Causal Cycle Structure, Criterion Selection, and the Limits of Behavioral Evidence in Consciousness Science

Recurrent Processing Theory predicts that disrupting feedback connectivity produces different effects on the functional correlates of experience than disrupting feedforward connectivity. This prediction is embedded in the shared assumptions of multiple research programs without being owned by any of them as an explicit testable claim. This paper asks what a decisive test of that hypothesis would require, and characterizes the barriers to designing one. Two barriers are identified and typed. The first is methodological: no current tool achieves the staged temporal disruption that would allow the feedforward phase to complete before feedback is selectively targeted. The Allen et al. (2020) sequencing problem is identified as the specific obstacle, and the development trajectories that would address it are characterized. The second barrier is in-principle, conditional on the phenomenological reading of IIT's foundational axioms: any attempt to select a matching complexity measure for the feedforward condition imports the experiential presuppositions the experiment is designed to avoid, and this circularity appears not only at the matching criterion but at any attempt to specify what confirmation would look like. The paper connects this finding to the unfolding argument (Doerig, Schurger, and Herzog, 2019), which establishes the same problem at the level of theory testability generally. The two are convergent: the barrier analysis locates the criterion selection problem inside a specific design element; the unfolding argument demonstrates it is structurally guaranteed for any behavioral experiment targeting causal structure theories. Together they characterize where the hard problem actually lives in the experimental practice of consciousness science — not as a philosophical position but as a locatable feature of experimental design. The paper does not propose an experiment. It characterizes what would be required to design one, why current conditions fall short, and what each of three collaborator communities — anesthesia researchers, methodologists working on IIT testability, and philosophers of science engaged with the unfolding argument — can do with that characterization. Version 2 corrects reference list errors and adds three citations. No changes to argument or conclusions. This paper was developed through extended human-AI collaborative inquiry. All citations were verified by the author against primary sources.