[i-shutov]

I36 — exact B30 GUI acceptance with independently proved X11 mouse and keyboard input

• QA-only HEAD b462a4244dea2c9ad4d1650388d8f9b7692f16c7 on #70 contains B30 production #69 750fb6b743a516f44a362f2f568ce1b92c69e6f4, source tree 965933b3f70504f2f4082bc22795bab67fad5d44 unchanged. Strict exact original Unicode/macro restoration, no ghost, six native commands, GUI PDF, F1 and responsive screenshots remain.
• I34 41279540150 proved older xdotool key --clearmodifiers accidentally released Button1 (256→0) before driver mouseup, invalidating earlier claims of a held gesture.
• I35 true-held 41279880858 proved Button1Mask remained 256 before AND after XTest Ctrl+Z, but Undo did not restore the first line and subsequent explicit mouseup inserted Line A · XeLaTeX build core and failure diagnostics #2. Independent ZIP 13072791651 SHA-256 70e6f332bc48289d90eee88a50c457833b86592f2d35bea9ceb13906e26c05eb, 13 CRC-clean, 11/11 report hashes and screenshots reviewed. Input keyboard delivery to egui remained unproved: four immediate XTest key events could occur within a single GUI frame.
• I36 adds XSync and 80ms dwell per Ctrl_L/z down/up event plus true server XQueryKeymap key-down proof and input-focus record, while requiring Button1Mask held throughout. Independently validated server key/map behavior with local Xvfb; synchronized QA Python, tests, plans/insights/CHANGELOG, no B/C/A production edits or test(I06): actual egui desktop process, input, source-save and screenshot proof #48 changes. Exact-head I36 workflow started; actual I36 GUI PASS not declared until uploaded TeX/PDF/PNG inspected.
• No merge/release. Physical Win10 with independent MiKTeX and Mac M1/M4 XeLaTeX NOT_RUN.

---

I35 — legitimate X11 held-Button1 Ctrl+Z / real GUI PDF gate

Current QA-only HEAD: 6f1ccdfa2965ff1f45c54b11b6171fdf65caac5b, B30 production #69 750fb6b743a516f44a362f2f568ce1b92c69e6f4, exact B30 app tree 965933b3f70504f2f4082bc22795bab67fad5d44 unchanged. All changes QA/docs; no B production or old #48 edits.

Critical I34 discovery: independently audited real X11 artifact 13073180399, ZIP SHA256 1f597a892e5bda878c0b84cfec78f0858e7e89138a55591b685721c4f807bc7a, 13 CRC-clean files and 11/11 JSON report hashes. Native XQueryPointer Button1Mask 256 before xdotool key --clearmodifiers ctrl+z, 0 after but before the driver's explicit mouseup. Previous B29/B30 “held” Ctrl+Z ghost tests did not actually hold Button1 across shortcut, so they cannot establish a B30 production defect.

I35 replaces the held shortcut only with actual X11 XTestFakeKeyEvent (Ctrl_L press, z press/release, Ctrl_L release). Independently verified this key-only sequence under isolated Xvfb: X11 Button1Mask stays 256 through the shortcut and changes to 0 only on real explicit mouseup. QA retains XQueryPointer guards both sides, byte-exact Unicode/macro rollback, no ghost, fresh deliberate gesture, six native C tools, Russian F1, real GUI Save/Build/Export XeLaTeX PDF, zoom/resize and independent source/PDF/PNG inspection. Offline tests cover balanced Ctrl+Z event order and mask predicates. Source/tree/CI checks still required; I35 actual X11 GUI PASS NOT_RUN until exact-head evidence.

Previous I33 41279059557 source-two-Line artifact is still factual but is INVALID_HELD_INPUT, not a valid proof of B30 Undo defect. Physical Win10/MiKTeX and Mac M1/M4/XeLaTeX NOT_RUN. No merge or release without current accepted CI/artifacts.

