30-question benchmark
Measure scene retrieval and timestamp correctness against a human-labeled question set.
Run setup
Load a labeled 30-question JSON file.
Measure scene retrieval and timestamp correctness against a human-labeled question set.
Load a labeled 30-question JSON file.