Skip to main content

status

This page carries the detailed red/green state that used to live in the README: what has been proven end-to-end, on which rigs, with which caveats. The ROADMAP says where the project is going; this page says what the evidence supports today. The verification ladder itself is defined in milestones.md.

Current release and acceptance boundary — 2026-11-02​

The latest full release is v0.1.0-alpha.1, published on 2026-08-22. Its assets are the Wails wootc.exe and branded *-Installer.exe executables; it has no native *-Installer-Setup.exe, wootc-legacy.exe, or native-SHA256SUMS.

The native package cutover (#345) merged and closed on 2026-10-31. That implementation does not prove a released native artifact or its GUI, setup, accessibility, hardware, and full-cycle gates. Issue #688 tracks the post-cutover defects in authority and publisher policy.

The maintainer reaffirmed VM-first as an RC requirement on 2026-10-30 in #596. Closure of #178 does not establish complete acceptance here. That still needs a desktop on Windows, persistent work, and native promotion of the same system. The soak has no recorded start; earlier native/Wails passes cannot count.

The following dated records remain historical evidence. References below to

209 or #399 as open describe their old run context; both issues are now

closed. Neither closure upgrades an old matrix row to a candidate pass.

The native-cycle matrix below records Windows seed → deploy → native Linux boot → seeded file readable from Linux. It does not prove Phase 1 inside Windows. ADR 0004 needs a usable Linux VM, with persistent user work, before native promotion. That gate remains open.

Development-worktree checks — 2026-11-02​

These checks exercise the current source work. They are not a released artifact, hosted matrix acceptance, or permission to open a support-policy gate.

CheckObserved result and boundary
Full fast tierPassed Bats, Python and Go modules, including catalog synchronization, recovery, import and session controls.
PowerShellRepository syntax lint, the new native observer parser, and all 136 PowerShell unit checks passed on Linux; not Windows mutation proof.
Native shell componentsAll 195 portable shell tests passed. Native/UI test projects cross-built; actual WinUI execution/setup remains unproven.
Legacy frontendAll 65 mock-backed GUI cases passed. Browser interaction also exercised editable secondary profiles, duplicate refusal and corrected input contrast; not native Windows proof.
Mounted migration scriptsIsolated Corral Ubuntu VM running privileged Fedora containers: main suite 90/90, auxiliary folder/profile suites 6/6, 6/6 and 12/12. Actual ordinary-user Documents editing/reopen, preserved Windows originals, collisions and refusal paths were exercised on seeded files, not real BitLocker.
Linux credential backendActual isolated Secret Service creation/reopen and locked, missing-keyring and unavailable-service states exercised. This does not prove a real authenticated browser/application session.

The fixture for nested Windows reached a Windows lock screen, but lacked a usable control channel for QGA. This record claims no pass for a native Windows full-cycle, real BitLocker/TPM, signed setup, firmware mutation or VM promotion of the same system.

Phase 1: Linux inside Windows​

GateStatusEvidence as of 2026-09-26
Windows-hosted VM readiness🟡Experimental helper preflight passed under nested WHPX: both disks, DHCP, scratch mounts, and matching serial/IPC marker; full target desktop unproven
Persistent target preparation and first desktop⚪No complete Windows-hosted target desktop proof yet
Guest work survives VM restart and native promotion⚪Same-disk lifecycle and migration acceptance remain open

The Windows preflight evidence used a disposable target and scratch disk. It did not install a distro. The 2 GiB helper reached its readiness marker in 396 seconds and powered off. These nested-server timings do not measure performance on older PCs. A helper result does not prove Phase 1.

Historical native-cycle evidence (KVM E2E rig: Windows 11 + TPM 2.0 + Secure Boot)​

These runs armed native installation from Windows. Earlier reports called that step “Phase 1”; they did not run the installed Linux desktop inside Windows.

