# LeanCast 0.3.0-beta validation

This document separates executed local checks from functionality requiring the user's own account and physical devices. No real profile, publishing key or account session is included in this release.

## Executed checks

- **2026-10-07 encrypted-profile regression: PASS, 42 assertions.** Production DPAPI CurrentUser code used fake data isolated under tests/profile-fixtures. Full profile fields round-trip; plaintext key/server markers are absent; replacement is atomic; corruption, oversized input, unknown versions, malformed scenes and duplicate hotkeys reject. Schema-1 migration preserves publishing/audio settings while clearing capture/camera/hotkey consent. No actual saved profile was opened or changed.
- **2026-10-07 hotkey regression: PASS, 12 assertions.** Parser/duplicate detection and fake registration callbacks verified diagnostic suppression, disabled settings, action routing, complete cleanup and rollback after a simulated conflict. No actual global shortcut was registered. Real conflicts still depend on Windows and the user's other applications.

- **2026-10-07 launcher/UI regression: PASS, 79 assertions.** Checked destination selection, default-scene persistence, unfinished setup, source/camera guards, cancel paths, mute hotkeys during native work/countdown, and diagnostics without capture or actual global shortcut registration.
- **2026-10-07 scene/editor regression: PASS, 1,526 assertions.** Generated image/placement/clone/serialization validation and simulated keyboard/mouse/resize controls. No physical sources were used.
- **2026-10-07 viewport regression: PASS, 10 aggregate assertions plus 302 control reachability checks.** All four tabs at 100% and simulated 125% control scaling remained reachable by scrolling in an 800x570 client viewport. Engine and preview remained inactive. Normal/small idle renders were visually inspected.

Native synthetic results are documented separately; they do not establish physical-device compatibility. The included tests are reproducible with tests/run-safe-tests.ps1. Generated local reports are excluded from release archives because they may contain machine paths.

## Native scene and encoded-output checks

The 2026-10-07 native scene probe passed with generated images and a generated audio tone. Four loopback RTMP outputs remained connected through source placement changes, a scene change, pause, and resume. The three widescreen routes shared one H.264 encoder; portrait used a second; all routes shared one AAC encoder. Invalid changes retained the previous composition. A fresh same-process x264 recording succeeded after the dual-view pipeline shut down.

All six preserved FLV recordings decoded completely without reported errors. Decoded image samples confirmed the synthetic camera fixture moved, its old position revealed the background, and the background selected during pause appeared on resume. The pause slate was visually inspected. Decoded audio peak was -11.8 dB before pause, -91 dB at the decoder's silence floor during pause and muted resume, and -12.0 dB after explicitly restoring the synthetic microphone bus.

The source archive includes `tests/NativeSceneProbe.cs`, `tests/NativeSceneSink.cjs`, `tests/verify-native-scenes.cjs`, `tests/NATIVE-SCENES.md`, and sanitized result summaries. This exercise did not activate a real microphone, camera, window, or display and did not send to an external service. The first sandbox attempt could not reach the local receiver; the successful run used permitted loopback networking.

The separate final performance run used the frozen 0.3 application sources. Three trials per case completed; all 18 retained benchmark files passed full decoding. See [the benchmark report](BENCHMARKS.md) for measurements, exact source hashes, and reproduction commands.

## Test boundaries

The safe test suite does not load the actual profile, initialize physical input capture or connect to public endpoints. Launcher fixtures use an empty selected scene so fully populated fake destination details still reject before profile saving or native startup. Rendering uses MainForm(false), which also suppresses global hotkey registration regardless of fixture settings.

Small viewport checks use an 800x570 client area and simulated 125% scaling, with scroll reachability inspected for every tab. Simulation does not establish real monitor-DPI migration behavior. DrawToBitmap may omit owner-drawn combo labels, so the offline renderer paints selected text from actual default control values. No private account data is loaded into screenshots.

Scene tests use generated images and simulated editor interaction. Native synthetic/loopback checks establish local routing and pause behavior only. They do not prove camera compatibility, window/display permission behavior, actual gameplay performance, microphone quality, platform eligibility/authentication, or viewer delivery. Benchmark results apply only to the measured source snapshot, workload and hardware in docs/BENCHMARKS.md.

## Local acceptance

Confirm dependencies, choose a source explicitly, test audio and mute, and inspect a short local recording before public use. Confirm any optional camera frames and both layouts. Test Pause/Resume locally. Inspect each selected platform's dashboard for live state and video/audio health; LeanCast's Sending label is transport status only.

The beta has not established universal GPU support, OBS compatibility outside major30, multi-GPU behavior or real-application frame-time improvement. Driver/network cleanup may delay shutdown.

## Package boundary

Packaging starts in a new empty staging directory and uses an explicit code/docs/assets allowlist and SHA-256 manifest. It excludes personal profiles, credentials, private account helpers, session notes, caches, recordings and raw historical logs. OBS binaries are not bundled. The installer refuses existing app directories/shortcuts and never starts capture or broadcasting. Historical0.2 archives remain unchanged.
