A Visual Dependence-Aware Framework for Multimodal Unsupervised Continual Post-Training | TickerVault