Verification — 2026-10-09

Source PR8 merged at b47f7a34706fa8c92a350f8c5bcbf0d875e189f0. Manual release preparation run37954427674 passed on that exact commit. Its three main assets are published at https://github.com/codepetca/zero/releases/tag/v0.5.1. Core0.1.1 and existing community Maven coordinates remain unchanged; optional community archives remain local. Zero0.5.0's three public downloads were retrieved and verified against their original sizes/hashes before publication.

Local90/90 Node tests passed. Linux PR CI and manual CI both pass90 tests, 3 release-boundary tests, all7 finite JavaFX examples under Xvfb, website5tests, typecheck/production build, packaging, canonical source/license parity, ZIP permissions, extracted Maven build and sealed exact asset hashes. The first Linux run exposed an existing fixed-event-loop test wait; bounded observable UI waits corrected it without changing production code or weakening assertions.

Final candidate archives have the same complete member contents as the local macOS-tested archives; ZIP container timestamps differ. Final MIT licenses, core/extension source and trusted examples are byte-identical. All3 public assets were subsequently downloaded without authentication and matched the manual-run candidate sizes and SHA256. The tag resolves to the reviewed source merge commit.

macOS extracted VSIX installed with isolated user/extension directories. Actual sidebar Run App launched Maven and its Main JavaFX child; actual Stop returned "App stopped", disabled Stop and terminated both processes. Native automation could not select the separate Java process window, so no manual app interaction is claimed. macOS ditto extraction preserves the ZIP's executable Maven wrapper; the initial Python QA extraction did not preserve modes and was unsuitable for that editor test. No release asset permission correction was needed.

Extracted Quiz passed two finite native behavior/lifecycle runs; real extracted extension plans added HealthBar0.1.0, updated0.1.1 and reverted0.1.0 through Maven and native JavaFX. Catalog MIT support, immutable JAR hashes/origins, fractional fill behavior and preserved student source passed. These use the local community repository and do not establish a publicly hosted community artifact repository.

Independent Sol/medium review covered the release/automation delta and the subsequent sidebar-test fix, with no blockers. Physical Windows/Linux installation, student GitHub authentication/upload, full profile import and novice classroom trials remain untested. Ignored .verification/release-0.5.1-* receipts retain the candidate, public downloads, CI logs and macOS editor/consumer evidence.

Student API reference, separate local community library, native Component Workshop and on-demand Browse/Try/Add/Update/Revert commands are implemented. Maven owns resolution; managed POM edits preserve student Java and unrelated configuration. HealthBar0.1.0's partial-fill bug and compatible0.1.1 fix remain immutable fixtures. Neither is a public/community-reviewed release.

Full Node85/85 checks passed (129.62s); the subsequent task ownership/cleanup delta passes19 affected checks, including one new Stop-during-launch integration test. Current coverage87 tests after review fixes;20 latest affected checks pass. The separate community admission suite now passes25 tests. Two consumer apps × install/update/revert passed six actual finite JavaFX stages. A coordinator consumer built from the same extension plans passed three additional Maven/JavaFX stages with source preservation and actual JAR hashes/origins.

Workshop checks render fresh named states, change controls/reset, compile a trusted local candidate explicitly, reject a fractional-fill mutant and invalidate stale source-bound receipts. Its Java ZIP passes independent Python validation. The extracted portable component kit passes with default sibling paths in a folder containing spaces. Maven plugin argument splitting initially broke that launch; relative defaults and file-URI custom paths corrected it.

The0.5.0 VSIX installs through isolated VS Code CLI; all extension source and packaged trusted example assets match local bytes. This proves installation, not native command interaction. Generated receipts stay ignored in .verification/ and the separate community .proof/. Final independent review/artifact acceptance is recorded in CURRENT:5 review turns,2 fix batches,6 P2s resolved, no remaining findings. Final69-member component ZIP/source/library hashes and installed VSIX bytes match the locally committed code. Whitespace/property Java packet independently passes Python validation; a stale initial fixture ZIP was discarded.

Physical editor clicks, chooser interaction, Windows/Linux, live CI/AI, authenticated community acceptance, public hosting/licensing and classroom trials remain unverified. Candidate Java executes with normal local permissions, not a sandbox. Offline AI feedback is advisory; no tool can approve or publish. No remote repo, push, publication, deployment or credential changes occurred.

Local implementation on codex/individual-workflow from 528ce8f. The owner agreed Start a change → Upload changes → Finish change into main, then requested this goal's orchestration. No team/PR mode or rebase interface. Initial main upload uses the existing reviewed path; Finish and optional local cleanup add guarded state transitions. Authentication/storage and local Java framework are retained. This goal authorizes local implementation and injected-remote dogfooding, not a real GitHub upload, account modification, source push/PR/merge or release.

