0.6.0 validation (2026-10-23)
User confirmed Dread shadows, Kirby shadows and silhouette correction, and smooth Kirby movement. Shared HUD GPU tests:588 cases per renderer including24font cases,52HUDvariants;734nonHUD shaders excluded.786captured Kirby programs compile/link.816receiver GPU cases recover expected positions across both eye signs and signed-depth conventions. Final text-to-panel visual confirmation is pending. No full-game or all-menu guarantee. Kirby60fps requires an external matching patch; stationary warm scene60pairs/sec, shader hitches observed. LinksAwakening F9 slowdown not reproduced in latest user test. Diagnostic captures and experimental persistent recording readback are excluded.

0.5.0 / 2026-10-22
TOTK 1.0.0, RTX5070, Samsung G90XF, 2x OpenGL stereo, FXAA off.
User confirmed Link lighting, matching tree shading, cloud cutouts, fade coverage
and D-pad menu background/flicker corrected. Menu validation requires external
DynamicFPS1.5.4 + UI Blackscreen Fix1.0.1. These mods are not bundled.
Background diagnostics disabled and temporary menu probes removed.
Full-game frame-rate stability and other hardware are unverified.

Ryujinx Stereo 0.4.0 experimental: validation scope, 2026-10-20

New 3D World profile: separate-eye native output; depth 100%, convergence 2250.
Local user confirmed good scene depth before the shadow correction. The shadow
patch targets the observed camera-ray row 22 and linear-depth scale/bias at
21.y/20.w. Nine cached fragment variants compile. A GPU computation test
reconstructs the original receiver X for both eye signs and zero separation,
at depths 100/600/900/3000 and convergence 600/2250. Odyssey isolation passes.
All 169 cached vertex shaders compile through the stereo renderer. Private
game shader inputs are not included. Bowser's Fury is not separately tested;
final visual confirmation of corrected shadows remains pending.

Ryujinx Stereo 0.3.0 experimental: validation scope, 2026-10-20

Both publication engines built successfully as self-contained Windows x64.
The existing Avalonia constructor and Windows-only audio API warnings remain.
The release binaries exclude WonderDrawProbe and its per-frame/per-draw hooks.

The release OpenGL renderer passed all 227 locally sampled Wonder vertex
shaders (184 scene/effect corrections), seven terrain variants, altered
resource-binding checks, and rejection of different terrain layouts. Hidden
GPU draw checks passed for eye separation, depth, convergence, sorting-offset
removal, original occlusion, UI plane, and the six-pixel edge inset at shipped
Wonder defaults. The same shader samples receive no Wonder corrections when
Odyssey is selected. Terrain and particle modes passed separately. Game shader
samples/captures are local test inputs and are not distributed.

ControlChecks passes shortcut parsing, duplicate rejection, foreground gating,
controller chord/repeat behavior, settings preservation, saved-default reset,
and transient feedback. Run after restoring/building with:
  dotnet run --project release-tools/ControlChecks/ControlChecks.csproj -c Release

During local development the user visually confirmed Wonder terrain, coins,
music blocks, text/balloons, cursor and screen edges in tested scenes. This is
not a full-game guarantee. Shader analysis caching reduces compilation work;
first-time shaders can still hitch, and the versioned cache requires warm-up.
No overworld-specific adjustment is included. No new FPS or fresh-PC gameplay
benchmark is claimed for the release. Shared F9/F10 recording is retained;
recording was not independently re-recorded for Wonder in this publication.

The package uses fresh portable configuration, public game profiles and an
allow-list of runtime files. No ROMs, keys, firmware, saves, shader caches,
recordings, personal controller configuration or diagnostic captures belong
in the archive. Dependency licenses/source archives remain included.

Prior 0.2.0 validation (historical, not rerun in full for this release):

Ryujinx Stereo 0.2.0 experimental: validation scope, 2026-10-19

Both current engines compile as self-contained Windows x64. GPU regressions
passed for both eyes/order, one presentation per pair, native shared-context
synchronization and graphics-state preservation. The user confirmed corrected
Zelda ground shadows, menu lighting, fog/smoke, stereo and colors, and smooth
native output in both games. Mario launched through the separate dEPTH
frontend was confirmed as smooth as the desktop emulator with correct 3D.

A real OpenGL capture test with synthetic two-eye quadrants and loopback audio
passed: upright eye order/colors, 3840x2160 H.264 at exactly 60/1 fps, audible
AAC, equal audio/video duration, preserved game GL state, and safe repeated
start/stop/immediate cancellation. The save-before-close test returned the
output path only after MP4 finalization. Both engines compiled after adding
foreground-only fullscreen hotkeys. The user confirmed F9/F10 closed Zelda
and showed the saved MP4. That gameplay clip was not independently located
or decoded; codec/audio assertions refer to the separate synthetic GPU test.

Publication source builds passed for both engines. Relocated library startup
loaded the clean portable configuration with OpenGL and FXAA; --help opens
the library in this frontend rather than exiting. No game was loaded.
Release preparation checks runtime hashes and archive exclusions. The download
uses allow-listed runtime files and fresh keyboard configuration, without
private profiles or game data.

No full-game benchmark or fresh-install gameplay on a second PC is claimed.
Stable 60 fps everywhere, other GPUs/displays and other games are unverified.
Capture repeats source frames below 60 fps. Display runtime and recording
FFmpeg executable are external dependencies.

0.4.0 publication: both self-contained Windows engines built; release-binary shadow GPU checks and control regression checks passed. Existing Avalonia warnings remain.

0.5.0 stereo display output / 2026-10-22
Both Windows x64 engines built and published successfully. Each installed engine
passed all 48 GPU output cases plus actual-engine presentation and live changes
between all formats and both eye orders. Coverage includes odd/even dimensions,
row orientation, full-format eye target sizes, SBS recording composition and
restoration of graphics state. Profile persistence checks passed for all five
game IDs. The expanded settings panel rendered and exited successfully.
These are renderer tests, not new gameplay validation of every game/display.
HDMI frame packing is not implemented; see docs/DISPLAY-OUTPUT.md.
