Live punchlist audit and repair status

Local audit-as-of record, not published acceptance

This page is a local audit-as-of record with its evidence integration complete. Main’s sole final /home/loca/dev/wikis/build.sh wrdp --check-only validation passed on 2026-09-09: exit 0 in 166.80 seconds, SSOT passed, 89 Markdown files parsed, 945 files emitted to scratch, and only the expected no-git warning; execution timestamp was not supplied. Installed source, observed runtime, and human acceptance are different columns. No row claims human acceptance or closes the punchlist. The scratch evidence roots were lost; surviving sanitized receipts preserve scoped historical observations, not restored originals. The final plugin packet and after-only harness supplement are noncanonical source artifacts, not runtime or release authority. The 50-asset public-safe corpus is complete for this round but remains local and unpublished.

This is the requested audit/task list, not another session handoff. It preserves 39 working identities: 8 S03 + 11 E03 + 13 Q05 + 6 A05 + 1 L05. Excluded and test-only entries remain visible rather than disappearing from that census. Existing context: remediation masterplan and probe authority boundaries.

Baseline, incidents, and evidence authority

• The starting tree was DIRTY at mirror HEAD 715474f: Main’s historical baseline records report 35 tracked modified files and 58 untracked entries. The archive originally at /tmp/qwbase-20260908/live-uncommitted.tar survives privately as evidence/2026-09-08-live-punchlist/patch-v3/live-uncommitted.tar, SHA-256 ab0c9460382a37e27863d512d00c3c9316f3c23a8f18cd0a95cc683937c84c83, 1,167,360 bytes. It contains 35 selected modified tracked-path members, including 17 language files; it is not a full bench/tree or untracked-file backup. HEAD is not a description of installed bytes, and these historical dirty counts are not a new working-tree inventory. Per-file patch provenance uses the preserved earliest/mirror bytes rather than treating this archive as complete rollback. No inherited working changes were committed by this round.
• Main’s historical private-copy observation at 2026-09-08 11:38:36.888 UTC recorded 42/42 preserved files matching source/receipt, directories 0700 and patch 0400. That original receipt was lost with scratch; this is not a fresh 42-file parity proof. The private v2 bundle survives under evidence/2026-09-08-live-punchlist/patch/, unchanged but superseded by v3. Both are dirty-baseline deltas, not canonical-HEAD-applicable patches or blanket mirror commits.
• The surviving private v3 packet (patch-v3/final-manifest-v3.json) records a 19-path dirty-baseline delta: 13 archive-backed and 6 mirror-backed before images, exact 19/19 header allowlist, apply-check/apply/recursive-diff exit 0, 19 byte/SHA round-trips and 19 unchanged before files. Patch SHA-256 is 6bc2b946a32094b143549e2e77dfc229584ab8e8dc2f81f93f4514cf3919e059 (149,219 bytes); proof-manifest SHA-256 is 3416f141d87fa95559e3dea6c3b6c954ac9f5f7fdceb93b100e57e6dbece7e67. These are recorded offline proofs, not commands rerun by this author. patch-v3/packet-status.json overrides the older final-sounding name with IMMUTABLE_INTERIM_NOT_FINAL_DELIVERY. V3 predates the following installed Dark-footer and Builder-tooltip repairs and is not promoted to the final source packet.
• Main’s fresh durable recovery-source/current-v3-parity.json at 2026-09-08 14:09:38.425 UTC records 19/19 installed files matching the surviving v3 after hashes, all www-data:www-data and mode 0644. This proves that source snapshot survived; it proves no database/history restoration, Settings acceptance, browser behavior or final source freeze.
• The later frozen Footer-repair installation receipt identifies PdfGenerator.php SHA-256 c296224893f5d304365bde1d609959e3904c8bf00456eeba42e17cae8359fec8, 17,933 bytes, with successful PHP lint. Its scoped four-line text-color repair leaves the decorative separator unchanged. The AFTER outer capture runner was bracketed at 2026-09-08 14:56:45.539084–14:56:55.792749 UTC; these caller-clock bounds are not PHP-internal execution times. Runtime evidence below verifies the repaired footer, not a final consolidated patch or mirror commit.
• Tooltip-fixture’s installation receipt bounds are 2026-09-08 15:28:38.365965–15:28:38.871535 UTC. assets/css/qwizz-builder.css changes from SHA-256 1bc37a0520bf192b9916b0de2f3cce0fa9f834b0aadfc5c95b79e582e482bb85 to 84e348ccc20a91ea3a564a305fa759d25974726867af308f084835c843a8dcbb (44,838 bytes). Its paired browser proof reuses exactly the same 100,517-byte fixture HTML, SHA-256 0153a444ff18561055f3993e62d99745240396231d552fb7f4e0b9112d47da43. The source HTML remains private; four request-local overrides did not open stored Hidden exposure or save Settings.
• The mobile Settings sticky-offset installation receipt is bracketed at 2026-09-08 16:13:46.302434–16:13:46.513140 UTC. assets/css/qwizz-admin-responsive.css changes from SHA-256 9ee0bf11be12322f012dcc56fac4bdd9a1a0ca8f8e87e12053479ed00902931b (9,603 bytes) to 636e08c0d05a9c0fde165de556fdcf08602a0231ed94bd9a64f03bb9ddec0091 (10,050 bytes), installed www-data:www-data mode 0644 with matching metadata and return code 0. The staged diff adds exactly four declarations inside the existing max-width: 600px block: admin-bar height reserved as #wpcontent block padding, rail padding zeroed, the sticky stats strip pinned to top: 0, and the toolbar shelf offset to the dashbar-height variable (settings-mobile-sticky-20260908/install-20260908T161346Z.json, settings-mobile-sticky.diff:3-23). No PHP file changed, and the receipt records cssLinterRun: false: this bench has no CSS linter, so none ran.
• Two inherited Settings defects were then repaired without expanding the 19-file installed-source allowlist because both targets were already present. For scroll-spy, assets/js/qwizz-admin-settings.js changed from pristine SHA-256 5b5061f156a2bad3fce7998bbe5843697d20c75091b70630079fd6cb02d335c2 to installed 70a0a621e6a7073b9b890a937b8a00db2a624fd8839f7dd198860a601cd644ea, www-data:www-data mode 0644 (settings-scrollspy-20260908/receipt.json). Its one hunk replaces the wp-admin-broken bottom-of-document test, which included #wpfooter, with a maximum-scroll comparison derived from the actual scrolling element (scrollspy.diff:3-14). The container lacks Node; the recorded container attempt exits 126, while host node --check exits 0. The final cache-disabled original a-d browser gate below supplies runtime acceptance. For policy-card spacing, assets/css/qwizz-admin-settings-core.css changed from pristine SHA-256 c126fb54b27b957d80653bb44660b1755e14eb93f44d17876a08a1c18c49f5df to installed 988fa3730e01bf3231f41f4ef316b5d1297e067cfb9cd4d735a8c5bb3d56bc35, the same owner/mode (settings-icongap-20260908/receipt.json). Its one-line hunk adds column-gap: 8px to the existing 24px icon track (icongap.diff:3-10); final a-d runtime measures 80 eligible pairs with no sub-6px gap. No CSS linter exists or ran.
• The Editor-specific sticky-outcome repair is installed and runtime-accepted for its exact product geometry (editor-outcome-20260908/receipt.json; fresh f receipt below). assets/css/qwizz-finetune.css changes from pristine SHA-256 ef0496c0d7fdf5ab029bb236c26a33ed2b82ca9ec5f3d756c5fc2b497b2d14e2 to installed 75c41eae82b803f4a9dd91dbd1848ca9c53d6480555bcc0ef7877424e2d5de0e, www-data:www-data mode 0644. The six-line addition scopes position: sticky, inset-block-end: max(16px, env(safe-area-inset-bottom)) and z-index: 2 to .qwizz-editor__outcome; the Builder outcome remains static (editor-outcome.diff:3-12). The later f product assertion observes the bar within the viewport with the three canonical actions and correct hover-tool opacity. That does not turn either red f transaction into an overall green lifecycle, and the installed-source allowlist remains 19 paths.
• The separate read-only database comparison, recorded at 2026-09-08 14:23:17.816780 UTC, compares a fresh 14:14:54 UTC snapshot with the recovered primary tool output from 12:57:39 UTC (recovery-current/database-comparison-transcript-baseline.json). Raw Settings, stored-JSON hash, all 94 full backup rows/string identities and the quiz/draft fixtures match that 94-row baseline; the runtime Settings getter was not rerun. The operator has 10 backups. That later-baseline comparison alone did not prove pre-task backup parity; the independent archive replay below supplies that narrower proof. That comparison provides no exact token/session-metadata parity; the 13:23 revocation is historical primary tool output, and disappearance of scratch proves no draft/artifact cleanup. Later exact-owned-session receipts are reported below and remain transaction-scoped.
• The independent pre-task archive replay at 2026-09-08 14:42:13.916009 UTC (recovery-current/pretask-archive-independent-replay.json) compares every current seven-column backup tuple directly with the Sep-8 nightly dump completed at 01:30:43 UTC. All 94 tuples occur exactly once, leaving 93 separators and zero unmatched tuple bytes; all seven column values match and both states contain 10 operator backups. This replay reused no archived-row decoder and mutated neither database nor runtime. There is no task-era backup loss or restoration needed: the older 95-total/11-operator wording cannot establish today’s loss. Main released the history-based Settings-probe hold on this evidence; single-owner sequencing and the operator no-write rule remain mandatory. This is backup-table pre-task parity, not blanket database, session, capability-cleanup or UI acceptance.
• Private harness-snapshots/manifest.json preserves five after-only harness/original-choice replay sources; harness-snapshots-v3/manifest.json preserves seven after-only harness/four-PDF replay sources and invocations, including the original failed receipt. Recorded copy parity is 5/5 and 7/7 respectively. Executable sources are not historical runtime output or an initial harness baseline, and remain outside the vault.
• The final after-only harness/worker supplement is sealed at evidence/2026-09-08-live-punchlist/after-only-final-policy-supplement/manifest.json, SHA-256 a7d8d50ba93dfe807070c435bc669b46dac6ebe53a7616321c63a8f56fbe6eb4; its proof-commands.json is SHA-256 7f18fe729965c13217b60ac10ebdb750a9aa47a47ea5701164b0f5e90b3f2e32. It records 39 source files with 39/39 copy/current parity, including f source b0962f…, current g/Editor harness 25a12d… and Editor hook 4d14fb…. It intentionally excludes fixtures, raw results, capabilities, config, PDF copies and answer content. This is after-only source preservation, not a historical baseline, runtime replay, canonical patch, mirror commit or release authorization.
• Mirror commit is NOT PERFORMED and blocked. The durable source-only assessment at 2026-09-08 14:19:39.856642 UTC (canonical-assessment-v3/assessment.json, SHA-256 9ac7f5a4ae1cb1be0c5d3121c3bc947efa22a17ffc50760b26bbba0b6784e80a) finds 13/19 dirty-baseline files differ from canonical 715474f; a disposable canonical git apply --check fails in 7 paths, exit 1. No actual index inspection/mutation or commit occurred. A complete-file mirror add would include inherited changes; naive selective staging is not proved safe. The release owner must decide inherited prerequisites and the canonical target/port, then authorize narrow staging after checking the real index for unrelated work. The grounded dirty-baseline packet remains deliverable as a noncanonical artifact, with releaseEligible: false; v3 remains an immutable interim packet and the separately sealed final packet is described next.
• The final plugin-source packet was sealed after midnight on 2026-09-09 while retaining this audit’s 2026-09-08 round identity (evidence/2026-09-08-live-punchlist/patch-final-plugin/final-manifest.json). Manifest SHA-256 is 1e43a3f36cbe7fa8c93d176673794794a5fafc0d650c71de86adf04903bab785; its 152,039-byte today-source-final.patch is SHA-256 d1bc84e75939f5fa449b8127c58f271bb59c0e26a7bfb03443d75a4e3083a93c. Against the exact dirty before images, all 19 allowed plugin paths pass strict apply-check, apply, recursive byte comparison and live-freeze hash comparison. Status is PLUGIN_SOURCE_PACKET_OFFLINE_PROVED_RUNTIME_GATE_INCOMPLETE, releaseEligible:false: this is not canonical applicability, runtime replay, public publication or release approval. The harness/worker supplement is sealed separately as the after-only 39-file, 39/39-parity source artifact described above; v2/v3 remain immutable historical packets.
• Historical baseline shell gate: 94 assertions, 6 failures were previously read at /tmp/qwbase-20260908/wave-all-baseline.tap:285-286. The TAP bytes are now lost; the surviving report/transcript observation is not a replacement receipt or a new run. The saved after-only harness cannot recreate that historical baseline. Later reconciled Wave 9 and all-wave results are reported below and do not retroactively replace this baseline.
• Page 1289 was accidentally published during verification, then restored to draft; its modification timestamp was not restored. Main historically reported wp post update 1289 --post_status=draft and guest 404. The previously inspected read-only snapshot at 2026-09-08 10:14:51.055227 UTC recorded page 1289 draft, its two original library shortcodes, stored post_modified 2026-09-08 07:51:13 and quiz fixtures 1284–1288 published (/tmp/qwaudit-20260908/library-fixture-before-final-gates.json). That original snapshot is now lost: these are retained transcript/report observations, not currently retrievable raw evidence or post-recovery state. The stored modification field is not action-time precision. Status restoration was not byte-exact restoration; fresh Library/unchangedness proof remains pending.
• The live public Builder/editor surface is https://wrdp.loca.zone/; the supplied /?page_id=57 fixture returned 404. Settings is https://wrdp.loca.zone/wp-admin/admin.php?page=qw-settings. A stale fixture is not proof that the actual Builder is unavailable (B entrypoint-404.png; E desktop-open.json:2-3).
• Builder and editor pre-repair reports describe cache-disabled Chromium at 1512×945 and 390×844. The recovered Settings corpus does not carry a usable capture timestamp, event log, or cache-disabled receipt; it is qualified inherited observation, not a newly established browser pass (S FINDINGS.md:9-15).
• The frozen source-only review (agent://ReviewFrozenUi) found no actionable patch-introduced defects in the eight frozen Builder/Settings files against their exact before snapshots. It traced timing/draft payload consumers, resize/offset handling, scoped mobile overflow, category styling and retained keyboard history. This is not a browser result, security sign-off, acceptance of the inherited dirty baseline, or human acceptance; no validation commands or mutations were run by that reviewer.