Checks:

• Focused sidebar tests passed 3/3: default simulation skips save/auth/Git mutation, cancellation, busy menus/commands, account and branch drift, update pause for Run App/fresh Finish, preserved conflict guidance and native Source Control/fallback, success-only optional cleanup, escaping and token exclusion.
• Coordinator's standalone starter trial used a temporary path with spaces and injected all 34 transport commands to a local bare repository. Two small Java changes compiled with the pinned Maven wrapper; two complete branch/finish cycles confirmed main and exact Java source bytes. Preparing then cancelling Start/Finish kept refs/files unchanged. A failed final push retained the feature and source; a new Finish review uploaded the preserved SHA. Optional cleanup removed only the local feature; the remote progress branch remained. Reusing that remote name was rejected. Final local-bare main was 7db23efdfcd76e8b41a970d495404848e6d082a8. Temporary trial repositories removed.
• Full exact implementation suite passed 56/56 (147.121s at 74bc2c5). Real local Git tests cover newer main fast-forward, disjoint merge/update pause, conflicts and resolution/retry, remote non-fast-forward races, forged/stale plans, local/remote/root/file drift, worktree/unrelated-history refusal, ignored-file preservation, duplicate remote feature names and atomic cleanup against a concurrently moved branch. Native-session transport tests cover the narrowly permitted pinned fetch, destination/session guards and credential isolation.
• Initial independent security review completed clean; state/UI review found one P2: native checkout left the displayed current branch stale until clicked. Coordinator added an exact-project Git HEAD/config watcher covering change, create and delete. The affected 3/3 UI checks passed (0.190s), including refresh without a Zero command and ignoring another project's metadata. Targeted review accepted that delta at ae1514c with no new findings. Unchanged engine/transport coverage is reused.
• Configuration, syntax, command contracts, document links and diff checks passed.

Review receipt: two independent GPT-6.1 Sol/high initial reviewers completed the full diff from 528ce8ff0206f9de21ec111843be89c9a242edd6 to 74bc2c58b45e8765f9fd2a72c5bc34788ff56434; security clean (~120s), Git-state/UI one accepted P2 (~3min). The state/UI reviewer accepted the targeted correction at ae1514cf50457860c7e27fa5121f4dc7a1e8c0ee (~1min). Three reviewer turns, one initial wave, one targeted wave and one fix batch, about 6.5 minutes elapsed including coordination; effective model and attributable token telemetry unknown. No final integration wave adds coverage: the correction is read-only display refresh and retains workflow/security contracts. Coordinator inspected the final evidence-only delta. Final 0.4.0 kit regenerated; extension source/manifest/README, every starter file, profile/optional bindings and all seven combined-kit members match local bytes. The Unix Maven wrapper is executable and no target/Git folder is included. Source publication remains unauthorized.

Local-bare transport is injected only in checks; production accepts the ordinary reviewed GitHub HTTPS destination and uses guarded native-session credentials. No real account/network trial is claimed. Final 0.4 VSIX installed through the isolated VS Code CLI (the first nonexistent-profile attempt failed; default isolated install succeeded). Computer-use selected a different Code process's Welcome window, so this does not confirm new-version activation or interaction. Installed JS bytes match the reviewed source after the final installation. Native 0.4 editor interaction and real GitHub start/finish, physical Windows/Linux, school restrictions and novice student flow remain untested. The Java framework is unchanged; these Java edits were compile checks, not physical app input or full framework GUI evidence.

Local change on codex/minimal-github-sidebar, based on the completed study branch 0cc62b5. The owner requested an account icon/initial with hover name and account actions, a connected repo name and create/connect options behind “Connect a repo”. The existing native authentication, upload review and transport are retained. Only a read-only account-label getter was added to the authentication bridge.

Checks:

• Focused UI/authentication tests passed 6/6. Account/repository menus support cancellation, signed-out local connect, sign-in before create, account change, delegation to VS Code Accounts for sign-out and copy-link when connected. Existing busy guards also block new menu commands during upload review. Account names are escaped and native tokens do not appear in the HTML. Connected repository name remains visible after session removal.
• The simulation regression now waits for its warning to be issued before checking command release, removing an IO scheduling race exposed by the added menu checks. Production still does not await warning dismissal.
• Full Node suite: 37/37 passed (34.66s initially, 28.41s after correcting the native account-command route). Configuration, syntax, document links and diff checks passed. Java source/dependencies are unchanged, so no Java GUI re-run was needed for this sidebar change.
• Packaged 0.3.2 VSIX installed into the isolated Zero profile. Actual Mac reload showed the signed-in account's initial, accessible account name and compact empty-repo row. The real repository quick pick displayed create/existing choices. Concurrent user interaction interrupted further menu automation; account actions are covered by mocks so far, not a completed native logout or account switch. A native sign-in permission prompt encountered during the interrupted trial was cancelled; no authorization was granted. No account credentials, repo connection or files were changed by the menu trial. The corrected VSIX was installed again; final activation/menu trial remains a gap.