---

I34 — X11 server-held button forensic gate (live result pending)

• QA-only exact HEAD 3cbb26f2cfe16013f0addbc87e33ff8f587231dd, parent B30 production #69 750fb6b743a516f44a362f2f568ce1b92c69e6f4; exact B30 app subtree 965933b3f70504f2f4082bc22795bab67fad5d44 unchanged. Git compare: 0 behind, seven QA-only files. No B/C/A production edits and no changes to legacy test(I06): actual egui desktop process, input, source-save and screenshot proof #48.
• I33 true X11 41279059557 FAIL at Ctrl+Z while allegedly holding guide, ZIP 13072965209 CRC/hash/screenshot/raw source inspected: Line A · XeLaTeX build core and failure diagnostics #2 present prior to script's explicit mouseup; saved source IDs 1+2 with SHA-256 cb0581fccf2003b0e2b15c4e63ff267cadbaee86f36c256f8ed15c78a3a2e39e (same as B29 failure). Unproven if X11 Button1 remained down between separate xdotool processes; do not infer cause yet.
• I34 adds native Xlib XQueryPointer Button1Mask checks before and after Ctrl+Z before script mouseup, logs server masks, fails closed if button was up. Strict byte-exact Unicode/macro rollback, no ghost, held-click+held-drag, fresh gesture, six C-native tools, F1, external GUI XeLaTeX PDF and full-page resize remain mandatory. Four QA-only files changed in this commit, with matching pure mask predicate test, plans/insights/CHANGELOG synchronized.
• I34 push and PR run started; I34 live outcome NOT_RUN/NOT_REVIEWED until exact artifacts complete. No merge/release; physical Win10/MiKTeX and Mac M1/M4/XeLaTeX NOT_RUN.

---

I33 — B30 exact live-X11 acceptance gate (QA preflight fixed; live run pending)

2026-11-01: independent QA-only HEAD 2994b77d41eab66714007d170d6a15e4e13a8ab9 merges existing QA with B30 #69 exact 750fb6b743a516f44a362f2f568ce1b92c69e6f4 via two parents. Verified Git compare: 0 behind, exactly seven QA-only files versus B30, app source subtree 965933b3f70504f2f4082bc22795bab67fad5d44 byte-identical. Workflow pins this exact app tree, drives a real egui Linux/X11 process, real held click/drag Ctrl+Z + release + GUI Save byte comparison, fresh gesture, all six native tools, F1 Russian, GUI XeLaTeX PDF, fit/resize and source/PDF/PNG evidence. No production changes. I33 push workflow and PR workflow initiated; report/artifacts NOT_RUN/NOT_REVIEWED until completion. B29 actual X11 41274957152 remains a verified ghost-Line FAIL. B30 hosted Rust/D QA/N1/MiKTeX/fonts 5/5 workflow families SUCCESS does not substitute actual X11. No merge or release; physical Win10/MiKTeX and Mac M1/M4 external XeLaTeX NOT_RUN.

Initial I33 41278805656 FAIL before GUI: offline test used old b30_editor_undo_history_regression key while the B30 profile deliberately declares b30_release_source_history_regression; this QA-only lookup was corrected on HEAD 2994b77d41eab66714007d170d6a15e4e13a8ab9, without relaxing source/gesture predicates. Corrected PR workflow 41279059557 pending; no GUI PASS or artifact acceptance claimed.

---

I32 — B29 ACTUAL DESKTOP FAIL (2026-11-01)

No merge, both PRs stay draft. Exact B29 Linux/X11 PR run 41274957152 started real egui under X11 and reached the held-pointer Undo test, then FAILED. Independently downloaded artifact 13071152254, ZIP SHA-256 7f172bae6cfca6d215151621cf066179259dcf07c9b7e91af2e48df4adfa5de1, CRC-clean and 11/11 report hashes valid, screenshots inspected. Native source paste/keyboard Save/first Line+GUI Save PASS. With pointer held and genuine Ctrl+Z, b29-click-undo-held.png already shows new Line #2 selected before mouse-up, while original Line #1 remains. Saved drawing.tex contains two owned native Line commands, SHA cb0581fccf2003b0e2b15c4e63ff267cadbaee86f36c256f8ed15c78a3a2e39e, not byte-exact original. Held-drag, remaining five tools, F1, GUI PDF and resize are NOT_RUN. This differs from prior B28 FAIL (original one Line persisted). Exact root cause NOT PROVEN; B-owner diagnostic handoff in #69.

