Original image dithering and saved camera exercises

Use a new output directory and keep the downloaded helper and build script together
The image builder uses Bun and Sharp
The original build used Bun 1.3.11 and Sharp 0.34.5
Use those versions when comparing reproduced bytes and inspect dither-runtime-record.json
Run bun build-dither-hynomas-assets.ts NEW_OUTPUT_DIRECTORY
It creates the original scene nine method and palette PNGs alpha fixture and camera review diagrams
The PNGs are exact palette references while WEBP comparison artwork is a viewing aid
Run uv run --no-project --with numpy --with pillow python verify-dither-native.py NEW_OUTPUT_DIRECTORY
It compares every published PNG pixel and checks palette counts and opaque output alpha

Camera source comparison
Download apps/frontend/src/cameraTrajectory.ts from the exact revision and URL in hynomas-source-audit.json
Keep the source in its own temporary folder and install three@0.179.1 beside it
The type-only types import is erased by Bun at runtime
Run bun verify-hynomas-camera-source.ts PATH_TO_PINNED_CAMERA_SOURCE NEW_OUTPUT_DIRECTORY
The harness checks the fixed source hash before importing the helper
It compares 32 original valid input fixtures with the local implementation
No frontend server camera recorder model weights GPU worker or cloud inference is executed

The saved CameraFrame array uses milliseconds and records position quaternion and FOV
The checker validates numeric fields and ordered indices and timestamps
The source pose rule ignores FOV when trimming the stationary suffix
Retain the untrimmed source when an ending hold or zoom is intentional
The trimmed frame array is not a complete camera path payload or a camera NPZ file
No encoded video duration is inferred from timestamp span