Source inspection caught that workbench.actions.accounts is a UI action rather than a registered command. The corrected route detects the available public workbench.action.manageAccounts command before opening its native account picker. On older editors without it, Sign out gives manual native Accounts instructions. Both paths pass focused tests. No internal session-deletion command is called.

One standard-risk GPT-6.1 Sol/medium independent review completed clean at 1534dd27b65f0f34088f85f5a2f9af6125066b1b against 0cc62b5b719c474cebb4a7217b0fa14bd28fd1ae: one turn/initial wave, zero remediation, about one minute estimated; token telemetry unavailable. Reviewer checked rendering, escaping/accessibility, cancellation/busy guards, session consistency, command compatibility, tests and guides, and independently confirmed the native Manage Accounts registration. Coordinator inspected the later evidence-only delta. Final ZIP/VSIX regenerated and byte-checked against the extension sources, manifest/README, every starter file and combined-kit members. The Unix Maven wrapper is executable; no build/Git directories are packaged.

This local request does not authorize push/PR/merge or release publication. Windows/Linux and signed-out physical UI remain untested. Native sign-out uses VS Code Accounts: the user selects the GitHub account and Sign Out there; Zero does not directly delete a native session or display a custom credential form.

After the 0.3.1 fix merged in PR #3, the owner authorized building the study app and explicitly skipped novice trials for this phase. Local implementation on codex/study-app-dogfood starts from merged main 52cd3be. The new example uses the existing SimpleApp unchanged: three Question objects in an array, retries, once-only scoring, Enter/check, Next, explanations, final summary and Restart. The shared caption constructor keeps the quiz's no-argument ScoreDisplay working; the practice tracker uses “Completed”. No dependency, extension or framework source changed, and no further upload or Pika action occurred.

Checks on the same Mac/JDK environment:

• New study finite real-window checks passed (25.80s): blank/wrong retries, trimmed/case-insensitive answers, Enter, duplicate scoring, locked Next, advance resets, mid-session restart, summary, finished-state guards and restart/rescoring. Quiz and practice checks passed with the shared constructor changes; animation, counter and drawing also passed. Shared/default component bytes match and the default quiz is unchanged.
• The normal seven-example run stopped at the existing keyboard assertion “Right arrow must move object”. Diagnostic state showed both window/canvas unfocused; the existing framework deliberately ignores unfocused key input. One worker diagnostic forced native window activation using unsupported UI automation; that pass is excluded from acceptance. No test/framework bypass or input-source change was made. This run does not establish a clean full suite or physical keyboard interaction; the focus-dependent keyboard check remains an environment verification gap.
• Extracted the packaged starter into a standalone path with spaces, copied study's exact three files, and added a fourth Question without changing the handlers. Full study behavior checks passed against four-question progress, scoring, completion and restart (10.046s Maven time). Restored the three-question source and removed the contributor harness from the ready-to-run local copy.
• Configuration, syntax, documentation links and diff checks passed. The full 37-test Node evidence from the fix is reused because extension code/tests are unchanged. Local packaging passed. In the isolated Zero profile, the standalone study folder showed Main, Question and ScoreDisplay in the minimal file tree; sidebar Run launched Maven and the JavaFX Main process. Sidebar Stop showed “App stopped” and both owned processes ended. This is editor/process evidence, not direct JavaFX typing/clicking.

One GPT-6.1 Sol/medium builder delivered the app, shared caption and checks in about seven minutes including focus investigation. Coordinator inspected the source and independently tried the packaged fourth-question exercise. One writer per component, no integration conflict. Token telemetry and attributable coordination time are unknown. One standard-risk GPT-6.1 Sol/medium independent review completed clean at 3f0d0c71e3b1f1471d25513f10e1d86e5c857754 against 52cd3be7ea2b4fcfa2ea8d84e21a83cd7107c67a, covering session transitions, compatibility, copy reuse, instructions and checks. The unchanged keyboard-focus gap does not block this SimpleApp change. One reviewer turn/initial wave, no remediation, about three minutes; token telemetry unavailable. Reviewer noticed the preliminary starter README needed repackaging. Coordinator inspected the subsequent evidence-only delta and regenerated the final kit, matching every starter file and guide/VSIX byte; no build/Git directories were included.

The starter still opens the beginner one-question quiz. Study remains an optional example and an extracted local project. Windows/Linux, school setup and direct JavaFX typing/clicking remain unverified; novice testing was skipped by the owner. Study changes are local only; no PR/push/release publication is authorized for them.

The user authorized a real teacher workflow on 2026-10-08 and completed native GitHub permission/browser sign-in. Environment: macOS 26.6.2 arm64, VS Code 1.141.0, OpenJDK/javac 17.0.14, Git 2.53.0, Red Hat Java 1.56.0 and Java Debugger 0.59.0. Started from packaged 0.3.0 in a fresh standalone path with spaces and isolated VS Code user data/extensions. Imported the actual settings-only profile through Profiles UI and activated Zero; installed VSIX and Java extensions by CLI.