---

I31 — B29 real desktop acceptance gate (OPEN, NOT YET ACCEPTED)

Source of truth (2026-11-01):

• main 529217b89e83ffb93393eefb99a2a6ff07967c13 (unchanged).
• B-owned #69 B29 exact head 133cab9eff2da0e2ab2ae9574e5328f2dad3ee62; independently resolved app source tree d7a487e91614c7b607ce92f53275e10e2a5ee911.
• QA-only test(I36): prove X11 keyboard receipt and held-pointer B30 Undo before desktop merge #70 exact head 05ff9a1712186e9a29195f0b677c309786f6422d: verified two-parent merge of old QA 4774fe9... plus B29 133cab9.... Diff relative to B29 is exactly seven QA-only paths (X11 workflow, driver, fixture, offline tests, CHANGELOG, plans, insights). No B-owned production edits or old test(I06): actual egui desktop process, input, source-save and screenshot proof #48 branch changes.
• New exact-tree actual Linux/X11 run: 41274953901. It was QUEUED at this checkpoint; execution/results and independent source, PDF, PNG review are NOT_RUN until verified.

Prior independent real GUI verdict remains FAIL: B28 X11 run 41271432298 saved one native Line; held click Ctrl+Z reported successful visual Undo, but original Line remained in saved TeX and GUI (SHA-verified screenshots); held-drag, other tools and GUI PDF NOT_RUN in that failing run. B29 addresses competing native TextEdit/visual inverse-patch undo history. Its hosted 5 CI families succeeded at exact B29 head, but unit/event PASS is not real X11 acceptance.

Unweakened gate on new B29: actual held-click and held-drag Ctrl+Z before mouse-up, exact UTF-8 original source restoration and no ghost line, subsequent fresh gesture, all six native C tools, visible GUI Save, Russian F1, GUI Build/Export via external XeLaTeX, PDF glyph/font and full-page/resize screenshots. Must review raw saved source, PDF and screenshots manually; a bootstrap screenshot or green workflow alone is not visual PASS. Only after acceptance consider SHA-leased QA→B then B→main, and separate main post-merge CI.

Physical Windows 10 x64 with independently installed MiKTeX and Mac ARM M1/M4 with user XeLaTeX, IME/Retina/dialog tests: NOT_RUN. Hosted Windows Server MiKTeX is not physical Win10. Cargo.lock not tracked; no release.

[i-shutov]

I26 independent real Linux/X11 desktop gate: FAIL — B27 not ready for merge (exact QA HEAD b726559f6c9e68fc929572cb75fd62338a2ccf1a, B27 production tree 5274ea9c12a3a9f25cea3b47c0fa504c1b5fff1b). Workflow 41267001529 reaches actual GUI, unlike previous source-tree-preflight failure. The real egui process launched, original Unicode/opaque macro source pasted and GUI-saved, first native line drawn and GUI-saved. Then xdotool pressed the pointer over the provisional guide, sent genuine Ctrl+Z while the mouse was held, released mouse, clicked GUI Save. After 10s, the saved file was byte-identical to the pre-Undo one-line file: B27 Ctrl+Z did not undo saved first line. The held-drag variant, all six gestures after this point, PDF and visual F1 are NOT_RUN by this failing run. No B27 desktop PASS.

