Original rolling shutter timing and Co-Director selector studies

Use a new output directory for reproduction
Keep the downloaded local helper and build script together
Use Bun for build-rolling-codirector-records.ts

Co-Director selector comparison
Download bandit.py from the exact pinned public source URL in codirector-source-audit.json
The source module is separately Apache licensed and retains its original notices
Run python3 verify-codirector-source.py PATH_TO_PINNED_BANDIT_PY NEW_OUTPUT_DIRECTORY
The verifier requires the fixed expected source hash before importing the local module
Only the standalone selector is executed with authored fixtures
No cloud model account or paid service is used by that verifier
Copy the resulting codirector-source-fixtures.json into the output directory
Run bun build-rolling-codirector-records.ts NEW_OUTPUT_DIRECTORY
It builds the default and alternate diagram and search records and compares 72 selected stage choices

Original artwork and analytic animation
Run uv run --no-project --with numpy --with pillow python render-rolling-codirector-study.py NEW_OUTPUT_DIRECTORY --font PATH_TO_ARIAL_TTF
The default font path is the macOS supplemental Arial location
The renderer requires FFmpeg on PATH and the generated Co-Director records in its output directory
It creates an original analytic row timing animation and explanatory artwork
The moving posts are synthetic point samples and do not integrate exposure blur
The enlarged still is decoded from frame 36 of the actual encoded MP4

Run uv run --no-project --with numpy python verify-rolling-codirector-native.py NEW_OUTPUT_DIRECTORY
It checks the row records with rational arithmetic all 72 MP4 timestamps and the decoded first and last row centers at the middle frame
The published threshold tolerance is 1.2 pixels and concerns only the saved synthetic comparison
No physical camera measurement footage repair cloud inference or paper quality benchmark is claimed