Observed passes:

• Minimal light sidebar, Java ready, Run while signed out, standard Cmd+Shift+B, Stop and compiler Problems navigation. Autosave persisted a UI edit without Save. VS Code continued a comment onto an import, causing a real missing-symbol error; double-clicking Problems navigated to its line/column. Coordinator repaired the source through file tools after unreliable repair-key automation and changed the question to 5+5/answer 10/5 points. Rebuild contained the modified question, no Problems, exactly one owned app process; Stop ended it.
• Cancel native sign-in, useful retry text and local controls retained. User then completed sign-in; Zero displayed the matching account. Create Repository opened official GitHub in the OS-default Chrome profile. Created a private empty teacher repository, connected from the standalone folder and set local test commit identity using a noreply email. No global identity/config changes.
• Simulation left Git status unchanged, with no index/commit. Live review displayed the exact account/repository, main and 29 paths. Cancel left no commit/staging and remote empty. Repeated Commit & Upload succeeded through Zero's native session transport. GitHub remote SHA matched local commit, and all 29 file blob SHAs matched; target/.git stayed excluded. No credentials persisted in Git config.
• Copy Repository Link pasted the correct page URL without .git into Chrome. The GitHub page showed the actual commit and uploaded files.

Found a reproducible UI issue in stock 0.3.0: awaiting the simulation warning kept Upload/account controls disabled until its notification was dismissed. Clearing notifications released them. Version 0.3.1 issues this informational warning without awaiting dismissal; upload review/confirmation still waits. The strengthened deferred-notification regression failed against old code and passes after this correction. Full Node suite: 37 tests passed (22.66 seconds); targeted UI tests, configuration/document-link checks and diff check passed. The packaged 0.3.1 VSIX installed into the isolated Zero profile. After Reload Window, a real simulation left Upload and Change GitHub Account enabled while the warning remained visible. Git status was unchanged and no new commit was created; restored the pilot's live setting and verified its clean worktree.

One independent GPT-6.1 Sol/medium reviewer completed the bounded correction and guide review with no actionable findings at 00901afba57e8a3cbe88df381e21824fac403f96 against merged MVP 3be9d5295b4f21b43d2b2b7077e40c612bf1403c, independently running the deferred-notification regression. Approximately three minutes, one review turn, no review remediation; attributable token telemetry is unavailable. The coordinator inspected the subsequent evidence-only documentation delta. Regenerated 0.3.1 VSIX, starter ZIP and combined kit locally; byte checks match the extension, starter and packaged guides, with no build/Git directories.

Limits: direct interaction with the unbundled JavaFX quiz window was unavailable through computer-use tooling; editor/process checks are not physical quiz input checks. Native permission cancellation was tried, not denied browser OAuth or account switching during a real upload. Pika submission awaits a chosen classroom/ assignment; no submission, student impersonation or grades changed. Windows/Linux, school proxies/cold caches and novice students remain untested. The private pilot repository is retained as evidence. The owner subsequently authorized the fix PR and merge: PR #3 merged to main at 52cd3be7ea2b4fcfa2ea8d84e21a83cd7107c67a; the local main checkout was synced. Its final head reused the unchanged code review and coordinator inspection of later documentation-only changes. No outstanding checks/reviews/threads or merge conflicts were reported. No release or Marketplace publication occurred.

This local implementation separates event-driven SimpleApp from animated SketchApp, starts with a quiz, and includes six examples plus a reusable ScoreDisplay and student exercises. The extension adds native GitHub session onboarding, website repository creation and transient authenticated Git transport. No real authentication, GitHub upload trial, publication or account change was performed for this milestone. The previous publication is recorded below.

Checks performed before independent review:

• npm run check: configuration, syntax, command/task contracts and local documentation links passed.
• npm test: initially 30 tests passed (~15 seconds); after the first batch, 32 passed (33.1 seconds), and after the second, 36 passed (40.8 seconds under concurrent build load). After the final ordering correction, all 37 passed (14.12 seconds). New coverage includes silent/native sign-in cancellation, session/account changes, sidebar states, upload-session capture/revalidation, destination/rewrite guards, transient credential config, hook cleanup, trace suppression and sanitized transport failures. All sessions and network operations were mocked/intercepted; local disposable Git staging and commits are real.
• Clean Java compile passed on Mac arm64/JDK 17. Finite GUI smoke passed (3.784 seconds Maven time): shared startup, actual layout resize, canvas pixels, canvas/control focus, held-input release, useful configuration/setup failures, one-error animation shutdown and cleanup.
• Six finite real GUI example checks passed (21.95 seconds): quiz, practice, animation, keyboard, mixed-control counter and drawing. Canonical shared ScoreDisplay and default quiz copies match. GUI input was synthetic.
• Coordinator tried exercises in disposable copies: changed quiz question, accepted answer and points; added caption constructor retaining the default; used “Completed” in the tracker while quiz retained “Score”; changed Player to immediate movement. Three finite behavior checks passed in 9.55 seconds. These checks verify the exercise instructions, not novice comprehension.