  • ✅ Arm (rung 1): the real wootc.exe arms a virgin Windows VM over QGA — root disk, signed chain, one-shot BCD, state.json = armed (24/24).
  • ✅ Deploy: the Secure Boot chain launches the deployer and fisherman lays down a full bootc image into root.disk — every post-deploy check passes (dracut hook, services, loop-root BLS args, ESP kernel-sync, host-esp.conf).
  • ✅ Native Phase-2 boot (rung 2): the signed Windows BCD → shim → GRUB chain boots the installed Bluefin system from the NTFS-hosted root.disk. The initramfs mounts NTFS with the kernel driver, attaches the raw disk with losetup, resolves the root UUID, runs OSTree prepare-root, switches to the real deployment, reaches the graphical system, and exposes Linux QGA.
  • ✅ Graduate to native disk (Phase 3 / rung 3): the VM boots Phase 2, independently verifies a blank /dev/sdb, runs the native bootc install, reboots into the graduated system, and confirms the file seeded in Windows survived onto the native disk — Windows and root.disk untouched (29/29).
  • ✅ GUI-driven native cycle: native install → native Linux boot → graduation, armed by the real wootc.exe GUI (drive mode), green end-to-end on bluefin:lts. The timelapse on the walkthrough page is that run.
  • ✅ E2E-gated releases: v0.1.0-alpha.1 was published only after a fresh full GUI-driven run passed on the exact tagged commit; nightly green runs cut automatic pre-releases from the SHA they proved.

GUI controls have Playwright coverage. Migration bridges and the planner have unit coverage. These component checks do not prove the VM desktop or that user work survives a change from VM to native boot.

The GUI follow-up run 36255744397 proved the expired-password repair, an interactive Windows session, and GUI launch. It then timed out on the management screen without an Install click. The app classified staged payload files as a previous installation. Issue #399 remains open; this run did not reach the deployer or Linux.

The later GUI run 36265248670 passed on 7e13b976651893e3079959c76e8314a83652a39e (2026-09-26). It restored Windows 11 with BitLocker off and repaired password expiry in the fixture. It observed an interactive session and clicked Install through the real Wails GUI. The deployer completed, then installed Bluefin lts booted. The run graduated Linux to a blank disk that the harness had verified. The graduated system booted from a non-loop root and retained this run's file from Windows Documents.

The failure ledger was empty and the publish job uploaded the timelapse. The retained provenance and pass marker bind that evidence to its source and artifact. This was a branch dispatch; #399 requires three passes on the main schedule. It ended in graduated Linux, so it does not prove a Windows return after graduation.

It does not prove the Windows-hosted VM journey, a GUI editor save, the new installed-boot record, or the WinUI shell. Later candidates still need their own gates.

The full-install native follow-up run 36290894695 passed on b90e46b (2026-09-27, #423). The firstboot service in Linux succeeded and wrote healthy. After Windows returned, wootc.exe status also reported healthy. This was the scripted native cycle with BitLocker off. It does not resolve the GUI failure in #399. An edit of Documents with BitLocker and Phase 1 inside Windows still need evidence.

Native Windows preview components, 2026-09-27​

Run 36328543336 passed all nine native jobs on head 7aee354. The tested merge and build identity was 68e6f452. It observed RPC at startup after peer authentication. It did not create an installer root. ACLs stayed the same. It refused malformed records.

Each package passed the startup check for its brand. These are component checks with the same elevated user. Interactive UAC, original-user collection, the minimum OS, and the full native journey remain open. The source record retains the exact source, jobs, and artifact hashes. This September component run used Wails as the default. The later native cutover is described above; #343 and #344 still require their own acceptance evidence.

Build/test matrix​

Red/green status per combination, from the KVM E2E rig (laptop runners) and the hosted-runner matrix (.github/workflows/e2e-matrix.yml). Legend: ✅ proven green · 🟡 in progress / partially proven · 🔴 known-red (tracked issue) · ⚪ not yet run.

Last full matrix sweep: 2026-07-25 — 12 of 22 cases green, with targeted re-runs since (dates on the affected rows).

⚠️ The ✅s below predating 2026-07-27 are currently UNVERIFIED (per-row exceptions are called out — e.g. the dakota composefs cell was re-proven on the failure-ledger harness on 2026-08-02). Until commit 3d7f9e2, fail() only printed — a failed check that did not itself abort could not stop the run reaching "ALL TESTS PASSED". A real BitLocker run was recorded PASS with [FAIL] User data NOT visible in Phase 2 $HOME in its own log, which is the North Star assertion itself. The harness now records every failure to a ledger and gates the banner on it, so this class of false green cannot recur — but the counts above were produced by the old harness and are being re-run. Treat them as claims awaiting evidence, not as status.

