The Unscheduled Promise: Why Persistent Beliefs Track Their Reinforcement Schedule, Not Their Evidence

Why are false beliefs not extinguished by the evidence that disconfirms them? The dominant answers, that disconfirming evidence is discounted (motivated reasoning) or integrated under strong priors (Bayesian accounts), model belief as a single evidence-sensitive representation whose strength declines, however slowly, as disconfirmation accumulates. This paper proposes that a persistent belief is two dissociated objects. The first is an engagement schedule: the acts the belief licenses (explanatory closure, social alignment, near-miss reinterpretation, imaginative rehearsal), each of which delivers reward reliably and on a defined timescale, whether or not the belief’s promise is ever kept. The second is the distal promise, which licenses the forgoing of alternatives but is never entered on any schedule: it specifies no time at which delivery is due, and therefore generates no event-locked negative prediction error. In a temporal-difference formulation this yields a fixed point at which belief strength is proportional to the rate of delivered engagement reward and independent of accumulated disconfirmation. The account makes a crossed prediction that separates it from all attenuation-based rivals: interrupting engagement should reduce belief strength in the complete absence of any disconfirming evidence, resuming it should restore belief strength without any new evidence, and disconfirmation with the schedule intact should not reduce it. Three predictions and an accompanying program of designs are derived, including an engagement-interruption study executable within a single semester, and explicit falsification criteria are stated for each.