Two independent GPT-6.1 Sol/high reviewers completed the initial change set at 353c3cb4082aa8b406f479de4dcc50f5bfd70993. The credential review found two P1 gaps: persistent Git Trace2 config could record runtime credentials, and a session removed during asynchronous preparation could still dispatch credentials. Both were fixed in one batch: explicit trace-disable environment values and final native-session validation immediately before dispatch. Actual local Git probes with fake credentials test trace files; removal/account-switch probes at three preparation boundaries require zero dispatch and temporary-directory cleanup. Targeted auth/transport tests passed 10/10 (~4 seconds).

The framework review found two nonblocking corrections, also accepted in this batch: historical credential guidance now has explicit historical scope; failed animation calls outer app shutdown to clear held inputs. Updated GUI smoke passed (2.506 seconds Maven time). A disposable mutant restoring the old stop call failed the new assertion, confirming it detects the cleanup bug.

Initial local packaging/source-byte checks passed. Scripted extraction into a path with spaces checked and honored Unix wrapper permission metadata and Windows wrapper CRLF; extracted GUI smoke passed (4.402 seconds Maven time). The local 0.3.0 VSIX installed through VS Code's CLI into an isolated extensions directory. The running new editor reported VS Code 1.141.0. Computer-use tooling selected the older editor process, so new-version physical sidebar/keyboard interaction has not been verified; earlier 0.2.0 interactive evidence below remains historical. The first targeted review confirmed those fixes but found a new interaction: the last asynchronous session lookup followed the URL-rewrite check. A rewrite introduced during that lookup could change Git's effective destination. The second batch performs a fresh asynchronous lookup, finishes Git/temporary-folder checks, then requires a synchronous captured-ticket check immediately before dispatch. Omitted/asynchronous checks fail closed. The transport requires Git 2.31+ for environment-based runtime configuration, rejecting old/unknown versions before its credential lookup. Targeted auth/transport/UI tests passed 16/16 (4.3 seconds), including this rewrite interval and unsupported Git versions. See Git's runtime configuration documentation and Git 2.31 release notes. The second targeted review identified one remaining preparation interval: temporary hook-directory creation still followed the destination checks. The coordinator reproduced it in a failing regression, moved directory creation before the final root/rewrite checks, and placed those checks inside cleanup. The new regression and full 37-test suite pass. This is the third fix batch. The third targeted review accepted the final correction. A cumulative integration review completed clean at d0127761507500abddc4e4ea680d66eef7770f62 against base fdacdd46eb560fd8ef36e29027a0b116cb35b32e, reusing the completed checks. No remaining actionable blocker was found. Review covered the shared native-session, transport and upload contract, simulation isolation, Run/Stop compatibility, framework cleanup and documentation/package contracts. Six reviewer turns, one initial wave, three targeted waves, three fix batches and one final integration pass took approximately 25 minutes; attributable token telemetry is unavailable.

Final local artifacts were regenerated after the review fixes: Zero 0.3.0 VSIX, zero-starter.zip and combined zero-bootstrap.zip. Byte checks matched every starter source file, current extension module and combined-kit member; no build or Git directories were packaged. Extraction into Final MVP with spaces retained Maven wrapper executable metadata and Windows wrapper CRLF. The extracted starter's finite GUI smoke passed (2.553 seconds Maven time). The final VSIX installed successfully through the CLI into the isolated test extensions directory. Final configuration/document-link and diff checks passed. This CLI installation is not evidence of physical new-version sidebar or authentication interaction.

Physical Windows/Linux, actual authentication/network upload and student pilots remain unverified. Use CLASSROOM-PILOT.md before adoption. The original Zero code's public license remains undecided.

Zero 0.2.0 is a locally packaged prototype. Initial verification ran before any kit publication. The owner subsequently authorized creating the public codepetca/zero repository and publishing through a PR into main. Disposable upload-engine test repositories used test-only identities and intercepted every transport; these tests did not upload student code. No account credentials changed. The earlier Processing template and its teacher verification record were inspected and preserved before bootstrapping Zero.

Environment: macOS 26.6.2 arm64, OpenJDK 17.0.14, Node.js 22.21.1, Git 2.53.0 and VS Code 1.138.0. Pinned JavaFX 21.0.12, Maven 3.9.11, Maven Wrapper 3.3.4, compiler plugin 3.14.0 and clean plugin 3.2.0.