ℹ️ Resolved (2026-07-27): the hosted-runner TPM blocker (#59) was ours, not GitHub's — our sshd wrapper took PID 1 from tini, so swtpm's -d daemonization never wrote its pid file and dockur silently disabled TPM. Fixed in 4a087eb; GitHub-hosted runners are back in scope.

Historical native-cycle matrix (Windows 11 Pro, Secure Boot + TPM 2.0):

Image familyBackend / rootfsNative armDeployPhase-2 bootPhase-3 graduateGUI-driven native cycle
bluefin:ltsostree · ext4-sealed✅✅✅✅ (29/29)✅
yellowfin:gnome (EL10)ostree · ext4-sealed✅✅✅✅⚪
yellowfin:kde / :xfce (EL10)ostree · ext4-sealed✅✅✅⚪⚪
bonito:gnome / :kde / :niri (Fedora)ostree · xfs (unsealed)✅✅✅⚪⚪
dakotacomposefs-native✅✅🔴 #209⚪⚪
marlin (Arch) / flounder (Debian)ostree · xfs (unsealed)✅🔴⚪⚪⚪
bazzite:stable (Bazzite-branded GUI)ostree · GRUB2, Universal Blue MOK✅✅✅⚪✅ (38792319332, 2026-10-15)
bazzite-deck:stable (Bazzite-branded GUI)ostree · GRUB2, Universal Blue MOK⚪⚪⚪⚪✅ (38844455285, 2026-10-16)
aurora:stable (Aurora-branded GUI)ostree · GRUB2✅✅✅⚪🟡 generic exe only (32613697020, 2026-08-23)
bluefin:stable (Bluefin-branded GUI)ostree · GRUB2✅✅✅⚪🟡 generic exe only (32604804979, 2026-08-22)

bluefin:stable and aurora:stable now queue the MOK enrollment of the Universal Blue key too (mokEnroll in app/data/images.json). Their kernel is Fedora-signed, but both ship akmods modules signed by /etc/pki/akmods/certs/akmods-ublue.der. In hosted run 39886118219 (bluefin:stable, Secure Boot on), the deployer skipped the enrollment, and the kernel refused v4l2loopback, so Phase-2 systemd-modules-load.service failed. The green cells above predate that change; no run with the enrollment has been made for either image yet. dakota ships no akmods cert, and this change does not touch it.

bazzite:stable has a green GUI run on 2026-10-15, 38792319332. It ran on b102db9, the head of PR #609, which also contains the launch fix of PR #516. The harness built the Bazzite-branded installer, and the GUI clicked Install.

The deployer queued the MOK enrollment, and the harness drove MokManager. After the enrollment, Phase 2 started with the Bazzite kernel under Secure Boot. Linux read the Documents file from this run, and then Windows started again. Phase 3 was off, and the image download on Windows (WOOTC_PRELOAD=0) was off.

The earlier green run was 32679831098 on 4fd4bc8 (2026-08-24). On 2026-10-15, 38803985996 also passed with the branded deployer menus of PR #610. bazzite-deck:stable has a green GUI run on 2026-10-16, 38844455285, on 139c29a, the head of PR #633. The GUI install, the deploy, MokManager under Secure Boot and Phase 2 passed. Linux read the Documents file from this run, and then Windows started again and wootc.exe status said healthy.

On 2026-10-16, 38832644743 ran on main with the scratch sizing of PR #629 (97552e0). The GUI install and the deploy passed. MokManager enrolled the key, and Phase 2 started under Secure Boot. Then Windows started again, and wootc.exe status said healthy. That run was inconclusive because the Phase-2 guest agent never answered, so the harness could not read the Documents file in Linux.

