The retrieval geometry of representative judgment

Economic models of representative judgment explain how comparisons distort beliefs, but whether this distortion arises during memory retrieval or at report remains unclear. We evaluate whether episodic retrieval can generate representative judgment. In a key--value model, centered Hebbian learning forms category addresses, and the target and reference jointly determine a retrieval query. Competitive retrieval favors target episodes that distinguish the target from the reference. This mechanism generates a covariance law for judgment bias and, under whitened Gaussian codes, recovers the likelihood-ratio distortion of \citet{bordalostereotypes2016}. It also predicts that changing the reference rotates the cue-period query and changes within-category reinstatement during recall. Simulations show that these signatures distinguish retrieval-based representative judgment from prototype, availability, and response-stage accounts. The model gives a neural interpretation of the weights in economic theory and a test of whether they are present during human memory search.