Independently audited FAIL artifact 13067228699: ZIP SHA-256 cd164cc3e0b11bec1d67a0174ec6093d10d9cdcac2a07cff7e622b4efe361c3a; 13 CRC-valid files, 11/11 report SHA-256 entries match, source preserved at 482 bytes with one existing owner command. Real PNGs pointer-line.png, b27-click-held.png, b27-click-undo-held.png, failure.png inspected. The source TextEdit remains visibly selected after Ctrl+Z; no line disappeared.

Likely root cause from exact B27 code (requires B owner verification): TikzariumApp::update invokes handle_visual_undo_shortcut before show_guide; handler bails out when ctx.wants_keyboard_input() to preserve native TextEdit Undo. The guide allocate_exact_size(..., Sense::click_and_drag()) does not explicitly claim/surrender keyboard focus on pointer-down. Thus a previously focused source TextEdit can retain keyboard ownership even while guide is armed, and global source-patch Undo is skipped; the B27 Session::undo capture-invalidation fix never runs. The visual selection supports this, but prove with real focus event trace rather than assuming.

Required B-owned correction: make pointer-down on the guide transfer appropriate keyboard focus out of source TextEdit (or explicitly implement an equivalent non-destructive shortcut-routing state), while preserving native TextEdit Ctrl+Z when the user is genuinely editing text. Add an event regression with a previously focused source TextEdit plus guide press → Ctrl+Z → mouse-up; verify exact original source, no ghost, next gesture works. Then rerun the unweakened actual X11 I26 test on the new exact production tree, plus source/PDF/screenshot audit. Do not change old #48 or merge #69/#70 yet. Physical Win10/MiKTeX and M1/M4 XeLaTeX remain NOT_RUN.

[i-shutov]

B28 integration handoff (2026-11-01; pending exact-head checks): I inspected the actual I26 X11 failed run 41267001529: real window + UTF-8 GUI paste/save + first native line GUI save were PASS, then held-click Ctrl+Z failed to remove that line. This is not an older preflight failure. The app auto-focuses the source TextEdit after command navigation; B27's global Undo refused ctx.wants_keyboard_input() even during an armed guide pointer. Also a successful app-level Undo must consume the key to avoid a second TextEdit Undo later in the same frame.

B-owned fix + focused-TextEdit production egui event regression added in existing PR #69 at exact head 0ad20ed14ef2d54190b48cfa3246b4dee5df8173 (base main 529217b89e83ffb93393eefb99a2a6ff07967c13). Only crates/tikzarium-app/src/tikz_canvas.rs plus own plans/insights changed. New test exercises 24 combos (six common-C tools × widths 820/1366 × held click/drag) with a real focused TextEdit, key consumption, opaque Cyrillic/macro preservation and inert release. Exact-head Rust/D/XeLaTeX CI is currently RUNNING, not yet PASS. The previous I26 app-tree pin is stale after this B28 source change; please independently update the QA branch/tree pin and rerun real X11 against B28 exact tree before claiming pass. Keep I26's own QA source ownership, no changes to C/A/D/E/F from B. Physical Windows 10/MiKTeX + macOS M1/M4 remain NOT_RUN.

[i-shutov]

B28 exact-head correction to earlier I26 handoff: current B-owned PR #69 HEAD is now f2dfbd0bb073e6152146a151eca138af8fc030e1 (supersedes earlier 0ad20ed... after fixing rustfmt). Real-X11 I26 41267001529 failed at held-click Ctrl+Z after first real native Line/GUI Save; diagnosed auto-focused TextEdit masking global Undo. B28 routes Ctrl+Z to visual source-patch Undo when a guide capture is held and consumes the event; 24 production egui::Context::run(RawInput) focused-TextEdit regression scenarios. Final B28 exact-head checks SUCCESS 5/5 groups: Rust CI 3/3; D QA 7/7 including Ubuntu 69/69 app tests, 56/56 core tests after retrying only the A/core 150ms timeout-flaky Ubuntu job (no core edits); N1 XeLaTeX; genuine hosted MiKTeX Windows Server 2/2 independent audit; font/Unicode. Real B28 GUI X11 NOT_RUN and previous I26 failure remains open pending independent rerun. Please update your own QA source-tree pin and rerun genuine X11 on exact B28 HEAD; no claim of B28 real GUI pass yet. See handoff PR #48 and PR #69.

