While predictive processing offers a compelling framework for understanding Autism Spectrum Disorder (ASD), the field is currently fraught with competing theories and mixed empirical evidence. We aim to compare and test current predictive processing theories and establish the empirical basis for a new unified theory of predictive processing in ASD. This theory strives to integrate mechanism-level explanations while also taking into account and identifying the cognitive and neural underpinnings of individual variations in autism symptoms. Our proposal may provide a basis for future individualized therapy. First, we aim to identify the predictive processing approach that best explains ASD by examining anticipatory eye movements via eye-tracking (WP1) and employing advanced EEG methods (WP2). This evaluation promises to offer a nuanced understanding of ASD. However, there are varying ways individuals with ASD might engage in different predictive processing mechanisms. Thus, WP3 focuses on individual differences, exploring how specific predictive mechanisms predict the diverse autistic traits expressed by different individuals. Using a large sample size, this study surpasses group-level analyses to unearth the intrinsic heterogeneity within the population. Our proposed research intends to construct a non-parametric Bayesian model of autistic predictive processing using the datasets of WP1-3. Our investigation uses a well-established probabilistic learning task capable of distinguishing learning-dependent and not-learning-dependent errors, facilitating a detailed analysis of prediction mechanisms. By explicitly comparing various predictive processing approaches, our work advances theoretical understanding and provides valuable insights with potential implications for the refinement of diagnostic criteria and personalized therapeutic interventions. This project can illuminate ASD complexity, address debates, and enhance the well-being of those on the autism spectrum.