# Compact Help native evidence

Branch-specific evidence for [issue 39](http://217.177.34.108:3000/sorokin.oleg.edu/ARCADIA-Studio/issues/39).
The [design record](../../compact-context-help-20260925.md) owns the decision and current acceptance boundary.

`native/` contains the Windows baseline captures from published source
`715036e132677dd2bc2fb37ed316d93dc9987aec`.
`candidate-f830/` preserves the independent review of source
`f83068757af885572f6b3f932d07233dd1b94b7e`, including its two failed Help workflows,
original screenshots, exact executable/runtime hashes and successful build log.
These observations must not be relabeled as acceptance of a successor executable.

`candidate-5657/` preserves the next frozen source's build, strict all-target
Clippy, full desktop test log and independent native review. Formatting, Clippy
and build passed. Tests: 1,538 passed, one failed, seven ignored. The sole failure
was the new caret test's synthetic typing setup before Help opened; `help-failure.log`
records its exact rerun. The next correction uses one character per key event.
The native report records corrected Help keyboard/caret behavior, minimum-window
layout and real operation-refusal feedback, with explicit scope limitations.
Copied text logs have normalized line endings and trailing blank lines removed;
original local logs are retained. Image and fixture hashes are unchanged.

`candidate-6e7c/` preserves final source
`6e7c4ddaf0775198384a8c4882809ea8e14b2976`: formatting and strict desktop/all-target
Clippy pass, the recovered GUI build passes, and 125 focused Help/allocation/guide
tests pass with zero failures/ignores. `build-diskfull.log` preserves the first
failed link; no code change was needed for the recovered build. Its native report
and 15 screenshots cover cancellation, guide consistency, group/primary selection,
minimum-window long names, exact caret return and Undo/Redo/Undo. The runtime
receipt confirms exit 0 and empty logs. Post-close baseline comparison confirms
all 37 copied fixture files unchanged. The transient OSK observation happened
before loading the model and is explicitly retained. Prior source-specific
workflows remain attributed to their actual binaries.

Reproduce final code checks on the documented Windows gnullvm/Rust 1.98.1 toolchain:

```powershell
cargo fmt --all -- --check
cargo clippy -p oa-desktop --all-targets --locked -- -D warnings
cargo build -p oa-desktop --locked
cargo test -p oa-desktop --locked -- context_help_tests allocation_drag guide::tests --test-threads=1
```

The initial test compilation began at that commit; successor work was edited
while it ran. `tests.log` (nine passed, one failed fixture precondition) and
`geometry-tests.log` (nine passed, one failed destination-width invariant) are
diagnostic evidence from that cycle, not frozen-source acceptance. The geometry
run reused its test executable without recompiling. The first command's
`allocation_drag_full_tree_tests` filter matched no tests; the second used the
actual `full_tree` filter plus the affected mounted State/Process checks.

The review used a copy of `examples/release1/project.arcadia` from the published
Windows package. All 37 fixture files matched the baseline before the trial and
again after closing without saving. `baseline-proof.json` records those bytes.
The executable is a local development build, not a published release artifact.

Reproduce the build using the documented Windows gnullvm toolchain and
`cargo build -p oa-desktop --locked`. Place its matching `libunwind.dll` beside the
executable, launch it, and open an isolated copy of the release1 example. Perform
the workflows and use the window sizes in [the native report](candidate-f830/review.md).
The baseline captures establish layout comparison; the report records actual
interaction results and explicit unverified workflows.