[Cuarroc]

Die AppData-Qualitätsanzeige ist lokal wieder in ihrer ursprünglichen Darstellung ohne Rahmen.
Alle 29 betroffenen Tests und der strenge Vergleich der 396 Designfälle bestehen.
Der Entwurf bleibt gesperrt, weil Red-first den ausgegliederten Gruppierungstest aus der erhaltenen Commit-Historie weiterhin verlangt.

Report

Plan-ID: KR-10e-2-APPDATA-QUALITY

State: BLOCKED, same owner and Draft #1975; not delivered remotely. Review: ausstehend (Stufe B). Root owns the final decision and the ONE non-OpenAI LAST delta review of da390b08685df56bc7ef5deb880199318168e770..cd2cbfbd4464f03ab4385cef8cd0ffae29dd7e88 after checked delivery. No Ready, review label, queue or integration action.

• Original source base: 86401bd81ec62e5e684903f4d29db18730dfbd86.
• Preserved local normal merge: 0a65f1c230ac0a15c42221b6c519cfdcc48584a8, still an ancestor; no reset, rebase or history rewrite.
• Final local candidate: cd2cbfbd4464f03ab4385cef8cd0ffae29dd7e88; clean worktree. Its parents are 82581d770c4095cf2e8d4ec21a16f7c326c5eff2 and main bc3a9ee6a26219a3a91c37ac21028d7a847b73e4.
• Actual remote / Draft head: 64ee625ba6bd8e90ca6fd5a2d05452ed92e04a52, unchanged. The remote conflict remains; local merges have not been published.
• Owner / coordination ticket: srv-kr10e2-quality / W1-03f-api: scoped-credential delivery route POST /agent/delivery #337 / 54c24620-dd9b-4078-ab1f-e11fe3dd9d57.
• Size: 143 RAW, seven package files including tests and P1/P2, against the candidate's actual merge-base bc3a9ee6; below 300 and the authorized review-fix cap 400. No overrun. Forcing findings: R1975-a1 / D1, R1975-a2, R1975-P1/P2.

Exact bounded split and preserved quality

Root split msg-0mvrfiyl2-9a1f86d7 explicitly removes ONLY the failed grouping correction. BetriebPanel.tsx is now byte-identical to the original reviewed da390b08: remove AppData from boxed facts, restore its original separate UNBOXED KeyValue. This actually removes the eight increases; discarding the earlier private prototype alone would not have done so.

The original quality feature remains intact: partial => mindestens, complete => exact size, absent/null/malformed quality => explicit unknown completeness; absent/null/invalid/negative/nonfinite size => unknown, true zero => zero. CPU-average, free-volume and RAM meanings, exact wording, one resource read across rerender and the command whitelist remain. The original IPC implementation/test, resource texts and existing BetriebPanel test are byte-preserved. All three original public quality tests are byte-preserved.

The original four compiling assertion-RED failures (24 existing tests passed) and all historical failed design evidence remain preserved. The obsolete public grouping-only regression is retained separately in private evidence, with its unchanged original text. Under the explicit split it is replaced by keeps AppData in its original unboxed description list, which checks separate lists, the unboxed presentation, literal lower-bound zero and one read. This is an intentional scope removal, not a weakened quality/null/zero/read assertion. No skip/retry/timeout was introduced.

New compiling behavior RED: the new test failed at the shared-list assertion, exit 1, before the source correction. Test commit 0e4cb1f9f8c0449a6c12aecc7e1f3b72b988f1b3 precedes fix 82581d770c4095cf2e8d4ec21a16f7c326c5eff2. All 29 targeted tests pass at the exact final candidate.

P1/P2 were touched only after the actual Composer389, Empty390, Font388 and Split1979 native/process-group termination checks. A real conflict existed only in PLAN. The normal main merge resolves only those two rows manually: P1 describes the local bounded split and pending publication/review; P2 retains the exact prior bounded acceptance. Reversing P1/P2 reproduces the main parent's PLAN exactly. Live PR #1927 and its Root acceptance were rechecked: merged bc95e556bc3f5ef5fa6fb2d9412183dc152a4ef3, 2026-10-26T16:56:38Z, head 3a6d73475a375e59a6030bf054b2927f697e46ed.

All 1,866 foreign modes/blobs equal the main parent. Main's font, rules and comment-editor changes entered solely through the automatic normal merge; this package contains no DiffView production/test repair. No README, design-system, stylesheet, backend, dependency, lockfile, configuration, baseline, threshold, profile or product-harness change. Architecture drift: no new allowlist entry.

Evidence and exact exits

Commands use public-safe output-directory placeholders. Complete commands, raw output, timestamps, environment, source/probe manifests and exact exits are retained privately.

