Skip to content

g10-qualification.md

Latest commit

 

History

History
268 lines (237 loc) · 23.8 KB

g10-qualification.md

File metadata and controls

268 lines (237 loc) · 23.8 KB
Raw

G10 release qualification

G10 is Complete under the approved scope, with the MVP release published. This checklist records what is actually verified; passing one row does not complete the other requirements. The owning contract is G10 in the implementation plan.

Approved scope

  • Qualify the available Wine/Proton environment and native Windows GitHub Actions.
  • Native Windows streaming-PC performance is explicitly deferred by the operator.
  • HTTP works locally and with authenticated LAN access; HTTPS is not required.
  • VTube Studio work is on hold. Streamer.bot remains the production action authority.
  • The tested Speaker.bot voice alias is local english.
  • Use temporary application data and owned examples. Never broadcast, record, spend money or send public chat merely to manufacture qualification evidence.
  • The operator selected “Use owned examples and explicitly leave actual paid-platform delivery unverified.” Reviewed Rant and owned Ko-fi/Bits examples qualify ingestion, totals and local automation in disposable data. They do not prove actual paid Rumble, Ko-fi or Twitch delivery, subscription/gift behavior, or effects on production totals. No test payment is required.

Acceptance checklist

Requirement Current evidence Remaining evidence
G00–G09 prerequisites Maintained entries and protected merges prove completion; integrated runtime evidence below qualifies their combined MVP behavior None for approved scope
Self-contained ZIP and per-user installer Run 37073729890, source 649b2d1, passes native ZIP/installer, startup off/opt-in, reinstall/uninstall and retained data; both package checksums and all uploaded asset digests/sizes match None; v0.1.0 is public
Guided setup without normal-user commands Configuration/review persistence and session-only entry pass; actual final Actions EXE under Wine passes both real-browser bot tests without action/speech/queue receipts None
SQLite-safe backup and restore Actual Actions EXE under Wine passes restart, restore, paused consumers and extra restart; final Windows run passes browser-connected restart/restore and late-handshake regressions None
Repository, wiki and offline documentation Ten chapters/eight owned screenshots published on main and wiki 3913754; actual final ZIP guide passes pages/images, navigation, UTF-8 and main plan link with networking disabled; preparation warnings removed None
Windows CI and SonarQube gate Final source 649b2d1: 75 core/384 host/175 frontend tests, no backend skips, imported OpenCover/LCOV, SonarQube and all protected PR checks pass Completion-record changes follow normal protected checks
Rumble/chat/Rant/ledger/crown and Ko-fi/Bits together Actual shipped production libraries/workers deduplicate owned Rant/chat and project exact/nominal totals and widgets; real bot trigger/forwarder receipts pass. Current real Rumble API reports healthy/enabled/baseline established; operator confirms it None for owned-example scope; actual paid-platform delivery remains unverified
OBS rendering and audible output Actual source reconnect/rendering screenshot proves owned chat, crown/ranking and estimated totals. Fresh saved-rule receipts record OBS sound completion and Speaker.bot dispatch; operator confirms “Heard both” None
Authenticated LAN HTTP Actual Actions EXE under Wine passes remote HTTP sign-in, credential rotation, DPAPI persistence, scoped token revocation and owner-only 403 restrictions; restored to loopback None; HTTPS is not required
Available-environment performance Actual Actions EXE with real Rumble polling, both bots and OBS: 4.616% idle/5.600% chat CPU (one core = 100%) and peak RSS 321.35 MB. Complete workload and separate normalization are recorded below Operator accepts documented Wine CPU/memory deviations. Targets are not claimed as passed; native Windows streaming-PC performance stays deferred

Final release evidence

Protected PR #13 merged on 2026-10-02 at 22:59:22 UTC as be467726b251fa31bb80f4d1eef1711e693b0870. Its Git tree exactly matches tested source 649b2d105bc3bda0bfd1012adcc40bbf190ddef6. Release v0.1.0 was published at 23:02:03 UTC with the verified ZIP, installer and checksum file; the tag points to that merge. GitHub-reported SHA-256 digests and sizes match all three local artifacts. The final downloaded EXE passes the real-browser read-only bot checks and graceful exit under Wine, and its offline guide passes with networking disabled. No production payments, broadcast or recording were used. Completion-record updates are documentation only and follow protected checks.