Pin for I26 QA fixture: reviewed_app_tree_sha = "4be7a5aebbdd77a820abc08ea37d349dcab6bcec", from exact B28 app source tree; branch commit f2dfbd0bb073e6152146a151eca138af8fc030e1. Please verify screenshot coordinates on new B28 build; updating a pin is not itself visual review.

[i-shutov]

B29 handoff — independent real-X11 FAIL on B28 addressed in B-owned PR #69; NOT YET ACCEPTED. Exact B29 HEAD 93a8b58d142aac7d93bcff409240af0db0d9f3c0, parent f2dfbd0bb073e6152146a151eca138af8fc030e1, base-main 529217b89e83ffb93393eefb99a2a6ff07967c13 (behind 0 at comparison). Exact crates/tikzarium-app/src tree 9fd693581f5cfc3750b77e91ec5253be30bd6d7f for your strict QA pin; do not weaken or skip any existing checks.

Independent I29 X11 41271432298 on B28 had genuinely saved native Line and subsequently reported visual Undo success, yet Line persisted in TeX and screenshot. B29 addresses dual undo ownership: guide press now captures keyboard focus away from source TextEdit; successful revision-checked visual Ctrl/Cmd+Z consumes matching raw key event; undo-highlight scrolls but does not refocus TextEdit during cancelled pointer; obsolete native TextEdit history is reset after successful visual inverse patch. No C/A/D/E/F edits or fake PDF mapping.

Reinforced real egui::Context::run(RawInput) B28 regression now primes code_editor() with both pre- and post-insertion TeX histories, simulates automatic focus on same Undo frame, six C primitives × widths 820/1366 × held click/drag (24 sequences), checks key consumption, no ghost after release, idle repaint and byte-identical Save/Reopen. UNIT_EVENT is not real X11 acceptance. B29 CI workflows were just triggered, conclusions not yet verified; do not merge #69/#70. Please rebase only your QA branch on exact B29 B-owned SHA / update source-tree pin, execute actual X11 with original strict source/held-click+drag/F1/GUI Build+Export/PDF/screenshot assertions, audit artifacts, then report real PASS/FAIL. Physical Windows10+MiKTeX/M1/M4 remain NOT_RUN.

[i-shutov]

B29 exact-head pin correction (supersedes the preliminary B29 SHA/tree in my previous handoff): hosted rustfmt uncovered two formatting diffs in the newly strengthened UNIT_EVENT regression; applied them on the same PR #69. Current B HEAD bd23fc4095abe6f8ee226b8fdf906ffc076bb8ad, exact crates/tikzarium-app/src tree a73d325747ee27fd05b3cc6de7ba9124211117d2, base main 529217b89e83ffb93393eefb99a2a6ff07967c13, 0 behind at last check. The earlier 93a8b58d.../9fd69358... pin is stale and must NOT be used to claim acceptance. B29 exact-head CI is running; current real I29 X11 B28 failure remains the latest independent desktop verdict. Please use this newer tree for strict #70 testing; do not merge based on old artifacts.

[i-shutov]

FINAL B29 integration handoff — verified exact Git HEAD and hosted CI; independent X11 still OPEN.