The deployer had also warned that qemu-ga was absent after the ntfs-3g injection. That warning was a probe defect: the image ships /usr/bin/qemu-ga, and the bazzite:stable run that passed logged the same warning. #633 fixes the probe. Its run above has the guest agent and the user-data check. Before that, 38804321935 (2026-10-15) stopped when the deployer ran out of space. The test record for the Bazzite brand has each check.

The Aurora and Bluefin current runs installed their image through the GUI. Phase 2 started, and then Windows started again. Phase 3 was off. Both runs used the generic wootc.exe. The hosted job did not build the brand installer for each image before 2026-08-24. Thus they prove the image, not the Aurora or Bluefin installer.

#226 needs a green GUI run of each branded-* cell in tests/e2e/matrix.tsv. None of the three cells has that run yet.

dakota (composefs-native) went green on the failure-ledger harness — composefs full chain, run 30710282014 (2026-08-02); #28 is closed. Its Phase-2 first boot after a clean GUI deploy currently hangs CPU-bound (#209, 2026-08-22) — being root-caused for v0.2.0-alpha. marlin/flounder fail in bootc install with "bootupd is required for ostree-based installs": they are ostree images that ship no bootupd.

Axes (against the EL10 / bluefin:lts baseline):

AxisStatusNotes
Windows 11 Pro✅primary proven path
Windows 11 Home / Enterprise / LTSC✅all three green on EL10
Windows 10 Pro✅green in 37 min via the restored Windows base image
Windows 10 Home / Enterprise / LTSC🔴Setup stops on its edition picker — the answer file's product key matches no image in the ISO for those editions — #58
Root filesystem: xfs (unsealed)✅mounted with explicit -t (a typeless mount tried ext4 on xfs)
Root filesystem: ext4 (sealed, fs-verity)✅proven sealed default
Root filesystem: btrfs (sealed)✅green cell on bonito via wootc.filesystem=btrfs (2026-08-22); blocked on EL10 kernels whose out-of-tree btrfs kmod is rejected under Secure Boot
Encryption: none✅
Encryption: tpm2-luks🟡#33 fixed (bffd284, 2026-08-10) — Phase-2 dracut regen works; green cell re-verification pending under the failure-ledger harness
BitLocker FDE (unencrypted-volume path)✅refusal path green 2026-08-22; full install path is the v0.3.0-beta gate — #34. Setup carves unencrypted volume E: for root.disk while C: stays encrypted
Secure Boot db: 2011 / 2023 / both / none⚪harness cells landed for #322 (firmware_db= in tests/e2e/matrix.tsv). The VM firmware gets each db, and Windows reads it back before the run. 2011, 2023 and both must boot the dual-signed shim. The none cell has no third-party CA, so the app must refuse with nothing written. No hosted run yet
Offline (-nic none, pre-downloaded image)⚪code path shipped (branded builds / WOOTC_PRELOAD=1); matrix axis tracked as #217

The historical native chain passed on bluefin:lts through the script path (29/29) and the real GUI. It checked the seeded file after native deployment, Phase-2 boot, and Phase-3 graduation. Upstream discontinued bluefin:lts, so the release gate now exercises that native chain on bluefin:stable. It does not establish the Windows-hosted Phase-1 journey.

Running the E2E yourself​

The E2E harness drives a Windows 11 VM (via dockur/windows) over the QEMU Guest Agent — no guest networking required. It needs a host with KVM, UEFI Secure Boot, and TPM 2.0.

just remote-sync # push + reset a runner to origin/main
just remote-e2e # fresh install + deploy (~30 min)
just remote-e2e-quick # reuse the installed disk (~5 min)

just remote-logs # tail the run
just remote-serial # watch the deployer serial console
just remote-status # grep PASS/FAIL markers

Every E2E run records a sped-up timelapse to tests/e2e/storage/artifacts/<run>/video/. To refresh the walkthrough on the README, publish a passing run's clip:

tests/e2e/publish-visual.sh --from-host <host> # or a local artifact dir
git add pages && git commit -m 'docs: refresh E2E walkthrough' && git push origin main

A GitHub-hosted workflow (.github/workflows/pages.yml) then deploys it to Pages — no self-hosted runner required. The README hero is a committed relative path, so it renders inline on GitHub even before Pages redeploys.