Desktop Material

Cheap LFS commit progress and push batching — 2026-07-23

This record covers the final local first-publication acceptance of the Cheap LFS commit terminal, bounded parallel uploads, OCI storage, provider routing, automatic ordinary-Git commit/push batching, and its compact Repository Releases companion surface. Initial integration commit c3db37ea55 is pushed; the isolated responsive correction is now exact-source build/UI accepted but not yet published. This receipt does not claim the correction's CI, CodeQL, Pages or wiki publication, new installer Release, GitHub Discussion summary, or the still-pending final Bambu live Action/fresh-clone proof.

Automated evidence

Scope Result
Release and OCI operations 80/80
Registry transport and app-runtime policy 77/77
Automatic commit/push batching and disposable-Git fixtures 117/117
Commit UI, settings, and English/Cantonese/bilingual localization 157/157
Pinned ORAS preparation scripts 8/8
Headless verifier contract 19/19
Compact commit-shell style contract 7/7

The historical combined rerun for the initial integration passed 151/151. The corrected Releases style/localization/UI plus Pages contracts pass 55/55. A final 152-test integrated rerun ran for 693 seconds without an observed failure, then was stopped cleanly during the disposable-Git batching suite at the user's explicit immediate-push request. It has no aggregate pass claim; a complete rerun remains a handoff item.

The full Cheap LFS folder aggregate completed 261/262 checks. Its only failure was a wall-clock policy case that exceeded the local harness's 2.5-second limit during concurrent heavy Git work. The isolated policy rerun passed 8/8, including that same timeout behavior. This distinction is kept explicit rather than reporting the aggregate wrapper as fully green.

Production build

The exact current worktree was built through the fixed Lowlevel MCP endpoint on an off-screen Windows desktop with:

npx --no-install cross-env RELEASE_CHANNEL=development DESKTOP_SKIP_PACKAGE=1 yarn build:prod

The initial first-publication command returned 0 after 400.46 seconds (404.3 seconds wall). Its compiled out/renderer.css was 1,178,671 bytes with SHA-256 6381556b36c295ba47ad90e8080f4079cbc61951bd7811ab9cb9fc3520638cb1.

After widening the combined compact gate from 760×520 to 800×560, the corrected exact-source command returned 0 after 390 seconds wall (Yarn 387.64 seconds). Its 1,179,200-byte out/renderer.css has SHA-256 6fba1434112ea5c02256a12e6ce8af42f5c870f0db5835155acb8075708d9d28, and out/renderer.js has SHA-256 424c928a6a0f6e3e2437f1549e55ec7e26d8cd98758f6ea22ca53e1d5fb5f32e. The accepted correction app launched from that exact out/main.js; no visible user desktop was used.

An earlier interim build returned 0 after 1,466.27 seconds. That timing and the 400.46-second initial integration build remain historical evidence; the 390-second corrected build and bundle hashes above are authoritative for the current compact Releases frame.

UI evidence

Historical Cheap LFS commit-progress acceptance at immutable source commit c3db37ea5524b91f9603151ae5d1107205f16a59

Historical compact Repository Releases acceptance at immutable source commit 513c5cc96aee045a218837530a11951e8466b618

The promoted rows below are immutable commit-addressed blobs. Their familiar gallery pathnames are labels only; a later capture at the same path on main does not change these dated byte/hash receipts.

Frame / immutable blob Role Dimensions Bytes SHA-256
cheap-lfs-commit-progress.png at c3db37ea5524b91f9603151ae5d1107205f16a59 Promoted wide English frame 1440×960 113,869 3d6358567126e3ce0504b04c4489abbfd473b77546bd82dac834553d50fe9333
Ephemeral inspected narrow frame Accepted local bilingual frame; not gallery-promoted 640×960 85,175 1b99c827d1b5b2cf05298fb1255873acdf0502f72a40437c378c0be7bb989e50
material-github-releases-compact.png at 513c5cc96aee045a218837530a11951e8466b618 Corrected promoted 200%-zoom Releases frame 960×660 89,856 8e29ac666a0832d353126d8dd759200ba7e853016a940501e5c7cbdbb1cf992a

