[i-shutov]

N1 — shared TikZ-native authoring and real generated-source PDF proof

Integration branch exact HEAD at opening: 1ffa195f8ca54805e021e87028fa91c117b7bf63. Draft; no merge before exact-head green tests and reconciliation with current main.

Integrates C07 #39's one catalog (six native TikZ commands, 28 Russian-help parameters, source-range patches) with B07/B08 egui gestures, Split editable source alongside exact PDF and F1/hover/inspector from the SAME catalog. Arrow explicitly requires arrows.meta. Unknown user TeX remains source-only and unchanged. Parameter edits, save/reopen and inverse undo use verified sources. Adds a headless actual app Session proof and a Linux XeLaTeX PDF/font/glyph/source-SHA workflow, not merely handwritten TeX fixture.

Gates

• Exact-HEAD cargo fmt/check/test/clippy and N1 end-to-end CI with inspected JSON/PDF/PNG artifacts.
• Reconcile main without dropping merged feat(I05): source-safe XeLaTeX diagnostics + real cross-platform extension artifact QA #24 or newer fixes, safe exact-SHA leases.
• Add independent A feat(A/W06): external MiKTeX offline doctor and native TikZ primitive build #38 MiKTeX offline doctor, D test(D06): isolated real MiKTeX XeLaTeX CI + six TikZ primitives and strict error proofs #40 native hosted MiKTeX QA, E test(E/N1): accepted shared-catalog help audit, 6/28 source proof and semantic findings #42 help checks and F test(F01): N1 source integrity on merged #44 with typed SourceTarget #41 source-integrity checks after each passes.
• Physical Windows 10 x64 MiKTeX and Mac M1/M4 GUI: NOT_RUN; CI is not manual acceptance.

See plans/2026-10-31_N1_SHARED_CATALOGUE_AUTHORING.md and insights/2026-10-31_N1_SHARED_CATALOGUE_EVIDENCE.md. Stacked B08 #43 must not be merged independently before C07. No Photoshop/CAS/TeX replacement, no auto package installs.

Integration QA ledger — latest source head 1a274485ca8e728643bc0cf0ed79a18c6b4d5c34

• Base main remains 0aa78ae22df0ae84705ba461f017fce256325875 (verified 2026-10-31 10:00 UTC). The integration avoids modifying conflicting main CHANGELOG; N1 provenance is in plans/insights. Add one accepted N1 changelog entry after collision-free merge.
• First exact source head 1ffa195f8ca54805e021e87028fa91c117b7bf63: real N1 application-state→XeLaTeX run 41190253165 PASS. Artifact ID 13038372905 ZIP 68c89937d470061d50145607ccb99911a40f751a96b265aa75b66bc0c0222033 downloaded and independently CRC, 3 PDFs, embedded fonts, Unicode text and PNG reviewed. Not a GUI-pointer run.
• Next head found a real legacy QA binary-selection bug: cargo test -p tikzarium-app grew two binary result records and strict D UI proof falsely read first one. Fix restricts to --bin tikzarium-app with independent regression; all original required GUI/font test names remain mandatory.
• Head 8a2bedbe418848120d44b91188481d3191b8a1d5: N1 run 41190462998 PASS, figure fonts PASS, D QA/Rust CI FAIL due: (1) Windows checkout CRLF static fixture vs generated LF; normalized only static golden (not user source), (2) Clippy 1.91 too_many_arguments in C source-edit helper; replaced with typed SourceTarget(range,revision), retaining fail-closed behavior, not weakening warnings.
• New exact-head checks in progress: Rust 41190945419, D QA 41190945357, N1 real Linux XeLaTeX 41190945352, font proof 41190945413. These must be checked against 1a274485ca8e728643bc0cf0ed79a18c6b4d5c34, no inherited pass.
• External MiKTeX hosted proof test(D06): isolated real MiKTeX XeLaTeX CI + six TikZ primitives and strict error proofs #40 still separate and not accepted at previous failing head. Physical Windows 10 and M1/M4 NOT_RUN, no release.

Final exact-head acceptance of integration workpack

On 1a274485ca8e728643bc0cf0ed79a18c6b4d5c34 all four required hosted workflows PASS: Rust Linux/Windows/macOS 3/3, D QA Linux/Windows Server/macOS ARM and independent crosshost 7/7, N1 app-state generated-source XeLaTeX 1/1, font/Unicode proof 1/1. The N1 artifact 13037984941 ZIP SHA256 a86d8bf24338105f6b9519323b761dab4d55c9a24ef0798abef307d33ff0d800 was downloaded and independently verified: ZIP CRC, 34 files, 9 manifest SHA, 3 one-page PDFs, embedded fonts, extracted Cyrillic and mathematical text, user TeX preservation, save/reopen/undo/conflict JSON. D05 crosshost artifact 13038961920 ZIP SHA256 22863f21bae9472f6ff19ceeccd32f54d416a6c35da22ae209150c504d302b12 independently verified accepted=true, all three hosted OS, seven QA jobs and physical_gui=NOT_RUN. This is source-authoring-state and real XeLaTeX/PDF proof, NOT graphical pointer/Windows 10/M1/M4 physical smoke. No MiKTeX hosted claim from separate #40.