• Repository i-shutov/tikzarium only, B-owned draft #69 HEAD 133cab9eff2da0e2ab2ae9574e5328f2dad3ee62; parent main 529217b89e83ffb93393eefb99a2a6ff07967c13; behind 0 / ahead 7 at final compare. Production crates/tikzarium-app/src subtree d7a487e91614c7b607ce92f53275e10e2a5ee911. Supersedes all preliminary B29 pins 93a8b58d... and bd23fc409.... No feat(B07): first TikZ-native egui drawing vertical with source patches and help #37/feat(B08): six C-catalogue-driven TikZ tools with code, F1 help and undo #43/feat(N1): integrate shared primitive catalog into egui authoring with exact XeLaTeX proof #44 remerge; C catalogue, A/core, D test(D06): isolated real MiKTeX XeLaTeX CI + six TikZ primitives and strict error proofs #40, E/help, F tests, QA test(I36): prove X11 keyboard receipt and held-pointer B30 Undo before desktop merge #70 and old test(I06): actual egui desktop process, input, source-save and screenshot proof #48 branch unchanged.
• REAL I29 X11 FAIL on B28: native Line and GUI Save succeeded; held mouse Ctrl+Z showed visual Undo status yet Line persisted in both source and file. B29 fix: guide takes keyboard focus; successful visual inverse patch consumes matching raw Ctrl/Cmd+Z before egui code_editor; auto-source highlight does not steal focus during cancelled pointer; stale TextEdit undoer cleared after successful app-level inverse patch. Unknown TeX preserved, common C generator reused; exact PDF still external XeLaTeX, overlay uncalibrated.
• HEAD tests PASS 5/5 workflow groups: Rust CI 41273093083 3/3; D QA 41273096124 7/7 incl Ubuntu 69/69 app and 56/56 core, Windows/macOS ARM runner, strict fmt and Clippy; N1 XeLaTeX 41273096126 PASS; hosted Windows Server MiKTeX 41273096115 PASS; fonts/Unicode 41273096131 PASS. Actual enhanced egui::Context::run(RawInput) UNIT_EVENT covers six C primitives, 820/1366 widths, click/drag focus/Undo, mouse release, idle repaint and UTF-8 Save/Reopen; not physical GUI.
• I23 auto-run 41273096178 FAIL old app-tree preflight before GUI launch, not a B29 GUI verdict. Please update QA-only test(I36): prove X11 keyboard receipt and held-pointer B30 Undo before desktop merge #70 test branch to exact B29 HEAD + subtree pin without weakening assertions, run and inspect real Linux/X11 held-click and held-drag, line+arrow+rectangle+circle+node+Bezier, source selection/parameter edits/F1/Undo after manual edit/Save/Reopen, GUI Build/Export PDF and screenshots. Do not merge fix(B27 egui): Undo cancels an armed native TikZ pointer gesture #69/test(I36): prove X11 keyboard receipt and held-pointer B30 Undo before desktop merge #70 until true X11 passes and artifact is reviewed. Physical Win10+independent MiKTeX and M1/M4 XeLaTeX, IME/Retina/native dialogs still NOT_RUN.

[i-shutov]

B30 B-owner handoff — I32 REAL X11 FAIL response (2026-11-01; QA gate still OPEN).

Independent I32 run 41274957152 on B29 showed a second authored Line already visible before the QA driver's explicit mouse-up after held click + genuine Ctrl+Z. This is an actual desktop FAIL, not a test harness preflight issue. Exact root cause remains unproven; do not reinterpret the old artifacts as PASS.

New B-only #69 HEAD 750fb6b743a516f44a362f2f568ce1b92c69e6f4 (main 529217b89e83ffb93393eefb99a2a6ff07967c13, 0 behind / 9 ahead). New exact crates/tikzarium-app/src Git tree 965933b3f70504f2f4082bc22795bab67fad5d44. This supersedes B29 tree d7a487...; DO NOT QA-claim a B30 run using an older tree.