Automated checks passed

• npm run check: source syntax, JSON/exported profile envelope, command/task contracts, matcher references and source documentation links.
• npm test: 21 passing tests. Covers project/root and URL guards; simulation; queued Run/Stop and configured task saving; folder-scoped upload settings; upload cancellation and Connect/upload serialization; real disposable Git staging/commits with intercepted upload/confirmation; source drift, conflicts, nested repositories, ignored staging, renames/deletions, identity failures and sanitized errors. Upload regressions preserve leading-space paths and bind the reviewed commit SHA and HTTPS destination. Git URL rewriting is refused.
• python3 scripts/verify-examples.py: three isolated starter copies compiled and opened finite real JavaFX windows. Keyboard checks key movement, bounds and pixels; counter checks native buttons/label; drawing checks painting and clear. All printed EXAMPLE_OK and exited successfully.
• ./mvnw -B -Psmoke clean compile javafx:run: finite real window checks lifecycle order, elapsed time, canvas pixels, held keys and a native control, prints ZERO_SMOKE_OK, then exits. Packaged starter extraction into a path with spaces also passed this check; wrapper executable permission survived extraction.
• Local VSIX packaging/installation passed. Starter ZIP source bytes, excluded build/Git directories and combined-kit members were checked. No Marketplace publication occurred. Dependency audit reported no vulnerabilities.

Interactive Mac evidence

A settings-only profile was actually imported through the VS Code Profiles UI. The export envelope was corrected after an initial import failure. Zero VSIX, Red Hat Java and Debugger for Java were installed in that test profile. A standalone starter copy outside the kit repository was opened in a path containing spaces.

The actual light sidebar, ordinary Main/Player editor, simple files tree and Java language support appeared. Run App launched a Maven task and a separate JavaFX process. Stop terminated both. Cmd+Shift+B launched the configured default task, and the sidebar Stop button owned and terminated that task too.

A deliberate Java syntax error produced a Maven error in Problems. Double-clicking it navigated to the correct source line and column (line 8, column 22). The disposable source was restored. Testing exposed Java language-server output in Maven's class folder; Run now uses clean compile javafx:run, and automatic Java background builds are disabled. This prevents a stale editor-generated class from masking a compiler error. Extra editor buttons, inlay hints and empty AI panels are hidden by the profile/extension settings.

The Mac initially locked during final verification. After access returned, a fresh standalone copy in a path with spaces passed Cmd+Shift+B launch and repeated shortcut restart. The first Maven/app processes ended and one replacement pair remained. Sidebar Stop then terminated both replacement processes. An editor comment persisted to disk without a Save command, verifying autosave; the test source was restored. The actual dark-theme sidebar/editor screenshot is saved at actual-zero-editor.png. Both light-profile import and a user-selected dark theme have now been observed in the real editor.

Independent review and remediation

Two GPT-6.1 high-reasoning reviewers covered the upload engine and IDE/configuration in one initial wave. Three P1 and two P2 findings were accepted: trimmed NUL filenames, mutable HEAD/destination transport, provider-returned tasks bypassing the resolver, and missing resource URI for upload settings. All were corrected in one fix batch. One targeted high-reasoning follow-up reviewed the combined delta and accepted it; four added engine regressions plus two delayed URL-rewrite probes passed. No remaining concrete blocker was found in that delta. The later native repeated-shortcut check also passed, as described above.

Review ledger: 3 launches, maximum 2 concurrent reviewers, 1 remediation wave, 1 fix batch. Reviewed engine SHA-256: 7a936e68d2464bb13fe3e7f9bee108e1002a50dbfdb80e14aeb5d5997e56fc23. Core SHA-256: 37844050205f033a9167bdc53f31cbec5d40355dfd59086479c99f6c02716682. The later IDE-only delta closes an empty AI sidebar on activation in the quiet profile; it does not change upload behavior.

Historical 0.2.0 limits and next platform checks

The following are the earlier version's observations and proposed checks. The 0.3.0 native-session bridge replaces its credential-helper requirement; use the current classroom pilot checklist for new trials.

Physical Windows/Linux, school machine restrictions/proxies, fresh dependency caches, GitHub authentication/network upload and Pika submission are unverified. Tests of intercepted Git transport are not evidence of a real upload. The installed Git subprocess needs a working credential helper; VS Code UI sign-in alone may not supply one. Public licensing has not been chosen (UNLICENSED prototype). An earlier passing Mac smoke emitted CVDisplayLink warnings; all display hardware support is not established.

Before classroom adoption, record OS/architecture, JDK and VS Code versions, then:

1. Import the packaged profile, install VSIX plus the two Java extensions, and open an extracted standalone starter in a path with spaces.
2. Run, change a visible value, Run again while active, and verify one app remains. Repeat through Ctrl/Cmd+Shift+B and Stop; verify build and app processes end.
3. Add a temporary syntax error, rebuild, open its Problems location, then restore. Observe autosave and try each example. Run the finite smoke check.
4. With separate explicit authorization, create a student-owned empty GitHub repository; connect, configure identity/credentials and enable live mode. Cancel one upload review, then confirm a real upload and inspect remote files. Check failure/retry guidance and paste the repository link separately in Pika.

