; no new acceptance run needed.

Merge order: #1484 → #1499 → this PR (Root retargets each to main after its base merges).