Drawing on an audit of 66 papers, 137 evaluation runs, and 780 metric-use records, this discussion traces how evaluation changes across short video, long video, and video world models, then identifies a benchmark direction suitable for a small team.