14 KiB
Integrated checkpoint — 2026-09-12
Codex-ytr83r, isolated branch codex/canvas-union-ytr83r-20260911. Current source 33e30c1 combines the a57 canvas, cb91525 grapheme and narrow saved-destination repair corrections. Its 530 frozen inputs match HEAD; independent repair review found no actionable finding. Windows desktop tests, strict Clippy, GUI build and bounded matching native/package acceptance passed. The canvas-repair Windows prerelease is published; ZIP, checksum and validation attachments were downloaded and hash-verified. Publication record preserves the exact identities. This is not a full user-goal completion claim. Live canvas integration, native/release acceptance, saved-destination repair and project overview retain ownership and plans. Earlier production 865483b and its package remain preserved.
Verified code and capabilities
Source 33e30c1 binds repair dirty state to the actual saved destination, preserving accepted-model history even when Apply is an identical-content no-op. It selectively incorporates peer 33c6d83 in three files. Windows focused repair 5, full desktop 761, strict desktop all-targets Clippy and GUI build passed; the five are included in 761. All 530 files / 9,727,025 bytes remained unchanged between gates and matched HEAD after the build. The preserved GUI 50486253… passed matching repair/history/recovery and Unicode selection/passive-save checks, packaged walkthrough/help smoke and normal exit0 with empty logs. All559 package files stayed unchanged; fresh extracted server25-request editing/persistence checks passed. Current validation and artifact hashes links the exact raw archives and independent reviews. Repair union and limits, closed gates and exact inputs, verbatim independent review.
Source cb91525 preserves extended grapheme clusters in the shared display helper. Its four focused tests, full desktop 756 and strict desktop Clippy passed; the four overlap the 756. Its GUI build was intentionally superseded by the repair union; matching combined33e native Unicode evidence is recorded above, without reattributing that result to cb91525. Correction and limits, closed Windows evidence.
The isolated a57c334 union selectively combines peer a31b40d canvas/chooser/form prerequisites with B4 lifetime rendering and contextual source/child-work corrections. Windows lifecycle 27, full desktop 752, strict desktop Clippy and Studio build passed on 529 unchanged frozen inputs. GUI 69DBB90F… passed bounded staged source/consumer creation, Cancel, Save, Undo/Redo and reopen checks: 13 oracles and 45 screenshots at the default wide window only. Canvas checkpoint, staged archive, independent review. A fresh a57 camera session subsequently passed bounded OA 100%/125%/Fit and minimum 1100×700 client Fit66 checks, OA/SA pan and moved Create/Delete hit selection, exact passive Save equality and normal exit 0. The 18 images support bounded clipping without toolbar bleed; labels are small at 66%. Camera archive and limits, verbatim independent review. The earlier picker failure remains historical evidence. These pre-grapheme a57 results do not establish current 33e30c1 native or complete usability acceptance.
Earlier published checkpoint 3ade397008cffe826d5f5a70e7c0233159e3bd42 contains production 29084050a7c020e0fb488be37c6e1ab235769cc0; push and exact remote readback passed. It combines prior named behavior creation/history integration (c81e285, e207b04), corrected creation-oracle evidence (b853a1e), CLI-only integration (023e3a8) and B4 lifecycle rendering.
B4 renders local lifetimes, branch-local activation termination, neutral incomplete lanes, Create capsules and Delete markers with original occurrence identity. Real Windows OA/SA six-case matrices and separate name-only Save/Undo/Redo/reopen workflows passed complete independent Project, reachable-source and physical-file comparisons. These are bounded fixture checks, not complete product usability acceptance.
- Native lifecycle observations and limits.
- OA evidence archive and SA evidence archive.
- Lifecycle implementation and independent findings.
- Creation and source integration evidence.
Source 321897fd55af9e07afa2c6ebf1e50be49e86f7d9 integrates the published contextual source-creation/minimum-card corrections from b2c3e540 with B4 and the header clipping correction. Its own Windows desktop all-targets gate passed 704 tests. Exact selective integration, independent union review and Windows results.
Published production 865483b4c66361a01fe0d9e731859220bafb4136 adds the reviewed child-work correction: first refinement transfers existing Allocation and RoleAssignment identities, metadata and artifact references to the child; later siblings inherit only a unique performer. OA/SA regressions cover nested children, roles, folded grouping, unallocated/mixed controls and exact history. The old extracted Studio BF9C2E61… passed OA first-child, sibling and grandchild Save/Undo/Redo checks, then confirmed OA reopen/Save and the SA first/sibling/grandchild, Undo/Redo and reopen/Save continuation. Original identity carriers and complete source/physical snapshots matched. These are native results for 865483b/BF9, not a57/69DB. Original OA evidence, closed continuation, independent continuation review, fixture/oracle and issue4 preserve the exact scope.
