Belief-Space Neglect:A Neurocomputational Framework for Epistemic Gating, Restricted Hypothesis-Space Access, and Model Revision

Model revision depends on more than access to evidence. Across spatial and representational neglect, frontal lesion syndromes, reversal learning, hypothesis generation, anosognosia and metacognition, information can be available yet exert less influence on inference or action than its diagnostic value would predict. Here we introduce Belief-Space Neglect (BSN) as a constrained neurocomputational framework for stage-selective failures in this evidence-to-model pathway. The framework distinguishes verified access to information from five candidate phenotypes: evidence-weighting, conflict/control, hypothesis-space, updating and metacognitive neglect. Its central formal distinction is between the task-defined set of possible hypotheses, the subset that becomes computationally active, and the probability distribution over those hypotheses. Candidate activation is modelled as an epistemic gate, whereas residual belief updating after activation is captured by a separate update-gain parameter. We outline a finite latent-cause Belief-Space Neglect Task that contrasts freely generated with externally supplied alternatives to test whether restricted candidate access can be dissociated from downstream under-updating. BSN is intended as a falsifiable synthesis, not a new diagnosis: it is supported only if stage-specific parameters predict behaviour beyond established accounts of prior strength, evidence reliability, perseveration and metacognitive noise.