Coordination receipt

Current goal started 2026-10-08 01:01 UTC; account weekly remaining was 79%. DeepSeek was paused through 2026-12-31 and was not used. Native GPT-6.1 medium workers owned Git engine/tests, Java examples and guides separately. The Java work completed in about three minutes; example checks initially took about 20 seconds. Git work needed a rename/deletion correction and the review safeguard batch. Coordinator verified deliveries; worker token telemetry was unavailable. Initial reviewers took approximately two and six minutes; targeted review about one minute. One writer per component and small context handoffs limited coordination overhead. The existing chat remained coordinator; no separate chat or automation was created. The publication request reused these completed reviews and checks; its PR records the final focused integration review and merge decision.

Local framework source is canonical under framework/src; generated readable starter copies are drift guarded. Original Java API and harness bytes match. Flat kit verification checks source parity, receipts, VSIX bytes, wrapper modes, absence of contributor tools and an actual extracted Maven compile in a path with spaces. Full Node suite89/89 passed139.128s. Seven finite JavaFX example checks passed24.15s; extracted69-member Component Workshop native check passes with core0.1.1. Prior immutable core0.1.0 artifacts remain unchanged.

The unchanged framework SmokeLauncher failed its canvas-focus assertion during background native launch. No assertion was weakened; full smoke success is not claimed. Physical Windows/Linux and final native editor interaction remain untested.

Next.js typecheck/build, four download/publication boundary tests and dependency audit pass. In-app browser checked desktop1487×1058 and mobile390×844, navigation, API documentation, visible keyboard focus and actual kit download. Downloaded ZIP bytes match the generated release receipt. Default production explains the unpublished release; local verified downloads require an explicit flag, refused on Vercel. Visual comparison passed; see website/design-qa.md.

No source push, PR, release publication, Vercel deployment, DNS or credentials changed in this phase. Local generated artifacts and browser evidence are ignored.

Owner selected MIT for original Zero. Canonical license text is included in the kit root/nested starter, standalone starter and VSIX LICENSE.txt; SPDXMIT and wrapper Apache notices verified. Two targeted source/license drift and extracted editable Maven checks pass. Main archives contain38/33/21entries respectively. Independent MIT/package integration review61d772b complete with no findings. Source PR4 merged8c54a3c. Public v0.5.0 contains only the three main artifacts; unauthenticated downloads independently matched their receipt sizes/SHA256s. Community archive remains local and its separate license unresolved. Website publication metadata is populated from those verified bytes; native platform limitations above remain unchanged.

• https://zero.codepet.ca serves the minimal landing page, /learn, /community and API docs. GitHub source PR4 and published-download PR5 merged.
• Vercel remote build at c482fc1 passed; canonical sources outside website/ were available during preparation. Authenticated deployed page checks passed before DNS changed. First Git deployment automatically targeted production.
• Cloudflare read-back matches Vercel’s exact DNS-only CNAME recommendation. Vercel domain ownership verified and configuration reports no mismatch. Public HTTPS200 and HTTP308→HTTPS verified without certificate bypass.
• In-app browser clicked DownloadZero on the live site and downloaded the 100306-byte kit, SHA256d296f25f7411f14bdbabfeb8c4015eb90086251102e823cb88854426ee783980. This matches both the manifest and independently retrieved GitHub release. Live Learn→Community navigation and /docs/api HTTP response verified.
• Release v0.5.0 contains kit/starter/VSIX only, all under original-code MIT with upstream notices preserved. Community archive stays local/unpublished.
• Published-version packaging intentionally refuses rebuilding; an actual attempt failed before artifact writes and all three hashes stayed unchanged.
• Native Windows/Linux setup, final editor authentication/upload workflow and novice classroom trials remain unverified; web checks do not establish them.

Owner selected MIT for original Zero Community source. Canonical notice copied byte-for-byte from Zero; current HealthBar metadata is MIT/experimental with null maintainer. License selection does not approve components or publish artifacts. Loader accepts MIT and historical UNLICENSED catalogs with unchanged status and maintainer guards. Focused component/editor tests30/30, website5/5 tests, typecheck, production build and root configuration/link checks pass. Community admission 32/32 boundary tests pass.

MIT schema1 packets carry root LICENSE as the eighth owned file. Native Workshop checks pass, including exact notice ZIP bytes, missing notice and license-drift export rejection. Python independently validates the actual Java-generated packet; its digest matches Python export5f457df0b1c38f7bd4ad6c2584ef82fa3d40878e215e3fbc463e697f7bcbb517. Historical UNLICENSED seven-file packets still validate. Acceptance remains false and publishing disabled without trusted maintainers/review.

Community install→update→revert proof passes6consumer checks in25.93s; immutable 0.1.0/0.1.1 artifact bytes retained. Prepared local catalog records MIT. Regenerated local component ZIP71members contains both canonical license notices and wrapper notices. Published Zero0.5.0 kit/starter/VSIX hashes remain unchanged. No community artifact release is included. Native Windows/Linux/editor input remains untested.