Earlier checkpoints below retain their original failures/pending states as historical evidence. This current acceptance table and final release record supersede those pending statements.

Current local environment

Inspected on 2026-10-02:

  • AMD Ryzen 9 7950X3D, 16 cores/32 logical processors, x86-64.
  • Linux 7.2.5-3-omarchy; Wine 11.17 Staging.
  • OBS MCP is connected to the local Linux OBS instance; streaming and recording are inactive. This is distinct from running the Windows application under Wine.
  • Streamer.bot 1.0.7 is reachable over WebSocket 8080 and HTTP 7474.
  • Speaker.bot 0.1.7 is now listening on port 7580 after starting its normal launcher directly. Its GTK desktop launch attempt did not produce a running process. No voice or application settings were changed.

Do not normalize CPU usage by all 32 logical processors without saying so. Report process CPU and memory separately from OBS/browser and Wine helper overhead.

Latest checks

  • All 170 frontend tests pass locally after media-test synchronization, with 88.93% line coverage (LCOV generated).

  • The rebuilt local Release host has zero warnings and errors.

  • The full isolated browser suite passes chat, visual editor, financial ledger, donor widgets, automation simulation, setup persistence, real backup download and validation, and permission/LAN settings persistence. It enables no live integration merely by navigating setup or saving unchanged permissions.

  • Local real-process restart and restore each create a new generation; restart retains later setup progress, restore recovers earlier progress, restored integrations/LAN are disabled, and quit stops the owned host.

  • python tools/qualify_bots.py --execute-local-test-trigger-and-queue --speaker-port 7580 passes against the current managed backend and actual installed bots: dedicated custom-trigger receipt, inspector observation, isolated replay, synthetic forwarder execution and Speaker.bot Pause/Resume. This does not establish final Windows-package behavior or audible output.

  • python tools/qualify_kofi.py --execute-synthetic-forwarding-action passes installed CPH forwarding through General.Custom into typed USD support; the owned simulated contribution stays out of the financial ledger. This is a synthetic bridge check, not a real paid donation.

  • Run 37030077956 passed recovery assertions but failed transient SQLite WAL cleanup on Windows. Bounded retries address that test cleanup race; the same recovery step passes in 37031475607.

  • The latest gate reported 82.7% new-code coverage. The offline-guide finding is resolved; the remaining executable-path findings prompted removal of that CLI argument entirely. Only managed, portable and installed modes are accepted; arbitrary paths are rejected before creating test data. The corrected final gate must still be observed, not inferred from these edits.

  • Windows run 37033816613 evaluated a65636a and passed the security gate. Packaging failed while scanning guide sources without its authenticated CI environment. Packaging and native package checks now receive the same conditional scanner environment as existing trusted scan steps; forks receive no secrets. Native package contents are scanned before semantic inspection.

  • Local Codex/VS Code MCP settings and standalone MCP configuration are excluded from public Git. The visual editor now offers manual URL copying when an HTTP browser omits or denies its clipboard API.

  • Run 37037428506 passed analysis and authenticated guide-source scans, then failed reading UTF-8 Markdown with Windows' default legacy encoding. Guide reads now explicitly use UTF-8; final native packaging is still pending.

  • Run 37040945969 passed backend tests but failed one frontend media assertion: the video element existed before React's playback effect applied its volume. The test now waits for the actual volume/mute/loop settings, including updated settings, while retaining its error and unmount cleanup assertions. A new exact-head Windows run is required; this failure did not reach packaging.

  • Run 37041732834 completed successfully. Its scanned, checksum-verified Windows package starts under Wine, but CSRF token creation fails because the default ASP.NET protector opens a CNG SP800-108 provider unavailable here. A standalone owned Windows probe confirms DPAPI round-trip succeeds, default protection fails at that provider, and managed authenticated protection succeeds. The application now selects supported managed AES-256-CBC/HMAC-SHA256 protection; five local host security tests pass, including payload tamper and purpose-isolation rejection. Actual corrected Wine and Windows CI checks are still required; no CSRF or credential protection is disabled.

