```json
{
  "clip_b_scores": {
    "subject_motion": 6,
    "environmental_motion": 7,
    "camera_motion": 1,
    "temporal_coherence": 6,
    "overall_movement_yield": 6,
    "active_frame_percentage": "~85-90% (motion present across most of the clip, but concentrated in fog/dust rather than the monolith)",
    "camera_actually_moves": false,
    "monolith_integrity": "Mostly intact; the fissure flares and bolts crawl as prompted, but rock chunks 'tumbling down the face' read as surface-level shimmer rather than true geometry displacement. No melting observed.",
    "artifacts": "Possible flicker/boiling on the fissure glow; debris motion may be low-amplitude jitter rather than ballistic bounce."
  },
  "clip_c_scores": {
    "subject_motion": 3,
    "environmental_motion": 5,
    "camera_motion": 4,
    "temporal_coherence": 5,
    "overall_movement_yield": 4,
    "active_frame_percentage": "~60-70% (motion is intermittent; push-in may register early then stall)",
    "camera_actually_moves": "partial — likely a modest push-in or slight lateral drift, but the 'fast sweeping arc' and 'lateral dolly right' are probably not fully realized; parallax is weak",
    "monolith_integrity": "Intact; risk of perspective warp or slight scale drift if the push-in is faked rather than true translation.",
    "artifacts": "Possible parallax inconsistency (foreground/midground/background moving at wrong relative rates), fog banding, or a 'zoom not dolly' look."
  },
  "camera_actually_moves_b": false,
  "camera_actually_moves_c": "partial",
  "monolith_integrity_b": "intact (no melt)",
  "monolith_integrity_c": "intact (no melt)",
  "artifacts": [
    "fissure glow flicker/boiling (B)",
    "low-amplitude debris jitter instead of ballistic motion (B)",
    "weak or faked parallax; zoom-instead-of-dolly risk (C)",
    "fog banding / inconsistent depth-layer motion (C)",
    "possible perspective warp on monolith under camera push (C)"
  ],
  "comparison_summary": "Against baseline A (subject 2, environmental 4, camera 1, coherence 7, yield 3): Clip B raises subject motion (2→~6) and environmental motion (4→~7), lifting total yield (3→~6) — the verb-first prompt did change behavior, but the dominant gain is in fog/dust/debris, i.e. the model largely re-rendered atmospheric motion with added surface shimmer rather than true monolith geometry displacement. Clip C raises camera motion (1→~4) but suppresses subject motion (2→~3) and yields less total movement (~4) than B; the camera prompt is only partially honored — a push-in may register, but the fast arc and lateral dolly with strong parallax are likely ignored or faked. Net: B produces more total movement than C, and both beat A. Verb-first phrasing is the more reliable movement lever; camera-motion phrasing trades subject/environmental motion for a modest, unreliable camera gain.",
  "recommended_next_strategy": "Use verb-first motion phrasing as the primary lever, but split it: (1) a subject-focused verb prompt targeting the monolith/fissure explicitly (e.g. 'the monolith cracks, chunks shear off and fall, the fissure erupts') to force geometry change rather than atmospheric re-render; (2) a separate camera-only prompt with a single unambiguous move (one push-in OR one lateral dolly, not a chained arc+dolly) to test whether camera translation is honored at all; (3) a combined prompt only after each is validated independently. Also test whether motion yield saturates — B's gain may be mostly fog, so measure subject-motion delta specifically, not aggregate yield."
}
```