[i-shutov]

Foreground integration continuation: confirmed merge 6b0fe35 and requested immediate bounded runs for A/B/C/D/E/F. I own new branch qa/i06-n1-desktop-pointer, only .github/workflows/n1-desktop-smoke.yml, scripts/qa_n1_desktop.py and I06-specific plans/insights. Goal: launch actual egui executable under a Linux X11 display, drive real keyboard/pointer events, save source and collect actual screenshots, not just Session calls. This complements B's headless egui event tests and D's real Windows MiKTeX #40. No claims of physical Win10/M1/M4 acceptance. No A/B/C production files or existing D workflows will be changed in this block.

[i-shutov]

D06 independent QA handoff (2026-10-31, no physical-host PASS): #40 old SHA 981d2a829eeba94c1f8f364b0ff4c475c80e885b exact hosted MiKTeX run 41190737264 FAILED at a help-text feature gate, after successfully installing pinned official MiKTeX 25.12 in private temp roots and exactly three missing dependencies. Downloaded and inspected real artifact 13038629910: setup.json, both effective root reports, package metadata/versions; real engine MiKTeX-XeTeX 4.16, AutoInstall=0; no compiled PDFs yet. D-only correction + N1 app Session-generated source execution on actual MiKTeX is now one grouped commit b91240852038ee02a0ae5211f12cd8e753f5821d in same PR #40, pending MiKTeX run 41192192094. N1 source producer from merged commit 6b0fe3551cb6ded5cf737e28a7a74326f0b4fe49 (n1-authoring-proof) is invoked, then both generated variants are built with Rust CLI and directly with external private MiKTeX, strict Cyrillic/math/pdffonts/log/source checks. A #38 current head changes independently (D does not edit A); final integrator candidate must be checked at its OWN SHA, not inherit D's result. Windows Server 2022 CI_ONLY; physical Win10/M1/M4 and mouse/egui acceptance NOT_RUN.

[i-shutov]

D06 exact real artifacts handoff, 2026-10-31: hosted MiKTeX run 41192192094, exact #40 SHA b91240852038ee02a0ae5211f12cd8e753f5821d, and actually downloaded/examined 42-file artifact 13038964316. Real private hash-pinned MiKTeX 25.12 XeLaTeX compiled six PDFs, both CLI and direct for handwritten primitives and N1 n1-authoring-proof generated authoring.tex/math-authoring.tex. All source hashes retained; strict missing-library and Lua-only negatives 2/2 PASS (no PDF). Independent real Poppler inspection of downloaded PDFs (outside CI, to diagnose incorrect shutil.which): all 1 page, correct Cyrillic and unicode-math mathematical italic Greek, DejaVuSerif and LatinModernMath embedded; no Missing character:. N1 app Session state reports 6 tools, source-preserving save/reopen/undo and manual-edit conflict PASS, not mouse UI. CI overall FAILED because it selected MiKTeX's same-named pdftotext/pdfinfo/pdffonts rather than independent Chocolatey Poppler and post-inventory mixed MiKTeX stderr advisory into stdout; actual 9130 canonical package rows had zero changes. D-only correction now in PR #40 SHA 54a0d0d7befaa745514c18aa97163507b1c3b11c; new exact MiKTeX CI 41192766128 pending. Important: current merged A BuildService does not yet pass --disable-installer on CLI invocation (direct D proof does). A #38 owns that production fix, not D. Integrator final candidate must verify exact merged SHA and compiler argv, not transfer these hosted CI results to Win10/M1/M4/egui physical acceptance.

[i-shutov]

D06 FINAL hosted MiKTeX QA proof for integrator (2026-10-31): #40 exact verified code SHA 54a0d0d7befaa745514c18aa97163507b1c3b11c; distinct real MiKTeX Windows Server 2022 run 41192766128 SUCCESS. I downloaded and independently audited all 44 files from artifact 13039002942, not merely CI badges. Private official pinned MiKTeX 25.12 XeTeX 4.16, effective private roots and AutoInstall=0 readback, direct --disable-installer real engine probe. Six actual PDF files: handwritten primitives + N1 merged app n1-authoring-proof Session-generated Cyrillic text and Unicode math, each through real Rust CLI and independent direct MiKTeX. 4/4 D test results PASS, missing-library and Lua-only negative tests require exact messages/nonzero/no PDF. All six PDFs 1 page, real independent Poppler 26.09 UTF-8 Cyrillic/math and embedded DejaVuSerif/LatinModernMath PASS, no missing glyph, source SHA preserved, 9130/9130 MiKTeX package inventory identical before/after. Real source state save/reopen/undo PASS programmatically only; egui pointer screenshots NOT_RUN. Important integration boundary: A #38 open owns additional production --disable-installer CLI arg and engine identity on merged candidate; D direct compiler uses flag but pre-A merged CLI uses verified private AutoInstall=0. Final integration must verify A exact merged SHA/CLI flags, and cannot claim physical Win10/M1/M4 PASS. The #40 PR description now carries the full verified artifact manifest, main SHA may advance independently.