  • A locally published corrected Windows EXE passes actual Wine LAN HTTP authentication, credential rotation and DPAPI persistence across restart, scoped viewing-token authorization/revocation and owner-only remote 403 checks. Real restart/restore recovers earlier setup progress, pauses restored connections and retains the recorded safety directory. Actual bot connections, dedicated trigger receipt, isolated replay and CPH chat/Ko-fi forwarding pass; the owned simulated support remains excluded from the ledger. Final Actions artifacts and OBS/audio/performance qualification remain pending.

  • Run 37044495358 / ff330f2 completed successfully with the managed protector. Available-environment profiling of the local corrected EXE with connected bots and an OBS preview source measured 4.216% idle CPU over 60 seconds and 5.183% chat CPU over 180 seconds at two synthetic messages/second (one-core normalization); peak RSS was 288.32 MB. CPU targets were not met. A short Warning-logging comparison did not show a CPU improvement and does not justify suppressing diagnostics as a fix. Performance qualification remains open.

  • An OBS-connected Wine restart exhausted the shutdown deadline, then crashed with TaskCanceledException in DatabaseShutdown.StopAsync. The operator's scanned references/backtrace.txt is consistent with that managed exception; it remains private/ignored. Shutdown now cancels browser subscriptions on ApplicationStopping, before Kestrel waits for connections. The expanded real-process qualifier keeps an overlay browser open and proves reconnect after restart and restore. Local managed qualification passes, and an actual locally published Windows EXE passes Wine recovery with the OBS source present; the captured console has no managed crash. A new exact-head CI run remains required.

  • Run 37046606099 / 679f855 completes successfully. Scanned downloaded packages match both checksums; the actual ZIP's ten offline chapters and eight images render with navigation and correct UTF-8. Actual Actions EXE LAN HTTP qualification passes. The local corrected Windows publish also passes the combined real-browser visual/media, automation simulation, financial precision, identity linking, donor appearance/update and reconnect checks. OBS screenshots show four-platform chat after recovery, and PNG alpha inspection confirms a transparent background. These separate checks do not establish all remaining integrated/live or final-release acceptance gates.

  • The Actions EXE restarts and restores with OBS open, then crashes on the next restart after bot configuration is restored. The last overlay request remains open for roughly 30 seconds. A connection can finish its handshake after the original shutdown snapshot, escaping that snapshot's cancellation. The hub now retains shutdown state, rejects new sockets with 503 and cancels handshakes that complete late. Six local socket/shutdown tests pass, including a deterministic delayed-handshake regression and existing concurrent-disconnect behavior. Corrected Wine and exact-head Windows checks remain required.

  • Run 37049814743 / d3a01e9541d35da5f78e0b5294b11f16a45191cc completes successfully, including Windows tests, SonarQube, ZIP/installer and native recovery with a browser connected. Both downloaded package checksums match. The actual Actions EXE under Wine passes authenticated LAN checks, restart/restore and the previously failing additional restart with owned bot configuration. The exact package also passes the full real-browser visual/media, safe automation, financial precision, identity linking, donor appearance and reconnect suite. The late-handshake crash is corrected in this candidate.

  • The operator approved owned paid-event examples with actual paid-platform delivery left unverified. A guarded private producer uses the actual shipped Core/Data/Rumble libraries against disposable application data. The reviewed synthetic Rant fixture yields one authoritative 125-cent Rant and one owned Rumble chat after baseline; reopening the store and repeating its snapshot accepts neither again. Packaged financial workers project exactly one $1.25 Rant, one $5.00 Ko-fi example and 100 Bits at an explicitly configured nominal one cent per Bit: 725 USD cents total. Actual OBS screenshots show the Rumble chat once, a $1.25 crown before the Ko-fi example, then a $5.00 Ko-fi crown, three ranked supporters and $7.25 total with estimates identified.