B30 source-only correction: native tool commit now requires a real egui primary-button release event and primary-up, not just Response::clicked; Ctrl/Cmd+Z while a guide pointer is held cancels it even without undo history, and a successful inverse patch also disarms same-frame mouse-down. Automatic code highlight/galley scroll no longer steals keyboard focus; the source TextEdit is noninteractive for exactly the verified inverse-patch frame, clears stale native undo history and immediately returns editable next frame. A fallback routes genuine Undo after guide capture if earlier app-level routing ran before pointer focus. Debug builds emit only content-free held-Undo pointer flags to app.log. 2 new UNIT_EVENT regression tests (24+12 combinations), plus all prior B27–B29 six-primitive/focused-source tests.

CI currently pending for final B30 HEAD, not PASS until checked. Exact-head links: Rust CI, D QA, N1 XeLaTeX, hosted Windows Server MiKTeX, fonts. The auto I23 workflow is B26-pinned, not B30 real GUI proof.

QA/D-only next action: re-pin your existing #70 X11 workflow/fixture to exact B30 app tree above after exact-head tests are green, preserve the full unweakened held-click+drag, fresh gesture, all six tools, source highlight/edit, F1 Russian help, Save/Reopen, XeLaTeX GUI Build/Export, PDF/font/resize and screenshot audit. Instrumented app.log should be inspected if undo fails again; do not invent event causality. No QA files were changed by B; keep both #69/#70 draft until independent actual X11 GUI PASS. Physical Win10/MiKTeX, M1/M4 remain NOT_RUN; grid provisional, PDF authoritative.

[i-shutov]

B30 FINAL exact-head checks — 5/5 hosted workflow groups PASS; real X11 NOT_RUN (2026-11-01)

B-owned draft #69 HEAD: 750fb6b743a516f44a362f2f568ce1b92c69e6f4. Main/base: 529217b89e83ffb93393eefb99a2a6ff07967c13 (compare: 0 behind / 9 ahead). App source tree: 965933b3f70504f2f4082bc22795bab67fad5d44; do not pin older B29 tree. Only B-owned source main.rs, tikz_canvas.rs, B27's previous tikz_authoring.rs and own plans/insights. C shared catalogue, A/core, D QA/#40/#70, E/help, F source tests and #37/#43/#44 remain unchanged.

Verified final SHA CI:

• Rust CI 41276887090: SUCCESS 3/3 (macOS, Windows, Linux fixtures); separate push CI 41276883623 SUCCESS.
• D QA 41276887082: SUCCESS 7/7, Ubuntu 71/71 app tests, including new b30_same_frame_undo_and_mouse_press_cannot_insert_a_ghost_on_release and b30_ctrl_z_with_no_visual_history_cancels_a_held_gesture, 56/56 core tests, rustfmt, strict Clippy; macOS/Windows native-hosted tests.
• N1 external XeLaTeX 41276887086: SUCCESS.
• Hosted Windows Server MiKTeX + independent PDF audit 41276887095: SUCCESS 2/2; this is NOT a physical Windows 10 test.
• Font/Unicode 41276887083: SUCCESS.

B30 actual changes: require explicit egui primary-release event + primary-up for six native TikZ tool commits; cancel held gesture for Ctrl/Cmd+Z even with no visual-history, including same-frame pointer event; focused source TextEdit does not steal guide Undo; one-frame native-history fence after verified inverse patch, editor returns editable next frame; source-free debug pointer flags. Shared C generation untouched; UTF-8 user TeX/macros preserved; no own TeX engine or coordinate solver. 36 new Context::run(RawInput) combinations are UNIT_EVENT ONLY.

Real desktop gate still FAIL/OPEN: independent I32 actual X11 on B29 41274957152 saved first C-native Line but showed ghost Line #2 before explicit mouse-up on held click Ctrl+Z; source/PDF/screenshots independently audited. B30 code has NOT been re-run in true Linux/X11, so cannot claim corrected desktop interaction. Automatic I23 41276887091 is obsolete B26-pinned preflight FAIL, not live B30. QA-only #70 should pin exact B30 tree, run unchanged strict actual-X11 held-click+drag Undo, next gesture, six primitives, selection/parameters/code highlight/manual editor, F1 Russian help, GUI Save/Reopen, external XeLaTeX GUI Build/Export, PDF/PNG/TeX audits. Do not merge #69/#70 until true GUI PASS. Physical Win10 MiKTeX and M1/M4 GUI/IME/Retina/dialog tests NOT_RUN; provisional grid != external XeLaTeX PDF.