Evidence keys and publication boundary

The current public-safe corpus contains 28 JSON receipts and 22 inspected/decoded PNGs (50 assets), plus a file manifest carrying stored-file SHA-256, roles, limits and reproduction commands. Asset count is not an item-pass count. All 50 asset digests/lengths match, every JSON parses and every public asset is mode 0644; all 22 PNG stored-byte and decoded-pixel digests are distinct. JSON hashes identify projections, never byte replacements for private originals. The first eight PNGs retain historical lossless-copy provenance; original-source parity is not newly re-measured after source loss. The other fourteen PNGs are explicit derivatives: four recovered Settings, two footer-only, three labeled conditional-fixture, two mobile-offset, one fresh Settings-surfaces and two paint-masked Editor viewport frames. No image count establishes interaction or human acceptance.

The original /tmp/qwaudit-20260908, /tmp/qwfix and /tmp/qwbase-20260908 scratch roots are no longer retrievable. Code-span paths and line citations into those roots below are historical provenance pointers, not surviving attachments. Unless an individually linked sanitized receipt or explicit durable private copy covers the claim, the underlying Builder/editor/Settings/install/library observation is now transcript/report-only. A preserved source digest or before-image replaces source provenance only, not the vanished browser, recovery or TAP output. The previously withheld interruption summary was never a public receipt; neither it nor source parity establishes capability revocation or database/history restoration. The three inline-only missing screenshots remain missing, and future captures must carry new timestamps rather than impersonate them.