The final wide receipt passed all 36/36 named surface assertions, including noBlockingDialog, plus the required deterministic selection receipt. It kept the Included, Excluded, and Cheap LFS candidate chips, Regex Builder, wrapped hidden-files warning, commit controls, storage recommendation, three distinct worker lanes, complete progress/actions, and the settled over-limit diff state inside non-overlapping regions. The selected MP4 showed the exact The diff is too large to be displayed. message without a spinner, and the undo surface was fully outside the capture.

The final 640×960 bilingual receipt also passed all 36/36 named assertions. After hidden-HWND activation, one real pointer attempt selected the MP4 and settled the same over-limit diff. All three 28-pixel worker rows stayed inside the 178-pixel terminal; the progress surface ended at y=942 inside the panel's y=944 bottom. The viewport had no horizontal overflow and the runtime contained none of the diagnostic style IDs used during earlier investigation. Its semantic receipt and hash are retained as local acceptance evidence rather than a second gallery asset; the ephemeral PNG was removed with its validated owned Temp root.

The corrected Repository Releases proof kept one constant 960×660 physical viewport while probing 100% (960×660 CSS), 125% (768×528 CSS), 150% (640×440 CSS), and 200% (480×330 CSS). Every scale had exactly zero document/body/root/ panel horizontal overflow and at least one complete release row. The 125%, 150%, and 200% cases activated the compact presentation and measured a 176 px list panel, 52.83–53.5 px rows, 30 px checkbox/control floors, a 9 px metadata floor, three metric columns, and a two-column latest card. All three populated rows carried locale-aware 24-hour HH:mm timestamps. Native Enter expanded and collapsed the localized disclosure; search, status, selection, and release-row actions retained focus semantics, while the no-next-page pagination action remained correctly disabled. The gallery source remains 76 images.

For historical audit only, the interim wide frame was 107,411 bytes with SHA-256 6d70fce553edcf54cef9bb806bc1d6f38bf8154a7ff2c859e236aba77afdb238. That interim pass had 35 named assertions plus one pointer receipt; its 640×960 bilingual attempt failed closed because the renderer stayed visibility-hidden. The final compiled-source receipts above supersede that narrow failure without erasing it.

Cleanup and publication boundary

For the interim run, after the then-tracked wide frame was promoted and hash-verified, the generic close API could not address the off-screen HWND. A helper launched only on the owned hidden desktop therefore posted WM_CLOSE to the exact revalidated HWND. The saved Electron process tree exited cleanly; Lowlevel MCP then proved zero owned windows, closed the named desktop, proved the CDP port free, and containment- and reparse-checked the direct Temp child before removing the complete run root. No process was terminated by name or PID in that historical run. For the final correction, the exact launch returned PID 20836, dynamically resolved HWND 1905774, CDP port 52613, provider PID 16700, and provider port 53748. The native close API could not address the revalidated hidden HWND, so only PID 20836 was stopped after exact executable/ profile/fixture/port provenance checks. Lowlevel MCP then reported zero windows. The synthetic credential was deleted and independently proved absent, provider PID 16700 was command-line validated and stopped, both ports were proved free, and the named desktop closed successfully on retry. The cleanup helper removed only the exact owned Temp root; an independent audit found it absent, both saved PIDs absent, zero referencing processes, and zero listeners. The separately owned installed Bambu acceptance environment remained untouched.

These results establish local source/build/UI acceptance for pushed initial integration commit c3db37ea55 and its pending isolated responsive correction. CI and CodeQL, Pages, synchronized wiki, the installer Release, and the required GitHub Discussion remain for the correction's publication step and must be recorded separately after they are actually verified. The final Bambu live workflow/Action and fresh-clone materialization proof likewise remains pending and is not claimed.