# Native lifecycle verification — 2026-09-11

Codex-ytr83r /root. This is an ongoing Windows verification record, not a release
or complete B4 acceptance. Linux setup, builds and checking remain paused by the
user until an explicit resumption instruction.

## Running application and method

The ordinary Windows application build completed exit0 at15:52:49.829UTC from
production29084050a7c020e0fb488be37c6e1ab235769cc0. Root launched the immutable
copy at15:54:23.838UTC, PID4532, with a separate LOCALAPPDATA profile. Executable
SHA256 is47332A6E6A0A496052754866A0863E5A08FE97AEB1D8A91C590A6AA5479035C2;
the adjacent current-toolchain libunwind.dll is
AA282FA47BB2C4E210CEA5E470CDBDCA58568F0BE0BC73E404D86AB9CD95EF6A.
Launch metadata and the original build log/status are in the
[OA archive](experiments/lifecycle-native-ui-oa-20260911/manifest.json).

All app input and observation used Computer Use against the real Studio window.
Working copies were prepared from the independent admitted fixtures before native
output existed. The unchanged core-only lifecycle oracle compares the complete
Project, reachable SourceBundle and every physical project file. Snapshot files,
screenshots and logs are outside working project directories. Expected names and
UUIDs are specified in [the fixture procedure](fixtures/scenario-lifecycle-native/README.md).

The initial normal client was1440×900. The minimum client1100×700 was reached by
window-border input and verified by the1102×732 full-window capture. A maximized
SA capture measured2245×1232; its normal100% geometry and the restored minimum
window were inspected separately. Fit scales observed were69% for the minimum
root/self cases and60% for the minimum conditional case. OA conditional Fit at the
initial normal size was90%. At100% the minimum canvas exposes ordinary scrollbars;
Fit presents the complete scene. Lifetime ends refer to the actual scene extent,
not all unused blank panel space.

## Observed lifecycle cases

| Case | Actual observation |
| --- | --- |
| Root Create/Action/Delete, OA and SA | Receiver solid lifetime starts at Create and ends at Delete; controller continues. Create capsule selects alias life_oa_110 or life_sa_110, Delete selects life_oa_112 or life_sa_112. Minimum-Fit Delete selection passed in both layers. |
| Empty preexisting, OA and SA | Both headers and solid preexisting lanes remain, with no invented birth/death. Clicking the receiver guide selects the Lifeline (life_oa_202/life_sa_202). SA Properties also displayed its exact UUID ending089a. |
| Create only, OA and SA | Receiver lifetime continues beyond its only message to the root scene endpoint; no Delete is fabricated. |
| Conditional Delete, OA and SA | The first operand's receiver activation ends at its own Delete. The empty alternative keeps an independent activation. Unresolved continuation and unanswered-call cues remain; no Reply is invented. The selected occurrence aliases are life_oa_414/life_sa_414. At SA minimum60% Fit, switching from header402 to Delete414 selected the correct distinct object. |
| Unassigned receiver, OA and SA | Header explicitly shows Unassigned participant. Create510, Delete511 and neutral Lifeline502 remain visible and individually selectable. Full Lifeline name remains in Properties. Missing representation/exchange does not remove the local lifecycle. |
| Self Delete, OA and SA | Single-lane return arrow and death marker align. Marker selects occurrence610, not Lifeline601. SA minimum69% Fit was checked by selecting the header, then the marker; both identities were correct. OA normal100% and minimum-Fit geometry were inspected. |

The intentionally unfinished fixture still reports64 objects,2 links and19 checks.
The input oracle admits19 warnings and no errors; these are not silently treated
as a completed engineering model. No scene insertion or deletion is claimed here.

## Actual persistence results

OA and SA use separate baseline working copies. The only authored change per copy
is the exact root Delete name `OA Delete receiver · reviewed` or
`SA Delete receiver · reviewed`, entered in Properties and applied. Name, Apply and
Save remained accessible at the minimum window size. One Control+Z and one
Control+Y invoked model Undo and Redo; status text confirmed each action.

| Native checkpoint | Independent result |
| --- | --- |
| OA passive navigation and Save | Complete baseline PASS. |
| OA rename and Save | Complete oa-rename PASS. |
| OA one Undo and Save | Complete baseline plus exact same against accepted baseline PASS, with accepted rename supplied as explicit retained-file origin. |
| OA one Redo and Save | Complete oa-rename and exact same against first rename PASS. |
| OA native file-picker reopen, select Delete, Save | Correct reviewed name/alias displayed; complete oa-rename and exact same against first rename PASS. |
| SA passive navigation and Save | Complete baseline PASS. |
| SA rename and Save | Complete sa-rename PASS. |
| SA one Undo and Save | Complete baseline and exact same against accepted baseline PASS, with accepted rename as explicit retained-file origin. |
| SA one Redo and Save | Complete sa-rename and exact same against first rename PASS. |
| SA native file-picker reopen, select Delete, Save | Correct reviewed name/alias displayed; complete sa-rename and exact same against first rename PASS after capture recovery. |

Baseline project bytes are20088, SHA256
444b31b04bbeb06605fee27729e1bc4a6e11fd2cb1288b0f41e59d901961d676.
OA rename bytes are20100, SHA256
3fbe484ecccec18cff1b5ac4f4dd18b3b1cdc161c75a5cc0b1b02c20ce453c86;
SA rename bytes are20100, SHA256
1f77009c68793c8870cac87be68ed4cd292dbbb3e5708cec66cf1da8b5006b82.
Every checkpoint retains the exact86-byte CRLF Unicode rationale, SHA256
8509f8d8a136d808cf98a1bb0fe50348b8a24409f30dee4ee65b62bf5cb9e6bf.
These hashes corroborate, and do not replace, the complete oracle comparisons.