Key	Existing evidence and timestamp boundary
WD	local://ledger-2026-09-05.md:1-7,9-49: condensed prior-assistant working digest, explicitly not verbatim. It records the 2026-09-03 and 2026-09-05 pick identities.
B	/tmp/qwaudit-20260908/builder/: pre-repair observations dated 2026-09-08; individual files without a capture timestamp are not assigned a fabricated time. FINDINGS.md:17-28 maps the raw observations to working IDs.
B-install	/tmp/qwaudit-20260908/builder-repair.json: installed-source receipt recorded 2026-09-08 08:05:14.635806 UTC, explicitly deferring browser validation (:3-4,179-184).
E	/tmp/qwaudit-20260908/editor/: pre-repair observations; desktop-open.json:2 is 2026-09-08 05:36:55.728 UTC. Other interactions are dated 2026-09-08, with their own capture fields authoritative; that desktop timestamp is not assigned to every interaction.
E-install	/tmp/qwaudit-20260908/editor-repair.json: initial layout repair recorded 2026-09-08 06:33:38.147608 UTC. This receipt predates the later toolbar/PDF repairs and cannot certify them.
S	/tmp/qwaudit-20260908/settings/: inherited Settings corpus; filesystem mtimes 2026-09-08 05:33:22.869909–05:42:29.882180 UTC, not capture times. FINDINGS.md:39-46 identifies raw per-check evidence and limits.
S-install	/tmp/qwaudit-20260908/settings-repair.json; chronology at S FINDINGS.md:13,21-29: CSS installed 08:12:27.927784–08:12:28.591054 UTC, JS frozen 08:14:18.869855 UTC, provenance check 08:16:22.050177 UTC, all 2026-09-08.
PDF-intermediate	/tmp/qwaudit-20260908/main-pdf-round-canonical.json:2-44: Main’s independent extraction receipt at 2026-09-08 09:38:13.571 UTC, explicitly intermediate, before server-owned PDF-context correction.
Native-print pair	Main’s independent extraction of private /tmp/qwfix/editor-controls/native-before.pdf and native-after.pdf: Correct answer labels 57→0, authored 57-question count label retained in both. Main supplied no separate extraction timestamp; do not borrow PDF-intermediate’s time for this pair. Publish only sanitized counts/hashes.
Theme-hold	/tmp/qwaudit-20260908/theme-proposals/q05-03-refined-b-approach.json:2-21: staged/on hold, installed:false, liveRuntime:false; no capture timestamp.
Choice-freeze	/tmp/qwfix/editor-recovery/installed-sha256.json:2-3,18-49,64-67: original-choice protection source installed/frozen; three changed files with hashes, ownership and mode. No capture timestamp or validation result; validationPerformed:[]. Predecessor interrupted-browser residue is explicitly unknown to recovery.
Theme-freeze	/tmp/qwfix/theme-ideas/installed-sha256.json:2-49: three theme files installed after rebasing on Choice-freeze; no standalone validation by that installer. Supersedes Theme-hold’s installation status, not its approval boundary.
Final-B	Builder receipt: geometry at 2026-09-08 10:32:16.320 UTC; timing action results have no separate timestamp. Actual keyboard/click/reload observations; no Settings writes. Desktop setup supplied by Main: 1512×945, DPR 1, cache disabled; setup timestamp unavailable.
Final-E	Desktop editor receipt: replace at 10:40:12.580 UTC, geometry at 10:40:52.473 UTC, choice actions at 10:44:35.323 UTC, boundaries at 10:45:37.639 UTC, all 2026-09-08. Badge action timestamp unavailable. Desktop setup supplied by Main: 1512×945, DPR 1, cache disabled; setup timestamp unavailable. Raw choice markup omitted.
Final-PDF	Restricted replay at 2026-09-08 10:26:18.213613 UTC; restricted actual-byte counts at 10:27:58.586 UTC. Positive replay has no separate timestamp; positive actual-byte counts at 10:49:39.454 UTC. All 2026-09-08. These sanitized historical receipts survive; original PDF bytes/text are lost.
Final-mutation	Intentional-red mutation receipt: process-only mutation, no live-source mutation; intended uniform-response assertion fails, cleanup assertions pass. Run timestamp unavailable. Public copy contains only allowlisted check fields and aggregate counters, not per-index response details.
Final-mobile	touch receipt: 390×844, DPR 1, touch enabled, isMobile:false, cache disabled explicitly configured; geometry/trusted touch at 2026-09-08 11:02:53.112 UTC, long-answer measurement at 11:13:33.774 UTC. Builder narrow receipt at 10:58:49.908 UTC records cache-disabled 390×844 with one touch point.
Final-Print	Native Print receipt at 2026-09-08 10:55:33.750 UTC: completed PDF attachment and exact canonical-preview bytes. Native multipart POST values were not established by the failed URLSearchParams parse and are excluded as evidence.
Oracle-provenance	Code-path reconciliation at 2026-09-08 10:52:22.005 UTC: identical embedded gate body, preserved old validator body minus exactly five opening-tag bytes; fixed response count 1, intentional mutant 12. No installed-source mutation; private source paths omitted.
Fresh-browser	Isolated browser setup at 2026-09-08 11:33:18.714 UTC; Kind result settled at 11:38:45.794 UTC, captured at 11:39:44.071 UTC, cache-disabled 1280×900. Fresh mobile Builder capture at 11:40:10.600 UTC and fresh mobile editor capture at 11:41:20.348 UTC are cache-disabled 390×900 with mobile/touch emulation, not the prior 390×844 actions.
Theme-fixture	Final theme fixture receipt: desktop mouse/keyboard at 11:53:12.748/11:53:12.943 UTC and 390px mobile at 11:54:31.694/11:54:31.953 UTC, 2026-09-08, cache disabled. Actual installed PHP template/CSS/JS with process-local six-theme exposure; no stored Settings write or PDF request. Live exposure remains Hidden. Earlier isolated-world/980px tooling sample discarded; top-level receipt time is not assigned to later mobile actions.
Raster-correction	Historical PDF raster correction: actual Dark/on page-1 source capture at 2026-09-08 12:54:05.560 UTC; inspection/correction at 13:04:56.013 UTC. Twelve requested captures, zero complete-page captures proved; automatic footer-only crops invalid. The separately inspected Dark/on footer contrast sample is a historical defect observation, not all-page proof. No original raster/PDF bytes from that run survive publicly or privately.
Footer-repair	Fresh footer-color comparison: BEFORE outer tool-event bounds 2026-09-08 14:33:59.667–14:34:22.147 UTC; AFTER caller-clock bounds 14:56:45.539084–14:56:55.792749 UTC, neither PHP-internal instants. Visual receipts began at 14:47:07.083 and 15:00:04.925 UTC. Each set has 29/29 PHP checks, four PDFs, 10 rows/10 distinct sources per PDF and 12/12 fully contained rendered pages. Both On cases show 1/3, 2/3, 3/3; both Off cases have no labels. Dark text contrast changes from 1.1631:1 to 15.2975–15.3267:1. Inputs match within each four-case set, not across independently randomized before/after sets. Public images are footer-only derivatives; no question/key content, body-identity or human-acceptance claim.
Recovered-S	Recovered historical Settings observations: four transcript-derived primary-tool projections, not restored scratch originals or a new run. The 2026-09-08 13:01:40.590–13:05:51.342 UTC manual surfaces lifecycle records separate S03-01/02/03/05/06/08 passes. Desktop S03-04/07 pass locally, but the enclosing 13:07:44.178–13:12:20.096 UTC lifecycle retains overallPass:false and restorationComplete:false. Both modes deliberately skipped built-in a-d assertions. Named contract results do not reconstruct missing full event/AX traces or current cleanup.
Tooltip-fixture	Paired conditional Builder tooltip comparison: BEFORE 2026-09-08 15:25:01.936–15:25:51.994 UTC; AFTER 15:29:03.566–15:29:48.677 UTC. Identical fixture HTML and paired geometry, cache-disabled desktop pointer and 390px native-touch emulation; white-on-white Toggle this option becomes readable dark text on white in OFF/ON states. Native footer field '' → '1' → '', miniature hidden/1/hidden and English→German selection/pressed states are retained on both surfaces. Main’s independent receipt at 15:45:39.316 UTC matches all 12 screenshot byte counts/digests but visually inspects only the three named public-source frames. Stored Hidden exposure is unchanged; no Settings save, quiz submission, PDF request, physical-device or all-theme claim.
Database-comparison	Fresh database comparison and independent pre-task supplement: the 14:23:17.816780 UTC comparison records the 14:14:54 UTC snapshot matching the transcript-derived 12:57:39 UTC baseline, 2026-09-08. At 14:42:13.916009 UTC, independent replay also matches every one of the 94 seven-column backup tuples to the Sep-8 01:30:43 UTC pre-task dump, with 10 operator backups in both and zero unmatched tuple bytes. No task-era backup loss/restoration needed; older 95/11 wording does not establish today’s loss. Neither comparison proves current token/session parity or draft/artifact cleanup.
Mobile-offset	Matched fresh mobile Settings pair, both cache-disabled 390×900/DPR1 with isMobile/hasTouch true. BEFORE final-runtime-20260908/final-browser-manual-mobile-before-20260908T155103Z.json observed 2026-09-08 15:38:42.364–15:39:33.564 UTC inside a 15:36:03.807–15:41:41.005 UTC lifecycle with overallPass:false, restorationComplete:false and zero accepted writes. AFTER manual-gate-recovery/final-browser-two-manual-history-mobile-helper-after-20260908T194538Z-observation.json observed 19:50:55.210–19:52:11.889 UTC inside a 19:49:23.978–19:52:57.901 UTC lifecycle. The 155103 and 194538 stamps are parity/folder labels, not capture starts. Both runs skipped the original a-d assertions, which remain required. The AFTER lifecycle is still overallPass:false with complete:false restoration for exactly one cleanup action, restore operator browser authentication, which failed with CDP Page.navigate timed out; the named localStorage, Settings-document, backup-table and disposable-administrator actions each passed. Its separate raw parity at 19:47:51→19:55:56 UTC matches 94 backup rows across all 658 seven-field values, the Settings option, both protected posts, zero former harness users and the two original operator session rows; that parity does not green the failed cleanup action or prove window-wide artifact cleanup. Public projections omit row, user and post identities, revision and state hashes, cookies and network payloads.
Shell-final	runtime-20260908T182215Z/: initial wave-6.tap 1..14 zero failures with nested 18/18 PHP; initial wave-9.tap 1..21 one failure on admin_tree output identity; after the authorized pin reconciliation, wave-9-reconciled.tap 1..21 and wave-all-reconciled.tap 1..98 with zero failures, plus bash -n pass, mixed Library mutation packets, a post-all tree-digest match and 19:32:12 UTC state parity. Provenance for the pin change is tree-pin-audit-20260908/receipt.json (18:48:00 UTC static audit) and tree-pin-reconciliation/change-receipt.json. TAP files carry no embedded execution clock, so directory names are not run times, and no CSS linter exists on this bench.
Fresh-Surfaces	Public-safe Settings surfaces projection, sourced from frozen private receipts: separate cache-disabled manual lifecycle 2026-09-09 00:30:52.886–00:33:08.035 UTC; browser observation 00:32:01.988–00:32:41.583. Its nine wrapper assertions pass, including trusted keyboard provenance and distinct S03-02/03/05/06 contracts on 1280×900 desktop and 390×900 touch mobile. Full AX-tree and visible-DOM denominators each show one Search and one Preview opener; trusted Enter/Escape proves open/close/focus return and Search filter/clear/collapse. Wrapper restoration is complete; owned Chrome exits 00:33:09.728, the exact session is revoked 3→2 at 00:34:16, and post parity at 00:34:32 matches Settings bytes, 94 backups, two protected pages, two operator sessions, zero former harness users and the debug boundary. Original a-d did not execute in this lifecycle. The raw desktop screenshot exposes the disposable account and remains private; the safe full-frame mobile derivative contains no account identity and visually shows adjacent, unclipped Search and Preview controls.
Response-observer	Corrected f authority is private standalone-gate-worker/private-final-browser-two-f-response-20260909T010816Z.json, 2026-09-09 01:11:16.633–01:12:26.696 UTC: 5/5, original/wrapper/overall pass, zero console/page errors and exact cleanup of one captured draft, one artifact and one private file. The offline synthetic observer proof at fetch-response-proof/run-b1deb7cf-7a92-40f2-9867-98d25338cf74/receipt.json passes 24/24 and kills four mutants; it is not live cleanup proof.
Final-g	Corrected g authority is private standalone-gate-worker/private-final-browser-two-g-repair-20260909T013431Z.json, 2026-09-09 01:37:30.133–01:38:32.161 UTC: 5/5, original/wrapper/overall pass, zero console/page errors and exact cleanup of one captured draft, one artifact and one private file. It awaits initial preview capture before Replace; this corrects a harness barrier, not the already-passing product behavior.
Final-Editor	Public-safe Editor projection, sourced from private final result/observation digests, covers the separate cache-disabled 2026-09-09 02:00:48.159–02:02:11.479 UTC lifecycle. It passes 8/8 including E03-08/10/11/12 across 1280×900 desktop and 390×900 touch mobile, with nonzero DOM/AX/tool denominators, native text undo, zero horizontal overflow, zero exposed history buttons and exact owned-capability cleanup. Original f is not executed or replaced.
Final-harness-source	Private after-only-final-policy-supplement/manifest.json, SHA-256 a7d8d50ba93dfe807070c435bc669b46dac6ebe53a7616321c63a8f56fbe6eb4, and proof-commands.json, SHA-256 7f18fe729965c13217b60ac10ebdb750a9aa47a47ea5701164b0f5e90b3f2e32: 39 after-only source files with 39/39 copy/current parity. Raw results, fixtures, capabilities, config, PDF copies and answer content are excluded; no runtime or canonical applicability follows.
August-S03	evidence/2026-08-15-visual-ia/audit/reports/S03SettingsShell.json:34-197, with caveats in adjacent verification.json:34-100: a separate dated canonical namespace, not WD wording.

The first eight stored visuals are historically inspected lossless PNG copies with recorded magic, dimensions and source/destination SHA parity; no byte/pixel duplicates were found within that original eight-image group:

• Final desktop editor: actual desktop surface; action proof is in Final-E.
• Final estimated boundary: preview-only boundary, not exact PDF pagination.
• Baseline wrong-command touch state: baseline, not the repaired touch replay.
• Baseline cross-row state: baseline obstruction context.
• categories: 11:40:10.600 UTC, 2026-09-08; not the Completion/icon-gap area.
• Fresh initial Mixed editor: 11:41:20.348 UTC, 2026-09-08; approved original-choice guidance visible; not post-Kind, old touch or long-answer evidence.
• Desktop theme fixture: 11:53:12.992 UTC, 2026-09-08; red FIXTURE ONLY banner, six named chips, Retro selected.
• Mobile theme fixture: 11:54:31.966 UTC, 2026-09-08; red FIXTURE ONLY banner, 390×900. Neither fixture is live visitor exposure or PDF theme-parity proof.

