In hypothesis sampling, individuals progressively constrain a space of candidate explanations, eliminating al-ternatives as evidence accumulates. Central to this process is the capacity to avoid returning to hypotheses already rejected, yet the cognitive mechanisms supporting that avoidance remain unresolved. We tested whether efficient sampling depends on working memory capacity (WMC), associated with the active maintenance of tested and avoided hypotheses, or fluid intelligence (Gf), associated with disengagement from them. A large community sample completed a hypothesis sampling task alongside latent-variable assessments of WMC, Gf, and attention control (AC), with AC further decomposed into selective and sustained components. Structural equation models showed that Gf uniquely predicted resampling behavior, fully subsuming the effect of WMC. AC did not mediate the relationship between Gf and resampling; in a reverse mediation, AC's association with resampling was itself accounted for by Gf. Within AC, only selective attention predicted resampling, while sustained attention showed no association. These findings indicate that efficient hypothesis sampling depends on attentional disengagement, the capacity to suppress previously activated representations and reallocate attention toward new candidates, rather than on active maintenance. The residual con-tribution of Gf beyond attention control further suggests that additional processes, plausibly involving relational integration, support the structured reconfiguration of the hypothesis space.Keywords: hypothesis sampling; fluid intelligence; working memory capacity; attention control; disengagement; selective attention; relational integration