[guycochran]

Summary

• correct the existing MXL-vs-NDI page's ratio wording and remove the unsupported picture-quality claim
• publish live ZoomISO CPU evidence plus a controlled identical-frame one/two-reader benchmark
• add the reproducible isolated harness and public-safe aggregate CSV
• document the exact scope and limitations: same-host CPU-only v210 switcher paths, not a universal transport verdict

Results

• live ZoomISO path: NDI-to-v210 MXL used 2.18x as much average CPU; native MXL reduced measured CPU about 54%
• controlled path: NDI-to-v210 MXL used 1.78-2.14x as much median CPU; native MXL reduced measured CPU 43.9-53.4%
• all accepted readers held approximately 30 fps with zero drops

Verification

• full Python test suite
• Node test suite
• Python compileall
• shell syntax checks
• git diff --check