Command / check	Exit / observation
npm run dev:start-check -- --min-free-gb 3.5 --max-cargo 2	0; actual model and supplied native quota/admission evidence checked separately
New named public unboxed-layout regression, before fix	1, compiling assertion failure
npm run test:unit -- src/screens/leitstand/betrieb/BetriebPanel.resources.public.test.tsx src/screens/leitstand/betrieb/BetriebPanel.test.tsx src/lib/ipc.resource-quality.public.test.ts	0, 29 tests at exact final candidate
npm run typecheck, normal final merge-commit hook	0
node scripts/plan-check.mjs --offline, normal hook	0; GitHub/age/running-job checks excluded by this command
bash scripts/ci/secret-scan.sh before each of three commits; all normal commit hooks	0 each
git diff --check; exact split/protected-file/foreign-blob/PLAN-inverse/historical-hash checks	0
npm run design:gate -- --out <base-capture>; same command for <head-capture>	1 / 1; 396 cases and 20 existing stored failures each
Strict paired comparison: every case, all five aggregates and stored-failure counts	0, nonincrease holds without exceptions
npm run test:unit -- src/components/DiffView.lines.public.test.tsx -t 'names diff lines while preserving literal hunks navigation drafts and empty files' on clean exact main 09eb5e14	0, exactly ONE run, one passed test
bash scripts/ci/red-first.sh at exact final candidate	1; historical grouped-layout specs no longer match a test at HEAD; every retained quality spec and new split spec passes at HEAD
Final bash scripts/ci/gates.sh lane prepush / normal push hook	NOT RUN this round, because required pre-push red-first is red; no green claim or bypass
git push -u origin codex/kr10e2-quality	NOT ATTEMPTED this round; no publication after failed required red-first
git ls-remote origin refs/heads/codex/kr10e2-quality	0, still exact remote 64ee625ba6bd8e90ca6fd5a2d05452ed92e04a52

The prior full-prepush/push failure at 0a65f1c2 remains historical: exit 1 in DiffView keyboard tab-stop assertion; 1,839 passed, one failed. It is not relabeled green by the isolated main probe or the later main merge.

Controlled clean-main probe

A fresh Git archive at exact 09eb5e144f3a3f257d7cd0aceac876937c67d713, tree f918a67c37a636803bd49f0803aea8aa61399040, had all 1,866 original blobs verified before and after; no source overlay or shim. Existing assigned-worktree dependencies were reused, with the same runner/configuration. Node v24.21.0, npm 11.19.0, Vitest 4.1.11, jsdom 30.1.1, React 18.3.1; dependency/lock hashes and raw environment retained. Guarded foreground command ran once, 2026-10-28T01:50:43Z; exit 0. Actual native header: OpenAI gpt-6.1-sol/high, observed before edits.

No KNOWN_ISSUES entry was added because the conditional red-main observation did not occur. This isolated pass does not prove a general flake, identify the historical full-lane cause, or justify a candidate-only DiffView repair. None was made.

Final exact-source design proof

Base 86401bd8, tree 715750615546e6e954ea8c3c489b4da5547588eb, versus final cd2cbfbd, tree c99979b0b0c7b43010279dd1872cf8cc33208ba2. Every original source blob was verified. The older archive's sole explicit exception is the accepted fixed capture from 07d8706b: scripts/dev/design-gate.mjs, Git blob 7dabe132b1c3697862345abcc2e8ada9926b26d1, SHA-256 9d408cad19ce6248e881fc1a7df6b4840593c09e978f4cb8afe7130b5bcc1d18. The final candidate already contains that exact probe. No update flag or favorable capture retry.

Chromium 153.0.8010.12, 396 identical cases: 0/2/12 items, 1440×900/1920×1080, light/dark. Both absolute gate exits remain 1, with 20 stored failures each; strict nonincrease exits 0.

Metric	Original base	Final candidate
slack	222	222
nesting	2532	2316
tabs	132	132
eyebrows	376	376
grids	20	20
stored failures	20	20

Every metric of every Betrieb case now equals the original base, including all eight former increases: two items 8→8, twelve items 28→28, in both sizes and themes. Only the twelve inherited classic-settings nesting reductions remain; they do not offset an increase elsewhere. There are no increases in any case.

All four actual final-source two-item Betrieb screenshots were inspected. The AppData label and full 0,4 GiB (Vollständigkeit unbekannt) wording are visibly readable at both widths/themes in the separate unboxed list. The pre-existing free-volume value is still ellipsized at 1440 pixels; no broader readability claim.

Blocking conflict and Root disposition needed

red-first.sh collects every Test-First line from non-merge branch commits. Historical commits afc69546 and d5e9d919 retain the spec groups AppData with CPU and free disk in the same description list. The Root-authorized split retires exactly that behavior/test, so the unchanged gate reports four skipped tests and no passed selected test for each historical grouped spec, then exits 1. The former trailers were valid when authored; this is a conflict between preserving branch history and removing their behavior.

