Figure 2: Evidence-grounded agent-as-judge evaluation framework. Given the user request, reference
assets, execution trace, intermediate artifacts, and final video, the evaluation agent conducts an agentic
evaluation. At each round, it invokes evaluation skills to inspect traces, extract video metadata, sample
key frames, separate audio, run ASR, analyze errors, and understand visual or audio content. The
collected evidence is then used to score process and output metrics and provide textual feedback.