Proceedings of the Annual Meeting of the Cognitive Science Society

Bayesian models gauge and potentially describe how people integrate source reliability and argument strength when updating beliefs, but no existing account captures their joint dynamics. We present GRASP, a process-level model that combines Bayesian-inspired weighting of source reliability with non-Bayesian extensions (a confidence-based gate and a faint-praise function), organised around five core mechanisms: confidence-based gating, source-conditional expectations, faint praise inference, reliability weighting, and bidirectional reliability updating. Our main experimental findings (N= 591) are: Stronger than expected arguments increased source reliability and moved beliefs toward speaker positions, which contrasted with weaker than expected arguments; argument quality dominated initial credentials in reliability judgments; source-conditional expectations govern reliability judgments more than belief updating. The model demonstrates how multiple phenomena emerge from one principle: evidence is evaluated relative to source- and argument conditional expectations.