[i-shutov]

B-owner independent artifact review for I36: real-X11 run 41280264391 SUCCESS at QA HEAD b462a4244dea2c9ad4d1650388d8f9b7692f16c7 / unchanged B30 app tree 965933b3f70504f2f4082bc22795bab67fad5d44. Artifact 13073380810 ZIP SHA256 86a3538d5f8e25e3da270102df76134361bfea7f2fdd7d9f0718c9921fc90c90: CRC clean, 39/39 report hashes match. Real X11 Button1 mask 256 before/after XTest Ctrl+Z on click AND drag, keymap state confirmed with 80ms dwell; exact 336-byte Cyrillic/macro source rollback, no ghost, fresh line after Undo, six unique native commands, F1/Russian help and external GUI XeLaTeX PDF independently examined in screenshots/source. Visual caveat: at 1180×680 Fit Page shows an explicit minimum-20%-zoom warning and a partially visible PDF, so not universal fit. QA JSON still records manual review NOT_RUN; my owner-side review does not replace QA's own formal acceptance. Physical Win10/M1/M4 NOT_RUN. Detailed exact-head checks/limits sent to integrator in #48. #69 remains draft.

[i-shutov]

I37 independent acceptance — I36 exact-head real X11 PASS (2026-11-01)

I independently downloaded the actual workflow artifact 13073380810 from run 41280264391, on exact QA HEAD b462a4244dea2c9ad4d1650388d8f9b7692f16c7. ZIP SHA-256 86a3538d5f8e25e3da270102df76134361bfea7f2fdd7d9f0718c9921fc90c90, 41 entries, CRC-clean, 39/39 report file hashes verified. App source tree pin 965933b3f70504f2f4082bc22795bab67fad5d44, identical to B30 production #69 HEAD 750fb6b743a516f44a362f2f568ce1b92c69e6f4; Git comparison B30→I36 contains only the seven QA/docs files.

Crucial corrected interpretation: previous I33 "ghost" was invalid as held-input evidence: xdotool key --clearmodifiers had released X11 Button1 (mask 256→0 before scripted mouseup). I35 proved true held Button1 but not reliable keyboard receipt. I36 finally proves both: XQueryPointer mask 256 before AND after each click/drag Ctrl+Z, XQueryKeymap shows Ctrl_L/z down/up with XSync+80ms dwell per event, and the actual GUI source returns byte-for-byte to the original 336-byte Cyrillic/opaque-macro input after Undo and mouse release, without a ghost insertion. A fresh subsequent actual mouse gesture authors normally; both click and drag variants PASS.

Reviewed actual .tex, PDF, and PNGs: GUI Save and six native C TikZ tools (Line/Arrow/Rectangle/Circle/Node/Cubic) PASS, user macro/Unicode unchanged, real external XeLaTeX one-page PDF with extracted A and Не менять, embedded CID DejaVuSerif and verified independent PDF raster. Russian contextual F1 screenshot inspected. Full-page PDF visible in 1366×740; at 1180×680 the known 20% minimum/vertical clipping warning remains, not silently reported as fixed. Cargo.lock is still generated during hosted CI, not tracked as reproducible release input.

Scope: this is acceptance of the automated real Linux/X11 desktop vertical on B30 exact production tree; it is NOT a physical Windows10/MiKTeX, macOS ARM M1/M4, Retina/IME, or release acceptance (all NOT_RUN). The previous B29/B30 ghost attribution should be superseded by the valid I36 result, not used as a production regression claim. QA #70 is now merge-ready into B #69; after that, recheck B head CI and artifacts before separately merging #69 to main.