multimodalcritiqueneutralMaking evidence available does not ensure that complementary cues across moments are integrated for answering in video understandingComputer Vision02 Aug 2026http://arxiv.org/abs/2607.28516v1