In this paper we present a theory to explain why competition enforcers may choose to sacrifice decision accuracy and instead adopt pessimistic beliefs; i.e. beliefs that overestimate the likelihood of infringement. Unlike models where beliefs are assumed exogenous, here beliefs are treated as potentially shaped by institutional or contextual factors. We argue that agencies’ skepticism toward evidence contradicting their beliefs is not the result of misinformation, but rather driven by cognitive dissonance. Surprisingly, perhaps, their pessimism and skepticism may prove consumer and total welfare increasing, though at the expense of compliant firms. We conclude by discussing if and how computational tools and algorithmic decision-support systems may help debias regulatory decisions through systematic evidence processing. We argue they may do so successfully, if they embrace epistemic humility, acknowledge uncertainty and are revised regularly in light of new information. The legitimacy of computational antitrust does not depend on its technical sophistication, but on its reflexive capacity to aid agencies to learn, self-correct, and remain democratically accountable in our digital era.