Four additional recovered Settings visuals are PNG derivatives, not replacements for lost original screenshot PNGs. Three desktop frames remove the complete top 32px admin bar; the mobile frame has no account/admin-bar region and is a full-frame PNG reencoding. All four public derivatives were visually inspected for account identity, admin bar, email and API keys; none is visible. Exact private tool-return bytes remain retained, and the manifest records crop/encoding lineage and source/output hashes:

• Recovered mobile General derivative: tool return at 2026-09-08 13:07:10.930 UTC; exact original capture time unrecovered. Full-frame derivative, not a new mobile run.
• Recovered desktop General admin-bar crop: tool return at 13:07:11.320 UTC; exact original capture time unrecovered. The complete top 32px was removed.
• Recovered desktop categories admin-bar crop: lost PNG capture event at 13:11:15.837 UTC; recovered WebP tool return at 13:15:37.524 UTC. PNG derivative after 32px top crop; not the lost PNG’s bytes.
• Recovered desktop error admin-bar crop: lost PNG capture event at 13:11:09.122 UTC; recovered WebP tool return at 13:15:38.968 UTC. PNG derivative after 32px top crop. A visible deliberate-error frame is not an indefinite-persistence or overall-lifecycle pass.

Two fresh 794×90 PNGs preserve only the authorized footer crops from Footer-repair: Dark before and Dark after. Both were visually inspected and contain no question/key page or identity. They show the text-color repair and unchanged decorative separator; private full-page captures, not these crops alone, support the separate 12/12 coverage and all-page-label observations.

Three additional conditional Builder tooltip derivatives add a 52px external CONDITIONAL FIXTURE ONLY / SAFE DERIVATIVE banner while preserving the complete original 900px-high image area. All three public derivatives were visually inspected; no crop or identity redaction was needed. They are new fixture evidence, not historical recovered screens or live visitor exposure:

• Mobile tooltip BEFORE: original capture 2026-09-08 15:25:44.433 UTC; ON state with an empty-looking white bubble.
• Mobile tooltip AFTER: original capture 15:29:42.062 UTC; native-touch ON state visibly reads Toggle this option.
• Desktop tooltip AFTER: original capture 15:29:18.540 UTC; native-pointer ON state visibly reads the complete hint. Main’s visual inspection covers these three original frames, not all 12 whose bytes/hashes were independently verified.

The earlier three mobile helper images displayed inline but did not persist; their JSON measurements remain durable, not invented stored screenshots. Fresh mobile/fixture frames have their own timestamps and do not replace those originals. Static frames alone do not prove interactions. No production publication occurred.

Current public-safe corpus

The local manifest, SHA-256 a3025a433ddb8fbf6e26b30a4e2a7f2c4dcd1271ede47261fcc4edd3e076c118, inventories 50 settled assets: 28 JSON projections and 22 PNGs. The manifest and directory agree exactly; every retained file matches its recorded byte length and SHA-256, every JSON parses, and every public file is mode 0644. All 22 PNGs decode without stored decoder metadata; their stored-byte and complete decoded-RGBA digests are each unique. The first 47 entry objects were retained and the three Editor entries appended.

The final three additions are Editor runtime, SHA-256 52856020258ce2c719a35dd1d6ca24ec753c93344a62ee87f99dabed4a2fc9cd, 8,978 bytes; desktop Editor frame, 2d4f4a93b5dead104316d3246768f447a847c64c8e0f237d464f177bd0a32fc4, 55,576 bytes; and mobile Editor frame, 5f84606332b7e7f7fee5a2817aaa7a18b87e4be729f73ee1e9f1008e46dd2d9f, 41,822 bytes. They project the corrected policy-aware Editor lifecycle only; original f remains separate.

The two Editor PNGs were captured after a temporary CSS paint mask hid the admin bar, editor text paint and designated private chrome. The hook verified all 49 eligible fields and six private-chrome nodes masked on each viewport, zero changes across 984 sampled rectangles per viewport, then removed the mask in finally and verified geometry restoration. Independent inspection found no visible account/admin identity or obvious personal data. This is visual masking, not content deletion: red spellcheck decorations remain, and the frames prove visible Editor geometry/control chrome only—not underlying DOM/value deletion, text readability, cropped pixels, interaction provenance or human acceptance.

The earlier 47 assets retain their stated provenance and limitations. The mobile offset comparison is still shallow BEFORE versus deep AFTER, not matched same-depth evidence. The corrected f and g transactions are current scoped green with exact owned-capability cleanup, but earlier unknown request/capability identities remain unresolved and no global cleanup is claimed. The public evidence remains local; no production publication occurred.

Decisions recorded during this round

These are recorded decisions supplied by Main from the current conversation, not reconstructed quotations and not acceptance of implementation:

1. Quantity remains Settings-bound. The current fixed-10 configuration remains fixed at 10; it is not to be bypassed merely to expose a slider.
2. Timing uses Manual/Auto plus seconds. The installed Builder contract preserves Manual as zero submitted delay and a remembered automatic delay (B-install :152-169).
3. Category controls use quiet hierarchical chips, with selection and focus distinct; this approves a direction, not an unobserved post-install appearance (S FINDINGS.md:101-107).
4. Page boundaries are preview-only estimates, not exact PDF pagination. Header question count means explicit document-header metadata, not question/answer ordinal visibility.
5. Theme taste was delegated after alternatives were presented/inspected. Main chose refined B; the user did not pick A/B/C. Theme-freeze records installed source: chips and Theme label absent when Hidden or only one theme is enabled, not replaced by disabled chips. Installed source is not human visual acceptance.
6. Protect original choices is the latest user-selected answer-integrity policy: prevent deletion of original pool choices while preserving text edits, reorder, Add, and deletion of added choices. Choice-freeze records installed source; Final-PDF and Final-E provide the scoped observed results below. Policy approval is still separate from human acceptance.

All 39 retained working identities

Wording rule: every requirement below is a WD paraphrase, not a verified original user quotation. The attachment was truncated around Q05-11; complete original Q05-12/Q05-13 and A05-01..03 wording was not recovered. A05-01..04 individual mapping remains incomplete. The report does not manufacture exact language from adjacent controls. “No acceptance recorded” means no item-specific human acceptance of the integrated after-state, even where a design choice was approved.

S03 — September working picks, not August canonical findings

The recovery’s REC-01..08 checks help locate evidence for these working picks. They are local recovery identities, not canonical August remappings.