Verification scope
On this Windows host, core all-targets passed433 tests with one intentionally ignored scaling probe; that probe separately passed with unoptimized timings. Production29084050 passed focused lifecycle22 and full desktop675 tests (the22 are included in675), and the ordinary GUI build/launch. An equivalent test-helper let-chain correction subsequently passed strict workspace/all-targets Clippy and its focused regression1/1. These results predate the newly identified header-rendering correction.
- Exact current-host logs and inputs.
- Test-only Clippy correction.
- Current-host toolchain setup.
- Independent Windows gap review.
The closed 321897f Windows gate passed 704 desktop tests with zero failed/ignored/filtered; all 261 recorded build inputs remained unchanged. Matching server ABA87A5F… passed semantic HTTP checks (204 schemas, 19 values, 9 rejected invalid shapes) and the corrected REST probe's 25 requests. Closed runtime evidence preserves both earlier source-version failures. Earlier integration checkpoint evidence preserves server/CLI 68 across 7 targets, schema 72/204 after the missing-jsonschema correction, and the old GUI's wide-name passive Save comparison.
The later 865483b Windows desktop gate passed 707 tests, including the three focused child-work regressions, with zero failed/ignored/filtered. All 261 recorded build inputs stayed unchanged. The matching debug GUI build and strict desktop-only all-targets Clippy passed; child-work and staged-source checker compilation/admission also passed. Closed gates and exact inputs. These checks do not establish the remaining native rendering or packaged acceptance.
The extracted package's unchanged server ABA87A5F… passed the actual HTTP schema checks (204/19/9) and 25-request REST editing/persistence/restart probe with a restricted process PATH and unrelated working directory. Seven raw response pairs and exact source snapshots matched; all five test-owned server processes stopped. The first REST run's PowerShell-added PATH directory is preserved as historical evidence; the corrected restricted-PATH rerun is authoritative. Packaged server evidence. This completed the earlier bounded server runtime acceptance. Fresh33e package/server/GUI results are linked above; the old receipts retain their own source attribution.
Material limitations
Earlier native wide Latin/CJK checks found neutral lifeline header text overflowing its card while Create labels stayed contained. The header correction is included in 865483b and retained in a57. Actual corrected wide-W/CJK observations in BF9 show bounded headers and Create labels, including selected zoom, scroll and minimum-window cases. Both complete wide and cluster sources remained unchanged after passive Save. Cluster checks exposed a separate scalar-shortening defect: an accent or joined emoji can be split at the label cutoff. The later cb91525 correction and current33e accent/ZWJ selection/passive-save checks resolve the observed string-splitting case. Long wide labels can still be hard-clipped and arbitrary font/DPI presentation is unverified; see the historical native typography record.
The earlier capture failures and pending Open dialog are historical: later confirmed OA and SA reopen/Save actions are preserved in the continuation archive. Its first OA reopened-1948 receipts remain data-only evidence. Bounded a57 staged and camera workflows above are complete; full focus/keyboard, every clipping edge/zoom and accessibility remain unverified. Bounded combined 33e30c1 native/package acceptance is now linked above. Small 66% labels are a recorded limitation. The BF9 Home page, scenario guide and bundled water-pressure example opened; one normal Alt+F4 then completed, with the process absent and empty stdout/stderr. The old Home/Guide/close archive preserves those old865/BF9 observations and package readback. No exit code was captured for that older BF9 process; the separate a57 exit 0 does not change its evidence.
The selected chooser/canvas union now has a57 source and Windows gate evidence; broader persistent graphical manipulation and separate text-navigation contributions retain their own integration/acceptance boundaries. Calculations, quantities/units, maintained libraries, practical large-model textual workflows, complete OA↔SA engineering parity, documentation and fresh cross-disciplinary review remain full-goal requirements. See the analysis proposal, library proposal, text-authoring design and decisions for attributed design evidence, not completed implementation claims.
Platform and published deliverables
Linux setup, downloads, builds and checks are paused by explicit user instruction until the user reports networking resolved and instructs resumption. Ubuntu26.04/WSL2 shell execution was observed; guest networking remained unavailable. Offline preparation was a workaround, not a network fix or Linux acceptance. Host VPN settings were not inspected or changed. WSL evidence and preserved offline inputs retain exact outcomes.
Peer Codex-9e8r6t published the separately verified Windows contextual prerelease, source b2c3e540; its publication handoff binds its artifacts and original-host checks. It is not a B4 release from this branch.
Record navigation
The one-time record migration preserves the exact prior STATE ledger and marks PLAN/TASK_LEDGER superseded. Earlier integration history preserves source-attributed older results. Historical leases, failed approaches and next-step wording do not override current Gitea coordination or the Linux pause.