Six tutorials, teacher guide, docs index, examples and read-only source pages integrated into the existing Next.js site; minimal landing retained. Root npm run check, website typecheck,7/7 tests and production build pass. 31 static pages generated plus dynamic landing/learn/source/download handlers. All31 lesson relative file/fragment links resolve. Production browser checks: landing→Learn→tutorial, one/all folds, next lesson reset, API fragment reveal, example→source, mobile menu/lesson selector, keyboard summary and tablet layout. All18 tutorial/docs readers default closed with unique IDs; desktop/mobile/tablet no horizontal overflow. Build-time source allowlist returns200 for a known Java file; traversal, .env, constructor, repeated file queries and missing routes404. No browser warnings/errors. Design comparisons and local receipts ignored in .verification/learning-site/; see website/design-qa.md for visual judgment.

Release0.5.1 manifest, download helpers/routes, dependency locks and all Java sources unchanged. Existing immutable kit is not rebuilt; new website lessons will enter a future kit. Native OS/editor/classroom evidence remains as previously recorded. Independent reviewed-head CI and live publication receipts follow.

Root configuration/link checks and 43/43 release tests pass (40 new publisher scenarios plus 3 preparation guards). Disposable-file fixtures intercept all GitHub/Git publication transports: exact three assets, realistic draft URLs, source/run/checksum drift, expired artifacts, existing branch/PR resume, immutable conflicts, failed CI/current-head guards, public and live corrupted bytes. Website typecheck, 7/7 tests and production build also pass.

The actual release:publish -- --version 0.5.1 --verify-only command retrieved canonical main, checked the pinned tag and three public asset URLs/sizes/SHA256s, and fetched the live Download Zero link and kit bytes successfully. It dispatched no workflow and changed no remote release, branch, PR or credential.

No new version was created solely for testing. Real new-version dispatch, draft publication and generated metadata PR/merge remain exercised through fixtures, not a complete real publication trial. No new Windows/Linux or interactive student/editor evidence is claimed. Existing immutable releases stay unchanged. Independent source review, CI and merge evidence will be recorded in the PR.

Both independent Sol/high initial reviews reproduced one publication blocker: a resumed draft could become public before a conflicting tag was detected. Three regressions failed against the initial publisher. The correction checks existing lightweight/annotated tags against the prepared source before uploaded byte verification and immediately before publication. Wrong tags and tag drift leave the draft private without a publication PATCH; matching annotated tags and absent draft tags retain normal publication. Targeted review follows.

Implementation adds the fixed public catalog/Maven transport and public default consumer while retaining explicit local catalogs, native editor guards and minimal sidebar. Root worker sweep105/105 passed; corrected public/Workshop schema delta 50/50 focused tests passed. Website17/17 tests, typecheck and production build pass. Gateway tests cover four exact artifacts, SHA1/SHA256 sidecars, corrupt/oversized bytes, no unchecked redirect, HEAD, unknown/traversal paths and retry/no-store.

Workshop worker verified historical real add/update/revert and a disposable committed 0.1.2 candidate with the finite native harness plus Java-to-Python packet validation. Community worker verified32 admission and12 public boundary checks, two reproducible builds, three behavior checks, two consumer apps and preserved historical bytes. Final canonical-source candidate/archive, independent review, CI and actual public Maven/VSIX delivery remain next. No new physical Windows/Linux or novice classroom evidence is claimed.

Zero source PR13 and community PR2 merged after independent paired review, one targeted correction review and current-head CI. A generated-free checkout passed documented historical/public preparation, real editor-plan Maven add/update/revert and the native Workshop harness. Canonical extracted 0.1.2 Workshop passed the native build/check/export harness; independent Python validation retained checked/communityReviewed=false/publishAllowed=false. All nine published community v0.1.2 files were downloaded without authentication and matched their measured SHA256s. The initial HealthBar remains experimental; actual authenticated acceptance check did not find independent human approval. Final public verification passed: the exact live catalog, all four Maven artifacts and eight checksum sidecars, portable Workshop and unknown-coordinate 404. A disposable student app used the public adapter's Add plan and ordinary HTTPS Maven from an absent community coordinate; explicit updates, fractional fill/reset, exact JAR hash and preserved Java source passed native macOS checks. Zero0.5.2 preparation run37983866959 sealed the exact three public kit assets. The publisher resumed the same draft/run after an initial GitHub readback miss, then verified public downloads, PR15 metadata merge and live landing kit bytes. No existing asset was replaced. GitHub's native release-lock setting remains unchanged; publisher/receipt checks enforce immutable release coordinates.

The released VSIX installed successfully and was listed as zero.zero@0.5.2 in an isolated extension directory. New native Browse/Try/Add click-through was not completed because UI automation selected the existing Code instance rather than the isolated test process. Automated native-editor mocks and the actual public Maven/JavaFX consumer passed; these do not establish physical editor input. First public0.1.2 has no older public version: future update/revert awaits another patch. The meaningful historical update/revert checks remain separate evidence. Native checks were finite synthetic macOS checks. CI uses Linux/Xvfb; physical Windows/Linux, typing/DirectoryChooser usability and classroom trials remain unverified. No live student upload or AI service was exercised.