Working identity / wording provenance	Installed state	Observed runtime evidence + timestamp	Human acceptance
S03-01 — WD: core-owned Screen Options stays top-right and unobstructed.	Retained unchanged; S FINDINGS.md:39.	Inherited pre-install overlap observation has no capture time. Recovered-S’s manual-only surfaces lifecycle at 2026-09-08 13:01:40.590–13:05:51.342 UTC records the distinct S03-01 contract passing; the original built-in assertions were not executed. This recovered primary output is historical, not a new run or full Screen Options event trace. Fresh helper-backed mobile AFTER, actual observation 2026-09-08 19:50:55.210–19:52:11.889 UTC at cache-disabled 390×900/DPR1 with isMobile/hasTouch true, maxTouchPoints 1 and a coarse pointer, records S03-01 passing: the Screen Options centre plus four edge midpoints are hit-testable at page top and the native control opens and closes under two trusted touches. Rounded-corner misses and ordinary scroll-under-sticky-header behaviour stay diagnostics, not this contract. The original a-d assertions were again not executed, and the enclosing lifecycle is not green.	No acceptance recorded.
S03-02 — WD: preview eye beside Search in tab bar; open/close returns focus, including narrow.	Placement retained.	Recovered-S’s 13:01:40.590–13:05:51.342 UTC lifecycle first records the separate contract passing. Fresh-Surfaces now supplies the missing causal trace at 2026-09-09 00:32:19.160 UTC desktop and 00:32:39.064 UTC mobile: the sole Preview opener holds focus before trusted Enter, the open surface focuses its close control, trusted Escape hides it and returns focus to the opener. Visible DOM and the full AX tree each have one Preview opener on both 1280×900 desktop and 390×900 touch mobile. The separate lifecycle restores successfully; original a-d is not executed there.	No acceptance recorded.
S03-03 — WD: one accessible icon-only Search trigger.	Existing trigger and decorative input glyph retained; not two actions (S FINDINGS.md:41).	Recovered-S records the contract passing and one visible trigger. Fresh-Surfaces at 2026-09-09 00:32:17.509 UTC desktop and 00:32:38.447 UTC mobile measures exactly one visible Search trigger and exactly one Search settings button in the full AX tree on each surface. The safe mobile derivative visually shows the focused Search control adjacent to Preview without clipping. Original a-d is not executed in this separate lifecycle.	No acceptance recorded.
S03-04 — WD: no idle saved message; brief accessible save confirmation, persistent errors.	v3’s recorded 12:32:40.731888 UTC conditional-controls install includes Settings JS; installed v3 hashes match at 14:09:38.425 UTC.	Earlier transcript-only observation: 12:05 success lasted 2633.4ms; an ordinary error visible at 12:06:22.850 UTC was absent by 12:06:26.205 UTC, a sampled 3.355s interval. Recovered-S later records zero idle messages at 13:09:45.026 UTC and polite atomic success at 13:10:00.927 UTC, gone at 13:10:03.623 UTC (2696.1ms). A process-local ordinary 503-style rejection before network remained visible at 13:11:05.738 and 13:11:09.108 UTC, 3.370s between samples; server revision unchanged. The S03-04 check passed, but its enclosing desktop lifecycle was false. These are scoped historical observations, not an all-errors or final-gate pass. In the fresh mobile AFTER run the polite atomic Settings saved. status appears at 19:51:27.572 UTC and is gone at 19:51:30.173 UTC, a measured 2601.4ms lifetime, with zero idle messages at 19:51:16.285 UTC. A process-local ordinary 503-style rejection before network at 19:52:02.394 UTC stays visible in a role="status" node at 19:52:04.615 and 19:52:07.839 UTC, with the server revision unchanged and the local value restored. S03-04 passes on the touch surface; this is one injected ordinary error, not all error classes, and the enclosing lifecycle is still false.	No acceptance recorded.
S03-05 — WD: companion preview-placement/focus-return pick.	No proven duplicate opener removed.	Kept distinct from S03-02. Fresh-Surfaces at 2026-09-09 00:32:17.509 UTC desktop and 00:32:38.447 UTC mobile records one visible opener and one Preview guest experience AX button on each surface; trusted Enter/Escape then provides the complete causal focus-return trace at 00:32:19.160 and 00:32:39.064 UTC. The separate lifecycle restores successfully; original a-d is not executed there.	No acceptance recorded.
S03-06 — WD: reveal/focus Search; Escape clears/closes/returns focus; sticky/narrow works.	Search handlers retained unchanged.	Fresh-Surfaces replays the interaction at sticky-scroll position on cache-disabled 1280×900 desktop and 390×900 touch mobile. At 2026-09-09 00:32:22.100 UTC desktop, Search opens focused at toolbar top 106 and scrollY 1114, trusted typing filters Science, and trusted Escape clears, collapses, hides the input and returns focus. The mobile trace at 00:32:40.150 UTC repeats this at toolbar top 74 and scrollY 1141. Every Enter/Escape keydown/keyup is trusted in the expected sequence. The separate lifecycle restores successfully; original a-d is not executed there.	No acceptance recorded.
S03-07 — WD: no history arrows; Settings keyboard undo/redo remains.	Keyboard-history source retained; no visible buttons added.	Recovered-S records zero visible history buttons at 13:09:45.026 UTC. Four separately accepted category writes—off, Ctrl-Z, Ctrl-Shift-Z, Ctrl-Z—have agreeing DOM and persisted REST states false/true/false/true, accepted at 13:10:04.721, 13:10:33.612, 13:10:46.410 and 13:10:56.960 UTC. S03-07 passed; the enclosing 13:07:44.178–13:12:20.096 UTC desktop lifecycle still has overallPass:false and restorationComplete:false. No overall green or current-session restoration follows from the scoped pass. The fresh mobile AFTER run records zero visible history buttons at 19:51:16.285 UTC and four accepted writes—category off, undo, redo, undo—with agreeing DOM and persisted REST states false/true/false/true, accepted at 19:51:30.825, 19:51:41.364, 19:51:50.781 and 19:51:58.421 UTC under four trusted key chords and two trusted touches. Every write was made by disposable administrator 1125, never the operator. S03-07 passes; the enclosing lifecycle remains false because operator browser authentication could not be restored.	No acceptance recorded.
S03-08 — WD: shelf adjacent to General; no blank band; anchor headings visible.	S-install publishes measured dashbar height; shelf geometry retained (S FINDINGS.md:89).	Recovered-S’s manual observations at 13:03:30.234–13:04:47.059 UTC measure General top equal to shelf bottom: 169.578125px desktop and 184px mobile; five reachable segments on each surface. Its manual-only lifecycle records S03-08 passing. At 13:09:45.026 UTC desktop equality is again 284.578125px. This establishes sampled adjacency. The later final original a-d gate at 20:56:29.383 UTC separately exercises all nine visible Settings workspaces: every clicked tab remains aria-current after settling, and each non-final heading lands at the measured toolbar offset; that fills the anchor-heading trace only for this cache-disabled 1280×900 gate, not the mobile Screen Options path. The installed mobile offset repair is verified by a before/after counterfactual pair at different scroll depths; no deep-before sample exists, so this is not a matched same-depth observation. BEFORE at 15:39:06.046 UTC, scroll 100: the absolute admin bar has already left the viewport at bottom −44, yet the sticky stats strip keeps top: 46px and renders 46–120, leaving a 46px phantom band, and only one of the sampled Screen Options points is hittable. AFTER at 19:52:11.706 UTC, deep sample scroll 2456: the admin bar sits at bottom −2410, the stats strip has top: 0px and renders 0–74 flush to the viewport, the toolbar shelf sits at top: 74px immediately beneath it, gap 0px, with geometry and pinned denominators 1 and zero phantom offset suspects; all three requested 0/46/100 rail-shelf samples have denominator 1 and 0px overlap. Main independently inspected two AFTER frames and confirms the admin bar gone, the stats strip flush at the top with no band above its 74px height, and readable, non-overlapping, non-overflowing 390px chips with the toolbar directly beneath. Native Screen Options still drives the category tier at Standard value 2 with the server revision unchanged. At scroll 100 the shelf is still at top 158 and Screen Options scrolls beneath the sticky header, so this is pinned-band correctness, not a claim of full sticky reachability. S03-08 passes as runtime acceptance only.	No acceptance recorded.

E03 — printable editor picks

Working identity / wording provenance	Installed state	Observed runtime evidence + timestamp	Human acceptance
E03-01 — WD: answer ordinal 2. beside letter B was tiny; legible at print scale (WD:31), not question numbering.	Retained editor answer-badge target; no new mapped repair. Original verbatim wording and whether “print scale” meant actual PDF or editor scale are not recovered.	E pre-repair desktop at 2026-09-08 05:36:55.728 UTC: editor answer ordinal 14.6667px/11pt, letter 14px (desktop-open.json; ordinal-badges.png; E FINDINGS.md:15). This is unscaled editor-sheet/narrow-view evidence, not physical-print proof. Main reports actual PDFs have no editor B2 badges; a separate inspection timestamp is unavailable. PDF question numbers and the fresh footer repair cannot substitute for this target.	Blocked on editor-print-scale versus physical-PDF intent; no acceptance recorded.
E03-02 — WD: one compact leading letter+ordinal badge.	Compact-badge repair retained in Theme-freeze.	E pre-repair: 60.03px badge, roughly 12px circle-edge and 21px glyph gap. Final-E, 2026-09-08, no separate badge timestamp: all 41 recorded badges width 44.03125px, gap 4px, text-bottom delta 0.	No acceptance recorded.
E03-03 — WD: same badge on one baseline.	Same compact-badge repair, identity preserved separately.	E pre-repair alignment alone did not satisfy compactness. Final-E, 2026-09-08, no separate badge timestamp: 41/41 samples have text-bottom delta 0 and 4px gap. No physical-print inspection claimed.	No acceptance recorded.
E03-04 — WD: Add answer is an accessible icon in row actions, not a full-width row.	Icon rule retained; original-choice protection installed with added-choice removal retained.	E pre-repair: named 32×32 icon, touch Add 4→5/remove→4. Final-E at 10:44:35.323 UTC, 2026-09-08: four originals plus A/B gives six; removing A gives five, preserves all originals and B, leaves one removable control. Final-mobile locates Q3 Add in a 44px-high row below the 32px handle; final Add action itself was desktop, not replayed as touch in this receipt.	Earlier icon ruling recorded; no after-state acceptance.
E03-05 — WD: actions below unchanged handle, never over answers.	In-flow tools row retained; mobile Q3 handle remains 32px, action row 44px high below it.	E pre-repair tools overlap first answer by 76.55×17.67px. Final-E at 10:40:52.473 UTC and Final-mobile at 11:02:53.112 UTC, 2026-09-08: each has 10 rows/70 controls, zero cross-row intersections, 40 answers/zero clipped, no horizontal overflow. Final-mobile trusted touch hits Q3 Reorder, count stays 10.	Below-handle ruling retained; no after-state acceptance.
E03-07 — WD: same below-handle/no-obstruction contract.	Same in-flow repair; original 32px handle retained.	E pre-repair Q3-handle tap duplicated Q2, count 10→11. Final-mobile at 11:02:53.112 UTC, 2026-09-08: 10 rows/70 controls, zero cross-row intersections; trusted touch event targets Reorder Question 3, count remains 10. This reproduces the specific wrong-command tap path without conflating it with a full drag/reorder proof.	No acceptance recorded.
E03-08 — WD: remove helper prose and presentation.editor_hint, not a no-op setting.	Obsolete presentation.editor_hint removal retained (E FINDINGS.md:21); the later explicitly approved original-choice protection explanation is distinct and remains visible.	The corrected policy-aware Editor observation at 2026-09-09 02:01:36.724–02:01:49.750 UTC passes on 1280×900 desktop and 390×900 touch mobile: 20 rendered nonblank rows and 76 original-choice controls retain the approved protection guidance while obsolete title/description/lede/hint chrome has zero rendered instances. Storage removal itself was not re-assessed.	Protection explanation approved as policy; no overall after-state acceptance.
E03-09 — WD: no visible editor heading; keep accessible region name.	Retained accessible region without visible heading.	E pre-repair desktop at 05:36:55.728 UTC; Final-E at 10:40:52.473 UTC, both 2026-09-08: region label Fine-tune printable quiz, zero visible editor heading in final receipt. Earlier dangling-reference check is not a full final AX-tree pass.	No acceptance recorded.
E03-10 — WD: compact truthful quantity/kind/theme/answers/count/boundaries toolbar; remove editor undo/redo.	Later toolbar/PDF/theme source installed; fixed-10, header-metadata, estimated-boundary and Hidden-theme policies apply.	The first authorized Editor lifecycle at 01:45:08.066–01:46:58.033 UTC remained red because its oracle unconditionally required Theme although Hidden exposure intentionally suppresses it. The corrected 02:00:48.159–02:02:11.479 UTC lifecycle passes 8/8 across two viewports: 20 rendered rows, seven in-frame toolbar controls per viewport, zero overflow controls/history buttons, and trusted native text undo on both. Theme is absent and policy-correct. Compactness is measured only; kind/theme mutation and Preview/PDF behavior were not exercised.	Semantics decided; no human compactness or final after-state acceptance.
E03-11 — WD: no editor hero/dead padding; sheet immediately after toolbar.	Retained reclaimed frame; approved help and print header are intentional content.	Corrected Editor observation at 02:01:36.724–02:01:49.750 UTC measures the whole desktop/mobile frame, with toolbar then approved help then sheet, one legitimate print header, zero obsolete hero/title/description/lede chrome and zero horizontal overflow. Approved help spacing and printed header are explicitly not counted as dead chrome.	No acceptance recorded.
E03-12 — WD: same frame result, historically absorbed by E03-11.	Retained separately for identity accounting; not deleted as duplicate.	The same corrected two-viewport observation passes the whole-frame contract and retains distinct item identity across 20 rendered rows and 12 in-frame row-tool controls. Question duplication was not tested, so no duplicate-item interaction claim is made.	No acceptance recorded.

