(A) Manipulation of sensory precision; stimulus probability density functions show our low (15 σ) and high (25 σ) variance conditions; stimulus mean and variance were orthogonally manipulated using a global-motion stimulus. (B) The performance was held constant using adaptive thresholding separately for disgust vs. neutral trials; conditions labels are neutral low variance (NL), neutral high variance (NH), disgust low variance (DL), disgust high variance (DH). (C) Degraded sensory precision reduces perceptual sensitivity; cues had no impact on either motion detection (i) or threshold (ii). Instead, disgust cues selectively increased rightward bias for low-variance stimuli (iii), suggesting arousal altered stimulus expectations. As predicted by interoceptive inference, arousing cues significantly decrease the impact of noise on uncertainty (M-bias) (iv). Curly brackets indicate F-test of 2-way interaction, square brackets indicate post-hoc t-tests (*** p<0.001, ** p<0.01, * p<0.05). All error bars +/- SEM. (D) Although performance was held constant (dark triangles, % correct), participants show considerable variability in metacognitive sensitivity (light diamonds, M-Ratio), reproducing previous results using the signal-theoretic confidence model. (E) Participants had no awareness of cue valence in a post-task forced choice test using identical trial parameters; 95% confidence intervals for d-prime on all four face pairs overlap zero (see Materials and methods, Valence Detection Task).