Can evaluation keep continuity, composition, lighting intent, camera feasibility, aesthetic preference and narrative purpose as separate dimensions without collapsing into a leaderboard?
VBench's 16 dimensions are visual; TTCW's 14 tests are textual; ShotBench's 8 are perceptual. None includes 'does this shot serve the scene's purpose'. What rubric and reporting format keeps the dimensions separate and makes the human verdict the anchor?
Why this is still open: Published benchmarks aggregate into a single average; the hub's contract (evidence as three independent dimensions) argues against that.