Q05 — Builder and editor review picks

Working identity / wording provenance	Installed state	Observed runtime evidence + timestamp	Human acceptance
Q05-01 — WD: readable language names with flag accents.	Conditional language controls retained; Tooltip-fixture renders installed shortcode/assets with request-local English/German availability. Stored Hidden exposure remains unchanged.	B pre-repair, 2026-09-08, exact capture time unavailable: selector count 0, only German enabled (q05-control-inventory.json, qw_exposure_settings.json; originals lost). In Tooltip-fixture’s AFTER run at 15:29:03.566–15:29:48.677 UTC, both desktop and native-touch surfaces retain visible English/German names with one SVG flag each. English→German selects backing qw_lang:de; pressed states become false/true. This is conditional-fixture interaction evidence, not opened live exposure or a complete AX-tree pass.	No acceptance recorded.
Q05-02 — WD: count slider, exact count and approximate pages.	Settings-bound decision retains current fixed 10; no bypass authorized.	B pre-repair: min/max/default 10, presets [10], no stepper/range. Final-B at 10:32:16.320 UTC, 2026-09-08: count 10; Final-E geometry has 10 rows. Process-local PDF replay’s 10–12 policy does not change live Settings.	Fixed-10 policy decided; no slider acceptance.
Q05-03 — WD: show better theme-selection ideas.	Refined B chosen by Main, installed per Theme-freeze; Hidden/one-theme mapping retained.	B pre-repair: picker absent under Hidden exposure. Theme-fixture desktop at 11:53:12.943 UTC and mobile at 11:54:31.953 UTC, 2026-09-08: six named 44px chips; mouse Dark→keyboard Retro updates selection/focus and PHP-sourced sheet tokens, no horizontal overflow. Actual installed template/CSS/JS, process-local exposure only; Hidden unchanged, no live-visitor picker claim. Separately, Footer-repair’s visual receipt begun at 15:00:04.925 UTC verifies the repaired Dark footer with Retro controls; it is not acceptance of all six theme designs or picker-to-PDF agreement.	Taste delegated; no user A/B/C choice, no final acceptance.
Q05-04 — WD: visual Page numbers on/off control with preview/PDF agreement.	Earlier empty-group suppression alone was not the requested card. Conditional controls were in v3 at 14:09:38.425 UTC; later PDF-footer color and Builder-tooltip CSS repairs are separately installed.	B pre-repair: absent footer control and empty 645×61px group. Final-B at 2026-09-08 10:32:16.320 UTC records zero empty Print details groups. Fresh Footer-repair’s AFTER caller bounds are 14:56:45.539084–14:56:55.792749 UTC; its visual receipt begun at 15:00:04.925 UTC proves 12/12 complete pages, sequential On labels in both themes, no Off labels, and Dark text contrast 15.2975–15.3267:1. Separately, Tooltip-fixture’s AFTER run at 15:29:03.566–15:29:48.677 UTC proves native footer '' → '1' → '' and miniature hidden/1/hidden on desktop/touch, with readable OFF/ON hint text. These scoped proofs do not create an end-to-end fixture-control→preview→PDF request or open stored exposure.	No human acceptance recorded; Main accepted the scoped footer and tooltip repairs.
Q05-05 — WD has no instruction; analogy to Page numbers was only proposed.	Empty-group suppression does not decide Name/Date/Class control design.	B pre-repair, 2026-09-08: fill control absent under exposure (q05-control-inventory.json). No interaction/output proof.	Intent unresolved; no acceptance.
Q05-06 — WD: improved auto-advance control.	Manual/Auto plus native seconds range installed; Manual zero and remembered Auto policy retained.	Final-B, 2026-09-08, timing actions lack separate timestamp: ArrowRight 1600ms, End 60000ms, Home 1500ms; Manual 0/disabled; remembered Auto 1600ms survives reload while Manual remains 0. Legacy 2555ms restores with step 0.001s, Manual→Auto retains 2555ms. Receipt records real keyboard/click/reload observations, not Settings saves or final mobile overlap.	Manual/Auto seconds approved; implementation not accepted.
Q05-07 — WD: icons on Builder completion choices; single-word New.	Existing icon-helper rendering and New label installed.	Final-B desktop at 10:32:16.320 UTC records Retry/New/Share SVGs. Final-mobile Builder at 10:58:49.908 UTC, 2026-09-08: all four Retry/New/Print/Share completion icons visible, 17×17px, 8px wrapper margin but measured icon-to-text gap 15.5px. Bottom Print/Play gap is 8px; no horizontal overflow. Margin and measured gap are not interchangeable.	No acceptance recorded.
Q05-08 — WD: nicer titles; date placement uncertain.	B-install normalizes en_US to an Intl-compatible locale while preserving placement; this does not decide title taste.	B pre-repair: Intl en_US RangeError and ISO fallback. Fresh-browser mobile editor at 11:41:20.348 UTC, 2026-09-08, visibly shows Surprise quiz · Sep 8. That is an observed human-readable title/date instance, not proof of every locale or a decision on catchiness/placement.	Catchiness/date placement unresolved; no acceptance.
Q05-09 — WD: one heading, no restated introduction.	Duplicate visible category heading removed; accessible section name retained.	B pre-repair: Create a quiz plus What is the quiz about; literal duplicate-string checker missed semantic duplication. Final-B at 10:32:16.320 UTC, 2026-09-08: headings Create a quiz, Questions, Shared quiz experience, Player, Completion; old category-question heading absent.	No acceptance recorded.
Q05-10 — WD: floating collapsed markers belong to click2agent, not QuizWizz.	Excluded from QuizWizz product repair, retained identity.	No QuizWizz runtime finding/verification asserted.	N/A; not product closure.
Q05-11 — WD truncated toolbar complaint; immediate truthful preview.	Later toolbar/PDF/theme source installed; fixed-10, header count, preview-only estimates, Hidden-theme and original-choice protection decisions apply.	Corrected policy-aware Editor at 2026-09-09 02:01:36.724–02:01:49.750 UTC records 10 nonblank rows, quantity matching rows, seven rendered in-frame toolbar controls per desktop/mobile viewport, zero overflow/history controls and policy-correct Theme absence under Hidden exposure. Trusted native text undo passes on both. This is truthful rendered toolbar policy evidence, not kind/theme mutation, Preview/PDF interaction, exact pagination or human compactness approval.	Semantics approved, not final acceptance.
Q05-12 — WD incomplete original wording: pool→real answers→printable output, no placeholder questions.	Later real-pool/PDF/original-choice source installed; old E-install is not final state.	E pre-repair Add produced blank stem/two choices. Final-E at 10:40:12.580 UTC, 2026-09-08: Replace HTTP 200, 10→10 rows, content changed and all nonblank. At 10:44:35.323 UTC, added A removed while originals+B preserved. Final-PDF uses 11 real unique sources and authorized 12-row duplicate in process-local 10–12 policy, not a live fixed-10 bypass.	Protect-original-choices policy approved; no final acceptance.
Q05-13 — WD incomplete original wording: usable row-attached actions below handle, never over answers.	In-flow row-action repair and Editor sticky-outcome repair retained in final source.	Historical Final-mobile at 11:02:53.112 UTC records 10 rows/70 controls, zero cross-row intersections and the Q3 trusted-touch handle targeting Q3 without count change. Pre-repair original f at 21:05:23.214–21:06:32.578 UTC remains 4/5 red because the outcome bar was outside the viewport. Corrected response-observer f at 2026-09-09 01:11:16.633–01:12:26.696 UTC passes 5/5: 10 rows, row tools opacity 0→1, and Cancel/Preview PDF/Print remain inside the 900px viewport; exactly its one draft, one artifact and one private file are removed with no outstanding cleanup. Earlier unknown capabilities remain unresolved.	Below-handle rule retained; no final acceptance.

A05 and L05 — incomplete wording, test-only captures, and retained Print

Working identity / wording provenance	Installed state	Observed runtime evidence + timestamp	Human acceptance
A05-01 — WD combined language request; original per-ID wording/mapping incomplete.	Language controls unchanged; no requirement-specific repair claimed.	S inherited pre-install, capture time unavailable: language-control facts at S-install :851-869; not proof of an unrecovered requirement.	No acceptance recorded.
A05-02 — WD same combined language request; original per-ID wording/mapping incomplete.	No separate implementation or guessed mapping.	Same inherited Settings limitation; no requirement-specific runtime test.	No acceptance recorded.
A05-03 — WD category-family consistency; original individual wording incomplete.	S-install quiet category-only border-box/weight and selected-ring treatment; existing hierarchy/focus semantics retained (S FINDINGS.md:101-107).	S inherited pre-install, capture time unavailable: 24/24 selected chips, no unselected sample; former outer height 62px (desktop-category-styles.json, mobile-category-styles.json). No post-install dimensions/selection/focus pass here.	Quiet hierarchical chip direction approved; no visual acceptance.
A05-04 — WD Science/category visual complaint; original individual wording incomplete.	Same scoped quiet-chip direction, not an invented Science-only rule.	S inherited pre-install, capture time unavailable: full Science focus images show expected 3px outline; tiny crops do not prove clipping (S FINDINGS.md:107). Hierarchy is source-derived, not guessed from names.	Direction approved; no visual acceptance.
A05-06 — WD Online Player capture, test only.	No product implementation/closure claimed.	No requirement-specific runtime acceptance applicable.	N/A; test-only capture.
A05-07 — WD SVG capture, test only.	No product implementation/closure claimed.	No requirement-specific runtime acceptance applicable.	N/A; test-only capture.
L05-03 — WD unspecified public Print pick; retain, do not change.	Retained Print identity, not a new output requirement.	B pre-repair, 2026-09-08: text/accessible name/aria-label/title Print, SVG; disabled empty/enabled selected. Print not clicked by B (desktop-occlusion-evidence.json, mobile-occlusion-evidence.json).	Retention instruction only; not PDF/editor acceptance.

Separate August canonical S03 namespace

These are the exact report titles, not quotations attributed to the user. Source: August-S03 :34-197. The August report contains 12 findings; only the eight colliding labels are listed here. None is closed by September REC or WD coverage.

