Actor-focused multi-frame temporal scan showed the correct action phase (left woman back to camera, right woman in profile) at t=2618.8s vs the previously used 2615.5s where both women were in different positions. Co-Authored-By: Claude Sonnet 4.6 <[email protected]>