Restoring the grouped behavior would break the split/nonincrease requirement. Faking the old test name, adding expected-failure/skip handling, weakening independent assertions, changing the gate or rewriting pushed history is outside this order. A normal push was therefore not attempted. Root must decide a separately authorized replacement delivery/history strategy; no replacement PR or new owner was created by this worker. The requested final full-prepush-zero and remote delivery remain unfulfilled.

Review disposition

• R1975-P1/P2: accepted meaning retained locally; only the own row's split/publication state updated. Remote remains the older Draft head.
• R1975-a1 / D1 and a2: grouped correction explicitly SPLIT OUT under Root authority. Original quality handling is retained; separate grouping redesign belongs to PM's existing assignment. This worker did not dispatch it.
• Historical round-one conditions at da390b08: Anthropic/Claude Sonnet 5.5 and Zhipu/GLM-5.2 transport records remain historical, not reviews of this candidate. Original reports and raw failures are preserved.
• R1975-A1..A3 remain follow-up, unchanged.
• Review: ausstehend (Stufe B). Root commissions ONE non-OpenAI LAST delta review after checked delivery; no author review, new review round, Ready, review-ok, Mergify or manual integration.

NICHT ABGEDECKT

• Actual final red-first is RED. No final full prepush or normal push was run after that failure. The final local candidate and its resolved PLAN state are not remotely delivered; Draft head/conflict remain unchanged.
• Both absolute design gates remain red with 20 existing stored failures; only strict paired nonincrease and the inspected AppData wording are demonstrated.
• Linux does not compile/run #[cfg(windows)] tests (run_in_pty, npm_shim, ConPTY, DPAPI) or Windows clippy arms. The Windows lane runs in the merge queue. No final full Rust-suite/clippy proof this round; final commit's cargo check/fmt and precommit checks passed.
• Precommit does not cover linux-lane browser smoke, frontend build or workflow gates. Updater, OmniRoute and mirror state remain unverified.
• No native directory scan/CPU/disk/performance measurement, installed/effective, real keyboard/AT or UAT acceptance. Full KR-10e, whole Glass and all parent acceptance remain OPEN.
• lage-jetzt mirror unavailable. The current thread's get_goal returns null and has no goal-id/resume argument; no supported cross-session resume tool is available. Prior goal 01a17ffc-1891-7b80-8771-cca6cf301f8d remains recorded as active/unfulfilled under W1-03f-api: scoped-credential delivery route POST /agent/delivery #337; no replacement goal was created and no unmet goal completed.

Root disposition: replacement delivery (ENTSCHEIDEN, 2026-10-28T02:31:54Z, after Stabschef report 02:31:23Z)

• Observed (author report App395, local final cd2cbfbd, remote unchanged 64ee625): a1 grouping removed per the split, original restored; 29 tests exit 0; design:gate 396 cases with no increase (stored 20 = 20); 1440 px readable; clean-main DiffView test exit 0. But the historical commits afc6954/d5e9d919 carry Test-First trailers naming the withdrawn grouping test, so red-first is RF=1. No final prepush/push was run.
• Decision, same precedent as fix: show persisted question history in NeedsYou #1893 → fix: preserve question history and legacy attribution (KR-06b-1) #1896: a REPLACEMENT PR, no history rewrite, no force-push, no gate waiver. The owner creates a new branch from current main and commits the final tree of cd2cbfbd (only the accepted P1/P2 content, without a1) as fresh red-first commits with valid trailers (the RED test first, then the fix). The new PR body links this PR and states the tree-identity proof against cd2cbfbd.
• Review: the earlier source reviews carry over only for unchanged content. One delta review by a non-OpenAI family over (last reviewed fix(betrieb): retain AppData measurement quality #1975 source) → (new tree), then the own PLAN row (serial). Root adds review-ok only to the new head.
• This PR: after the replacement PR is opened, Root closes fix(betrieb): retain AppData measurement quality #1975 with a reference (branch and WIP stay; reopening is possible). The a1 grouping stays split out as FU-KR10E2-A1 (W3-02d: installer adapter (exit/UAC/sharing classification, no retry) #348).

[mergify]

@Cuarroc Dieser PR hat einen Konflikt mit main und kann nicht in die Merge-Queue. Bitte main hineinmergen (kein Rebase, kein Force-Push) und den Konflikt loesen; das Label conflict verschwindet danach von selbst.

[Cuarroc]

Closed by Root (2026-10-28) as replaced by #1996 (codex/kr10e2-quality-replacement, source head 6590dba), per the Root replacement decision recorded in the #1975 body. The branch codex/kr10e2-quality and its commits are kept; nothing deleted. Reopen is possible at any time.