<p>On a narrow understanding, adjudication processes are all about assessing and containing uncertainties of various kinds. If one were to find ways of getting language models deployed in judicial settings to reliably quantify the many shades of uncertainty colouring their outputs, one could end up with a formidable support tool, or so the reasoning goes. This paper challenges this narrow understanding of uncertainty in LLM-augmented judicial practices. Rather than seeing uncertainty as something to be eliminated or quantified by judicial support tools, we argue that some non-quantifiable forms of uncertainty play a vital role in supporting moral perception. While the significance of moral perception to judicial processes has been amply demonstrated, less attention has been paid to how perceptual intelligence depends on the ways in which uncertainty is allowed to surface within institutional infrastructure. Contemporary research on LLMs’ communication of uncertainty offers a unique opportunity to examine this under-appreciated relationship in concrete terms. Can we design LLM-powered augmentation tools that express uncertainty in ways that enhance rather than diminish the relevant forms of moral perception? This question takes on particular urgency given the efforts invested in tackling the epistemic calibration challenges that stem from LLMs’ opaque ‘epistemic façade’. Most of these efforts center on developing methods to quantify and calibrate the reliability of LLM outputs. While such methods may help guard against unwarranted epistemic confidence, these quantitative approaches are unlikely to capture the types of uncertainty that play a key, enabling role within moral perception.</p>