Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The authors formalize a unified sheaf-theoretic framework showing that perceptual binding failures and contextual judgment anomalies stem from the exact same topological obstruction: a non-vanishing first cohomology class ($H^1$) of a presheaf lacking a global section. Grounded in predictive coding and quantum-like contextuality, the work utilizes the complete contextual fraction (CF) over $\mathbb{Z}_2$ to overcome the blind spots of standard Čech cohomology invariants. To validate this shared mechanism, they design a cross-domain experimental protocol that tests for correlated signed obstruction margins across perceptual and decision tasks while rigorously residualizing out baseline response consistency.
Perceptual illusions and cognitive decision paradoxes share the exact same mathematical obstruction—a frustrated cycle in sheaf cohomology—suggesting that contextual inconsistencies across human and artificial intelligence are structural rather than domain-specific noise.
Perceptual binding and the contextuality of judgment are studied apart, in psychophysics and decision research. We argue they share one obstruction: a nonzero class in $H^1$ of a presheaf with no global section -- though only contextuality is tested, since binding's obstruction vanishes. We build on sheaf formulations of predictive coding (Seely 2025) and contextuality (Abramsky&Brandenburger 2011): a cyclic set of pairwise judgments admits a global (noncontextual) explanation exactly when the cyclic (Suppes-Zanotti / n-cycle) inequalities hold. Its obstruction is measured by the complete contextual fraction CF (Abramsky, Barbosa&Mansfield 2017), not the weaker Cech invariant, which certifies contextuality but can miss it (Caru 2017). Penrose (1992) classifies the continuous tribar by $H^1$ over the multiplicative group $\mathbb{R}^+$ of depths -- but the $\mathbb{Z}_2$ case is what we test. We build the perceptual arena from two binary judgments per cyclic-dominance pairing, realizing the same frustrated-cycle obstruction as the survey. The central test is cross-domain: one cohort performs both; shared mechanism predicts a correlation between the arenas'signed obstruction margins -- CF before clamping -- which neither field has measured. Both are inconsistency scores, so a bare correlation is confounded by general response consistency. The prediction is therefore confound-residualized: the correlation must survive partialling out a general-consistency factor from a variance- and reliability-matched control. A sufficiently precise null would count against that account, provided the preregistered reliability checks pass and the control does not load on the obstruction. All three tests are designed but unrun.