August canonical identity	Original August finding title	Relationship to this report
S03-01	On mobile, the settings page has no settings above the fold	Separate from September Screen Options pick.
S03-02	Nine “workspaces” are labels for one undifferentiated long form, not an information architecture	Separate from September preview pick.
S03-03	The active section indicator lies as soon as the user scrolls	Separate from September Search-trigger pick. A current original a-d run at 20:06:27.517 UTC reproduced the inherited defect: after selecting the final Advanced workspace, settled position still marked Data current. The one-hunk installed scroll-spy repair replaces a footer-inflated document-end test with the real scrolling-element maximum; the tier-4 rerun at 20:56:29.383 UTC passes all nine visible workspace tabs, retaining Advanced as aria-current at the settled final position. This is scoped runtime acceptance of the August collision, not the September S03-03 Search identity or human acceptance.
S03-04	The page is a 12-screen mobile document with no progress cue or collapse strategy	Separate from September transient-save-feedback pick.
S03-05	The autosave model can silently lose an ordinary edit during navigation	Separate from September preview companion. August verification retains the source-backed race but withdraws unmeasured live-loss proof (verification.json:50-68); not a disproven race or new September live-loss result.
S03-06	Mobile removes the words from the save-state feedback	Separate from September Search/Escape pick.
S03-07	The legacy import deep link lands with its destination heading under the sticky toolbar	Separate from September keyboard-history pick.
S03-08	The mobile section rail hides the active destination and gives no cue that most choices are off-screen	Separate from September shelf-adjacency pick.

Explicit exclusions and locally tracked observations

• E03-06 was trashed in WD provenance; not one of the 39 retained identities, not silently renumbered.
• A05-05 is absent from WD provenance; not one of the 39, not invented from A05-04/06.
• Q05-10 is retained for accounting but excluded from QuizWizz product work. A05-06/07 are retained test-only captures, not product requirements.
• Three locally tracked PICKER observations are not original canonical IDs. No PICKER01..03 identities or mapping were recovered in WD or the three supplied audit reports. This report does not manufacture titles, map them onto A05/Q05-03, add them to the 39, or mark them accepted; a separately frozen local observation receipt is required to report their individual evidence.

PDF integrity: what was actually proved

At 2026-09-08 09:38:13.571 UTC, Main independently extracted the intermediate PDFs and checked seven file hashes. The recorded failing sample had 57 Correct answer labels; the intermediate answers-off sample had 0, with zero answer-key sections. The same receipt records answers-on with one key and 11 labels, duplicate answers-on with 12, removed answers-on with 10, and quiz-off/requested-off samples with zero keys/labels (PDF-intermediate :5-43). These are aggregate counts only; answer-key contents are not copied into the vault.

Separately, Main historically extracted the native-print before/after pair: 57→0 Correct answer labels, while the authored 57-question count label remained in both. That distinction matters: removing leaked answer annotations did not require deleting legitimate count metadata. The extraction timestamp was not supplied and is not borrowed from the intermediate canonical round. The raw pair is now lost; this aggregate observation survives only in transcript/report, not as available private PDFs or a restored receipt.

The 57→0 intermediate observation is historical, with its original files now lost; it is not final-source acceptance. It predates server-owned PDF-context correction and the latest Protect original choices policy. It does not certify every hidden-answer oracle, final theme source, arbitrary payloads, exact preview/PDF pagination, or all edited-choice/reorder/Add/removal cases. New runtime proof must identify its own frozen source and request policy rather than borrowing this round’s result.

Scoped original-choice and PDF observations

Final-PDF records 12/12 passing restricted-output checks, exit 0, at 2026-09-08 10:26:18.213613 UTC. Its process-local policy permits 10–12 questions solely for the replay; the live fixed-10 Settings value is unchanged. The scenarios edit/reorder original choices, add two blank-bound authored distractors, remove authored distractors while preserving originals, and duplicate a whole question to render 12 rows from 11 unique sources. Unknown, duplicate or missing original bindings are rejected uniformly before output is authorized. This is a scoped exercised contract, not an assertion about arbitrary untested inputs.

Independent extraction at 10:27:58.586 UTC measured three actual restricted PDFs: text lengths 2529/2460/2701, zero Correct answer labels and zero answer-key headings in each. The positive authorized-key replay also exits 0 with 12/12 passing checks; independent extraction at 10:49:39.454 UTC measures 11/11/12 answer labels and one key heading per PDF, with text lengths 3680/3611/3955. Nonempty actual-byte extraction and the positive control prevent a blank-output or always-hide-answer “pass.” No answer content is reproduced here.

Main’s historical process-only mutation deliberately made the uniform-response assertion fail: exit 1, 18 requests with 12 distinct responses rather than one. The surviving sanitized receipt records passing Settings-unchangedness and owned draft/artifact-transient cleanup assertions for that run. No live source was mutated. This shows that specific gate detected the original-choice oracle defect then, not mutation proof of every assertion or present-day state integrity. The public receipt retains allowlisted aggregates; private replay sources survive separately, but original response details/base64 PDFs/raw PDFs are lost.

The final source-only policy review (agent://ReviewEditorPolicy) reports no findings after inspecting the affected client guards again following Theme-freeze. It covers original bindings, actual rendered counts, native Craft quiz identity, and the server-owned narrowing PDF context, while preserving separately authorized Trivia HTML and quizless Forced-key behavior. It is not visual acceptance or an independent runtime replay; the actual-byte observations above are Main’s separate proof.

At 2026-09-08 10:55:33.750 UTC, Final-Print records an actual completed native Print attachment: HTTP 200, PDF magic, 143836 bytes, SHA-256 231e77f0e9d8cab2e12ec92f1de36677b9b57c62f8440cc869721fdbf99482c4, identical to the canonical preview. Extracted text contains the retained edited extra and not the removed extra; the header count is absent in this configured output. This authorized sample has 10 answer labels and one key heading, so it is not a hidden-answer sample. The multipart form was incorrectly inspected with URLSearchParams first; those null POST values are explicitly discarded, not treated as output proof.

Oracle-provenance independently records exact embedded gate-body equivalence and exact preserved old-validator body after removing only the five-byte PHP opening tag. The installed fixed service rejects the 18 attack requests with one uniform response; the intentional isolated old-validator replay produces 12 distinct responses and the expected assertion exception. This closes the code-path-equivalence limitation for that mutation result without implying that private PDFs or old source should be published.

The historical option-PDF round was not a theme/footer agreement pass. The surviving Raster-correction projection preserves a Chrome-rendered Dark/on page-1 observation captured at 2026-09-08 12:54:05.560 UTC and inspected/corrected at 13:04:56.013 UTC: black footer ink (0,0,0) against paper (23,24,29), contrast 1.1848:1; canonical SHA-256 507e94ce9dabce2d908a71a4c4ac59c636df52e773f8a9f2380d33abf24c5cd7. The original PDF and crop are now lost, not available private evidence. Main’s successful 12:41 UTC service run generated four PDFs and passed 29/29 checks, but its successful execution receipt and raw outputs do not survive; this fact is now transcript-only, not disproved or regenerated. The preserved invocation’s PREPARED_UNEXECUTED_UNLINTED_BY_EDITOR_RECOVERY describes that agent’s execution boundary, not Main’s later run. Generation success did not prove visual footer correctness. The surviving correction records twelve requested captures but zero complete pages; clipping displaced the embedded PDF view, and gray automatic footer-only crops were invalid. These historical deficiencies are preserved rather than overwritten by the following new run.

Fresh Footer-repair separately verifies the installed color correction. BEFORE outer tool-event bounds are 2026-09-08 14:33:59.667–14:34:22.147 UTC; AFTER caller-clock bounds are 14:56:45.539084–14:56:55.792749 UTC, not PHP process instants. The visual receipts begin at 14:47:07.083 and 15:00:04.925 UTC. Each run records 29/29 PHP checks and four PDFs with 10 question blocks, 10 distinct sources and 10 key entries each; Main reports independently matching all four AFTER PDF hashes. Each visual set has 12/12 painted, fully contained pages at 1280×1300/DPR1 with cache disabled. Both On PDFs show 1/3, 2/3, 3/3; both Off PDFs show no footer labels. Current pagination is questions 1–5, questions 6–10, then the key. Dark text contrast improves from the fresh BEFORE 1.1631:1 to AFTER 15.2975–15.3267:1; the decorative black separator is unchanged. Inputs match within each four-case set only: before/after were randomized differently, so no cross-run body identity is claimed. Selected-text extraction was empty and proves no retained-content identity. The historical extra-question-6 heading difference was neither reproduced nor explained. Main accepted the scoped repair; human acceptance, editor B2 print-scale intent and exposed-control agreement remain distinct.

Later shell verification and harness pin reconciliation

The fresh stored Wave 6 TAP records 1..14, zero failures, including its nested 18/18 PHP assertions. The initial fresh Wave 9 TAP records 1..21 with exactly one failure, admin_tree output identity; its query budgets and the remaining 20 assertions pass. The TAPs contain no embedded execution timestamp, so the directory name is not presented as a run clock. Sources: evidence/2026-09-08-live-punchlist/runtime-20260908T182215Z/wave-6.tap:1-37 and wave-9.tap:1-31. These are scoped shell results, not a final integrated green gate or browser acceptance.

The wrong-type Library mutation trace records a surviving fixture with the wrong post type, followed by a deletion marker for that same exact recorded ID and completed exact-ID recovery. The private ID and owner label are not reproduced here. Source: the same runtime directory’s library-wrong-type-wave9.tap:1-6. This trace is retained separately from the final residue comparison and does not turn a malformed fixture into an accepted Library result.

The four pre/post Wave 6 and Wave 9 debug captures contain identical file metadata: 7,049,129 bytes and modification time 2026-09-08 15:44:10.208650198 +0200. They establish unchanged captured size/mtime, not a log-content hash comparison, an empty log, or an execution timestamp. Sources: the same runtime directory’s debug-before-wave6.txt:1, debug-after-wave6.txt:1, debug-before-wave9.txt:1 and debug-after-wave9.txt:1.

The admin_tree mismatch was then audited statically before anything changed. The read-only causation audit recorded at 2026-09-08 18:48:00 UTC (evidence/2026-09-08-live-punchlist/tree-pin-audit-20260908/receipt.json) classifies it as inherited and pre-existing. Its cited proofs are that the stored exposure option was not modified today — current raw and semantic digests match the Sep-8 01:30:43 UTC pre-task dump across all 94 backup tuples — that the stored configuration has carried 14 enabled categories including animals:true since an operator save at 2026-09-04 17:10:04 UTC, that the 2026-08-27 remediation page pinned tree=505bf158… while only 6 categories were enabled with animals false, that none of today’s 19 installed files touch SubjectTaxonomy.php, the settings store/normalizer/schema or ExposureSettings::get(), and that bench includes/SubjectTaxonomy.php is byte-identical to mirror import 7ec58f44 and canonical 715474f. The audit marks the causal link between the September-4 expansion and the new digest value as inference, not proof, and explicitly forbids blind repinning; the identifying backup row is withheld here.