The [closed OA archive report](lifecycle-native-ui-oa-archive-20260911.md) binds
26 JPEG screenshots, five complete checkpoints, seven original oracle logs and
build/launch/failed-Clippy evidence. An independent archivist verified all48
payloads unchanged and byte-exact. OA baseline's first oracle result exists in
root's tool output; it was not retroactively presented as an original saved log.
The [closed SA archive](lifecycle-native-ui-sa-archive-20260911.md) preserves31 JPEGs, five complete checkpoints, eight oracle logs and corrected strict-Clippy log/status. All51 payloads passed independent original/copy and second-readback byte equality.

## Tool failures, current checks and next work

Native capture occasionally returned a16×16 transient owner image alongside the
full app image; input and evidence used the unique full screenshot. The file
dialog's UI Automation focus/cache was unreliable; visually confirmed filename
focus and literal file paths worked for the initial opens and OA reopen.
An attempted border expansion was rejected as outside window bounds; Maximize
provided the wide comparison. One cursor-manager input failure and one restore
capture timeout recovered by reobserving the selected window. Later SA reopen
capture and its one recovery both returned `window capture timed out: timed out
waiting on channel`. Root paused dialog input rather than using stale coordinates.
After strict compilation completed, fresh window selection and activation yielded
a valid capture. Root entered the exact SA filename and completed reopen/Save and
the independent oracle comparisons. This temporal association does not prove a
cause for the transient failures. No corresponding Studio crash is established.

Desktop675/675 and lifecycle22/22 already passed (the22 are included in675).
Strict workspace Clippy then failed on one test-only collapsible-if. The
[equivalent let-chain correction](lifecycle-clippy-correction-ytr83r-20260911.md)
changes no production bytes. Corrected workspace/all-targets strict run76858
passed exit0 at16:45:59.383UTC in1216.612seconds. The affected scenario workspace test then passed1/1 with674 filtered, exit0 at16:56:53.069UTC (408.285seconds). Original logs/status are in the [closed pre-correction archive](lifecycle-header-overflow-before-archive-20260911.md). These results predate the new header correction.

## Wide-name defect and bounded next verification

The typography-wide fixture exposed a real neutral-header defect. At minimum80%, the OA receiver card occupied approximately x515–679 while its18-W name extended to about730. Overflow remained at64%,51.2%, minimum Fit69%, and wide100/125/156.25/195.3125%. The Create capsule text stayed contained. Header and Create clicks still selected the distinct original Lifeline102 and occurrence110, with the full18-W name retained in Properties.

SA displayed the18-CJK-character name with available glyphs at wide100% and125%; its neutral header overflowed while the Create capsule stayed contained. Header102 and Create110 selected correctly and Properties retained the full name. A later attempt to increase CJK zoom was interrupted by the user; no156% outcome is inferred. The user then explicitly reauthorized Computer Use. A fresh main-window capture failed with `window capture timed out: timed out waiting on channel`; the prescribed fresh-window recovery failed with `computer-use request timed out: activate_window`. Root stopped UI input at that point. Combining/emoji native acceptance remains unclaimed.

The [pre-correction archive](lifecycle-header-overflow-before-archive-20260911.md) preserves exactly12 original wide-name screenshots and the affected-test log/status, with complete byte readback. The [bounded correction](lifecycle-header-overflow-correction-ytr83r-20260911.md) adds clipping to the existing shared card text path; it requires its own compile/test and matching GUI verification. Existing model and source checks do not certify corrected pixels.

Native toolbar steps provide51.2%/100%/195.3125%, plus minimum Fit; exact0.5x/2x are geometry-test cases, not selectable native settings. Corrected verification must inspect admitted Latin/CJK/combining/emoji labels, minimum/wide geometry, scrolling and distinct selection, then independently compare passive Save against immutable source predictions. This fixture matrix does not establish complete product usability or feature completeness.

## Git evidence preservation

Before checkpoint commit, root compared the complete index inventory and every Git blob byte against all122 files in the three closed OA/SA/pre-correction archives, plus the preserved historical STATE file:123 exact files,8,470,143 bytes. Nineteen raw logs required explicit inclusion because the repository ignores new log files. The final inventory and all byte comparisons passed; no screenshot or log normalization was accepted.

## Subsequent pre-correction passive Save

After the server/CLI compiler completed, a fresh selected-window capture initially showed an occluding application; root issued no coordinate input against that image. Explicit activation of the already selected Studio window recovered a valid1102×732 Studio capture at125% with the SA CJK header selected. Root clicked Save all, observed `All source files saved`, then froze the complete two-file wide fixture and ran the independent `same` oracle against its immutable admitted typography-wide prediction. Exit0: all identities, fields, reachable source and physical bytes match. The20128-byte project retains SHA2566a18ee1f246321bdf8ab0e7e8e5d4eba134ce486ed5224c4ffd61c7342c3a7d3 and the86-byte rationale retains8509f8d8a136d808cf98a1bb0fe50348b8a24409f30dee4ee65b62bf5cb9e6bf. This remains old production29084050/GUI47332A6E; it proves passive persistence, not corrected clipping or the cause of tool recovery. Original supplemental screenshot/checkpoint/log are preserved separately from the closed12-image archive.