Figure 1. Theoretical framework and task design.

(A) Schematics of the drift-diffusion model (DDM). Motion evidence is modeled as samples from a unit-variance Gaussian distribution (mean: signed coherence, Coh). Effective evidence is modeled as the sum of motion evidence and an internal momentary-evidence bias (me). The decision variable starts at value a × z, where z governs decision-rule bias and accumulates effective evidence over time with a proportional scaling factor (k). A decision is made when the decision variable reaches either bound. Response time (RT) is assumed to be the sum of the decision time and a saccade-specific non-decision time. (B) Response-time (RT) random-dot visual motion direction discrimination task with asymmetric rewards. A monkey makes a saccade decision based on the perceived global motion of a random-dot kinematogram. Reward is delivered on correct trials and with a magnitude that depends on reward context. Two reward contexts (LR-Left and LR-Right) were alternated in blocks of trials with signaled block changes. Motion directions and strengths were randomly interleaved within blocks.