Physics-consistency held-out on open video — public clip / critique ask

Hi maintainers —

I’m Mushood Hanif. Independent held-out physics-consistency check on CogVideoX-5b I2V (n=9, static_hover flag rate 1.000, Wilson 95% CI [0.701, 1.000]; one failure mode / scene family). Looking to score a second open model under the same frozen detectors (CoTracker3 → VIDEO_CHECKS).

Writeup: https://haga.mushoodhanif.com/article/sim-physics-consistency-v1#held-out Lab: https://haga.mushoodhanif.com/lab/physicsiq

Ask:

1. Critique welcome on whether freeze/hover-via-tracker is a fair diagnostic for open T2V outputs.
2. Pointers to public Open-Sora Plan sample videos (or permission to evaluate specific releases) with provenance — I’ll return aggregate flag rates + method limits only unless you say otherwise.

Note: the org email [email protected] currently has no MX record, so filing here instead. Thanks for the open release.