On that evidence the user authorized a harness reconciliation. scripts/quizwizz-probe.sh changed in exactly one hunk from SHA-256 4e2a635621da3875dddd94472fbf2b59d37de83b60bf150e3f3379761cedd828 to da3beb64c8ce9f7efac1399fa2fb50afd6b95e297759b73b31980ec952f41594, mode 755, replacing the frozen QW9_TREE_DIGEST value with e05e2fc52ba238853b1ae28925943ea5ced508db637f7ce548be814232459028 and adding provenance comments; serialization, assertions, query budgets and node-shape checks are byte-identical (tree-pin-reconciliation/change-receipt.json, quizwizz-probe.patch:3-16). The value remains a frozen constant, so future unexplained drift still fails. The editing agent ran nothing.

The reconciled runs are green within their own scope. wave-9-reconciled.tap:1-28 records 1..21, zero failures, including admin_tree output identity and preserved node shapes at the same 2/10/5/4-query budgets. wave-all-reconciled.tap:238-239 records 1..98, zero failures, with nested Wave 10/11 PHP assertions inside it. A separate read-only post-all capture shows the observed tree digest equal to the new expected value (admin-tree-reconciled-post-all.json), and bash -n scripts/quizwizz-probe.sh passes (quizwizz-probe-reconciled.syntax:1). No CSS linter exists on this bench, so none ran and none is claimed.

The recorded Library mutation packets are a mixed matrix, not a uniform kill list. Suppressed-style, no-cards and one-shortcode each fail exactly their targeted assertion out of five; both after-ID-failure packets fail fixture creation with published owned page unavailable; the deletion-failure and foreign-owner packets still record 5/5 assertions and instead exercise exact-ID recovery, emitting a survival marker followed by a deletion marker for that same recorded ID; the valid control records 5/5 with no failure. Each packet prints its own residue=0 cleanup line. Private fixture IDs and probe-owner labels are not reproduced here.

Post-all state comparison at 19:32:12 UTC matches its baseline on raw and stored-JSON Settings digests, 94 backup rows with an identical ordered digest, 10 operator rows, page 1284 published and page 1289 draft with unchanged modification stamps, and zero former harness users (post-reconciled-all-residue-parity.json). The companion marker check counts zero fixture-owner metadata rows (library-marker-reconciled-all-residue.json); that is a narrow marker scope, not a whole-window artifact inventory. This reconciliation is harness maintenance on an inherited red assertion, not a product repair, not an operator Settings change, and not human acceptance; releaseEligible: false is unchanged.

Final original-browser sequence: red-to-green history and bounded passes

The final cache-disabled browser sequence began on harness SHA-256 d9b111e31efe2f524c8d6e4a47c73ebbf7c7a3aaa61254164610437ac63e4773. The first original a-d run settled at 20:06:27.517 UTC with 5/7; a tier-4 diagnostic then checked 80 icon/text pairs and isolated two real 5px policy-card gaps while the scroll-spy repair made all nine workspaces pass. After the one-line gap repair, final original a-d runs 20:56:16.606–20:58:19.058 UTC and passes 7/7, overallPass:true, wrapperPass:true: nine settled workspace-current states, measured sticky-heading offsets, 80/80 icon gaps, zero console/page errors and complete scoped restoration. Final original e runs 21:01:26.582–21:02:33.169 UTC and passes 5/5 with complete restoration. Neither result grants human acceptance.

Pre-repair original f runs 21:05:23.214–21:06:32.578 UTC and remains 4/5 red: row-tool opacity works, but the outcome bar sits outside the viewport. Exact recovery later proves only its one captured draft identity already absent; the unknown second watched creation remains unresolved. The first post-sticky f rerun at 21:59:02.380–22:00:54.029 UTC proves the product geometry—10 rows, row tools opacity 0→1, and Cancel/Preview PDF/Print at viewport coordinates 804.96875–884—but its lifecycle remains 5/6 red because response capture had not settled before close.

Corrected response-observer f (standalone-gate-worker/private-final-browser-two-f-response-20260909T010816Z.json) runs 2026-09-09 01:11:16.633–01:12:26.696 UTC and passes 5/5 with original, wrapper and overall pass, zero console/page errors and the same in-viewport product geometry. Its two witnessed successful responses yield exactly one draft and one artifact; cleanup removes those two capabilities plus one private file, with zero errors, outstanding items or remaining owned objects. Exact session revocation, baseline parity and helper absence also pass. This is a scoped current-green transaction, not cleanup authority for the earlier unknown creation.

The response-observer synthetic proof (fetch-response-proof/run-b1deb7cf-7a92-40f2-9867-98d25338cf74/receipt.json) passes 24/24 and kills four mutants: acknowledgement before persistence, first-route instead of exact-header matching, trust in a caller-forged header, and ignoring an unwitnessed canceled-200 response. It proves the offline observer contract, not live cleanup by itself.

The first g response-observer lifecycle records the requested product behavior passing—pool-short Replace retries once without exclusions and preserves every row—but stays lifecycle-red because Replace invalidated the still-pending initial preview before artifact capture. Exact containment removed only the known captured draft; no unknown-preview sweep ran. Corrected g (standalone-gate-worker/private-final-browser-two-g-repair-20260909T013431Z.json) awaits the initial successful preview capture before Replace, runs 01:37:30.133–01:38:32.161 UTC and passes 5/5 with original, wrapper and overall pass, zero console/page errors and exact cleanup of one draft, one artifact and one private file. This fixes harness barrier ordering, not product behavior.

Fresh-Surfaces separately runs 00:30:52.886–00:33:08.035 UTC and passes nine wrapper assertions for S03-02/03/05/06 on cache-disabled desktop/mobile, with singular DOM/AX Search and Preview openers, trusted Enter/Escape focus return and trusted Search filter/clear/collapse. It does not execute original a-d or substitute for f/g. Cleanup is bounded to its owned Chrome, exact session and preserved Settings/backups/protected pages/operator sessions.

The first fresh Editor lifecycle runs 01:45:08.066–01:46:58.033 UTC and remains red because the hook required Theme unconditionally while the product intentionally hides Theme unless visitor exposure is enabled and more than one theme is enabled. E03-08/11/12 already passed there; E03-10 failed only that policy-blind oracle. A tight offline correction passes 68 checks with zero failures; its execution timestamp was not exposed, so 01:57:44.730 UTC is persistence time only.

The corrected policy-aware Editor lifecycle runs 02:00:48.159–02:02:11.479 UTC on harness SHA-256 25a12d994470163bcafb723a3cc2f82b27dc48d45e9270bf25bf38313ed0b30a and hook SHA-256 4d14fbda5321e32112df32eeb6824e13ec08ed2c1b6b21bac2998c4a6f4e719f. It passes 8/8 with original assertions explicitly not executed, zero console/page errors and two cache-disabled viewports. Denominators are 20 rendered rows, 76 original-choice controls, 100 rendered buttons, 154 AX buttons, 12 in-frame row tools and two safe frames. E03-08/10/11/12 pass within their explicit limits; E03-01 remains unresolved. Exact cleanup removes one draft, one artifact and one private file; baseline/session/helper handback is green. This separate manual lifecycle does not replace original f or create human acceptance.

No further state-mutating runtime was run. Earlier red receipts remain authoritative for their own lifecycles; later green receipts do not rewrite them. Unknown earlier request/capability identities remain unresolved, and no global cleanup is claimed.

Current verification frontier

The integrated evidence is now settled for this local report. Original a-d passes 7/7 after the scroll-spy and policy-gap repairs; original e passes 5/5. Corrected response-observer f and corrected g each pass 5/5 with exact current-run capability cleanup. Fresh-Surfaces passes 9/9 for S03-02/03/05/06. The corrected separate Editor lifecycle passes 8/8 for E03-08/10/11/12 within its stated limits. The earlier f, g and Editor red lifecycles remain preserved as red-to-green evidence, and earlier unknown capability identities remain unresolved.

The final plugin packet remains a 19-path dirty-baseline artifact, not canonical or release eligible. The sealed after-only harness/worker supplement has manifest SHA-256 a7d8d50ba93dfe807070c435bc669b46dac6ebe53a7616321c63a8f56fbe6eb4 and proof-command SHA-256 7f18fe729965c13217b60ac10ebdb750a9aa47a47ea5701164b0f5e90b3f2e32; its 39 files have 39/39 copy/current parity. It preserves final f/g/Editor worker source but deliberately excludes fixtures, raw results, capabilities, config, PDF copies and answer content. It is source preservation, not runtime replay, historical baseline, canonical patch, mirror commit or publication.

The public manifest now inventories 50 local assets—28 JSON and 22 PNG—with exact file/manifest agreement, all modes 0644, all JSON parsed and no stored-byte or decoded-RGBA PNG duplicates. The two public Editor frames are paint-masked viewport geometry evidence, not proof of underlying content deletion or human acceptance. No production publication occurred.

Remaining decisions are limited to matters the evidence cannot decide: E03-01 whether the condensed answer-badge “print scale” intent requires actual-PDF parity or editor-scale legibility; Q05-05 intended Name/Date/Class treatment; Q05-08 title/date taste; unrecovered A05 individual wording/language-control intent; and item-specific human acceptance. Screen Options scrolling beneath the sticky header at shallow scroll, matched same-depth mobile-before evidence, all-theme taste, question duplication in the corrected Editor lifecycle and end-to-end card/preview/PDF agreement also remain unproved. Fixed-10 policy, Manual/Auto seconds, quiet chips, estimated boundaries, header-count semantics, delegated theme taste and Protect original choices remain decided.

No formatter, linter, test suite, production wiki build, mirror push or publication was run by this report author. Main ran the sole final /home/loca/dev/wikis/build.sh wrdp --check-only validation on 2026-09-09: exit 0, wall 166.80 seconds, SSOT passed, 89 Markdown files parsed, 945 files emitted to scratch, check-only completed successfully and only the expected no-git warning occurred. The execution timestamp was not supplied; the private evidence/2026-09-08-live-punchlist/final-wiki-check-only.json records that boundary. No production publication followed. releaseEligible: false, humanAccepted: false and productionPublished: false remain unchanged.