  • The two explicitly enabled owned rules produce exactly two execution receipts: Speaker.bot dispatch acknowledged using local english, and actual OBS browser sound playback completed. The rules were disabled afterward. Audible output still requires the operator's separate confirmation; these receipts alone are insufficient. These are owned examples, not actual paid-platform delivery.

  • Inspecting the actual candidate's offline Home page exposed a guide-builder bug: external GitHub Markdown links were changed to .html as though they were local chapters. Only local chapter links are now rewritten. A real browser verifies ten corrected pages, eight images, local navigation, UTF-8 and preservation of the remote implementation-plan URL. Native package checks now validate packaged local links and that external URL. Updated Actions packaging must still pass; the earlier candidate does not contain this fix.

Final-candidate available-environment measurements

The actual d3a01e9 Actions EXE was sampled with connected Streamer.bot and Speaker.bot, two owned OBS browser sources, Info logging and three owned ledger contributions. Both owned automation rules were disabled after their bounded audio test. Rumble was disabled pending session-only credential entry; these results do not measure enabled Rumble polling. Streaming and recording stayed off. Chat traffic was 360 nonpersistent, side-effect-free synthetic text messages over 180 seconds, two per second across four platforms, rendered in OBS preview.

Process Idle CPU, 60.02 s Chat CPU, 180 s Peak RSS
Windows backend under Wine 4.449% 5.383% 297.14 MB / 283.38 MiB
Linux OBS, separately observed 3.749% 5.061% 1483.78 MB
Owned Wine-prefix wineserver, separately observed 1.083% 1.389% 16.88 MB

All CPU values above use one logical core as 100%. Dividing backend CPU by the machine's 32 logical processors gives 0.1390% idle and 0.1682% chat; those are explicitly different normalizations and are not substituted for the failed one-core targets. Backend RSS satisfies the 300 MB target in this workload. Idle/chat CPU does not satisfy the respective 1%/3% targets using one-core normalization. OBS and the wineserver are excluded from backend RSS; the table does not sum every Wine helper or OBS child process. This measured deviation remains visible, and native Windows streaming-PC performance remains deferred.

Source-spec completion audit corrections

  • Specification section 86 requires explicit integration connection tests. Setup now provides Test Streamer.bot connection and Test Speaker.bot connection. Each sends a correlated read-only GetInfo on the existing connection; it does not save form values, enable integrations, speak, change queues or run actions. Disabled, rejected and timed-out probes are reported truthfully. Older Speaker.bot metadata rejection does not disconnect an otherwise usable session. Rumble's Connect Rumble tests its actual URL and establishes the polling baseline.
  • Seven targeted backend cases and eight setup UI cases pass. A real browser against an isolated managed host connected to the installed Streamer.bot 1.0.7 and Speaker.bot 0.1.7 passes both new tests; their execution histories remain empty. The refreshed OpenAPI snapshot/generated types pass actual-process contract/recovery qualification. Full local suites pass 175 frontend tests (88.97% line coverage) and 380 host tests with four Windows-only skips. New probe behavior still requires Windows package qualification.
  • Specification section 91's required documentation filenames are now present: architecture, installation, Rumble, events, overlays, widgets, automation, database, security and development. Seven added entry points link canonical plain-language guides and detailed contracts instead of duplicating instructions. README and all public Markdown local links resolve. Stale architecture/index statements are corrected against actual merged G08/G09 evidence.
  • Wiki commit e017419 publishes the read-only test instructions. The regenerated offline guide passes ten pages, eight images, navigation, UTF-8 and preserved external Markdown links in a real browser.
  • Windows run 37054493247 / 86814ff passed analysis and SonarQube, then failed native package checking because the new PowerShell $home variable collided with its reserved, read-only $HOME. It is renamed to $guideHomeDocument. The failed run is not counted as successful package qualification. The corrected verifier and new setup probes require a fresh successful Windows run.

Verified final Windows package — 2026-10-02

  • Windows Actions run 37056770891, source e10b8fb, succeeds: 75 core tests, 384 host tests without skips, frontend checks, native ZIP/installer, restart/restore with an overlay open, startup opt-in/off, reinstall/uninstall and retained data. OpenCover and LCOV are imported; the SonarQube quality gate passes. All PR checks succeed.
  • Downloaded ZIP and installer both match SHA256SUMS.txt. The scanned/extracted Actions EXE passes real-browser read-only connection tests against the installed bots under Wine with empty execution histories, followed by graceful exit and a closed listener. Its bundled guide passes ten chapters, eight images and navigation with browser networking disabled.
  • The exact Actions EXE now serves port 17474 using preserved owned qualification data. An OBS screenshot after handoff confirms one owned Rumble message, three supporters and the $7.25 total with estimates marked. Previous d3 measurements included an owned helper polling status twice per second; that helper was retired before the new measurement.
  • Actual e10b8fb Actions EXE, connected bots, two owned OBS sources, Info logging, disabled test rules and disabled Rumble: idle 60.00 seconds averages 4.000% CPU; ordinary chat 180.01 seconds with 360 synthetic preview messages (two/second) averages 5.144%, with one logical core as 100%. Peak backend RSS across both phases is 298.04 MB (284.23 MiB). Memory meets the 300 MB target; CPU exceeds the 1%/3% targets. Separate all-32-logical-processor normalization is 0.1250%/0.1608%, not a replacement for the one-core comparison. No related-process samples were taken in this run. Enabled Rumble polling and native Windows streaming-PC performance are not qualified.
  • These successful checks supersede the fresh-Windows-package requirement following the historical 86814ff reserved-variable failure. Audible confirmation, real API health, performance disposition and release publication remain open.

Resume checkpoint — 2026-10-02

Final audio confirmation, real Rumble baseline and Wine CPU disposition remain unresolved across at least three consecutive goal turns. The owned host is running with both bots connected; Rumble has no session credential and remains disabled. Windows run 37061856073, commit c3f8cb2, is still active at the SonarQube step. Recheck its terminal result on resume. No merge, tag or release is published. Resolve the operator-dependent gates before protected delivery; do not substitute earlier audio evidence or dispatch receipts for the final audible check.

Resumed operator checks — 2026-10-02

The operator reports “baseline established showing.” Preserve this as operator-observed evidence. A subsequent independent HTTP check found connection refused and no TDSBLive process on this machine; an editor refresh check is pending before audio qualification. Windows run 37061856073 for c3f8cb2 completed successfully. This supersedes its historical active-run status above. Subsequent detached restart of the owned host restores editor HTTP 200. Independent status verifies both bots connected and Rumble healthy/enabled/baseline established, with its session credential present and forwarding disabled. The fresh audio qualification uses the actual e10b8fb production libraries and saved owned rules: one additional $5 Ko-fi example and 100 nominal one-cent Bits produce distinct receipts (speech dispatched/acknowledged-not-playback-confirmed; sound completed/browser-playback-completed). The operator confirms “Heard both,” supplying audible evidence alongside the receipts. Both owned rules are disabled afterward. These extra examples affect disposable test totals only. OBS streaming and recording were inactive. The operator selected “Release with documented Wine limitations,” accepting the measured CPU and memory deviations. Final enabled-Rumble measurement: 60.01-second idle at 4.616% CPU, 180.01-second chat at 5.600% CPU with 360 synthetic preview messages (two/second), peak backend RSS 321.35 MB (306.46 MiB). One logical core is 100%; separate all-32 normalization is 0.1443%/0.1750%. Both bots and two owned OBS sources are connected, Info logging is enabled, rules are disabled and no recurring qualification observer is active. Native Windows streaming-PC performance remains deferred. Final release-guide packaging and protected release delivery remain open.

Final guide preparation

Wiki commit 3913754 publishes the final ten chapters/eight owned screenshots with the accepted Wine limitations. The generated offline guide passes all pages/images, navigation, UTF-8 and the main plan link with networking disabled. Final-source native packaging and protected release delivery still remain; no goal-completion claim is made from these documentation checks.

Completion rule

Keep pending rows pending until their required evidence exists. Record any failed check and its correction; never replace final-package evidence with an earlier candidate, a mock, an acknowledgment or a successful build. Only then update G10 in the implementation plan and publish the qualified release.