Physics may be approaching a coherence cliff—a point where the rate of new, mutually inconsistent models exceeds the field’s capacity to reconcile them. Using a functional-state-space (FSS) framework, this preregistered study proposes the first empirical test of coherence loss in physics by modelling research activity as a transition graph inferred from arXiv, INSPIRE, and related open datasets. In this representation, algebraic connectivity (λ₂ of the normalised Laplacian) serves as an operational coherence metric: declining λ₂ indicates fragmentation into disconnected sub-paradigms where corrective dynamics stall. Preliminary measurements suggest that several high-energy and condensed-matter clusters already exhibit decreasing spectral gaps. The protocol outlines a low-cost ($10k) retrodiction pilot that forecasts imminent connectivity thresholds within specific subfields and evaluates whether FSS/DCI—a minimal set of adaptive functions implemented through decentralised collective intelligence—can diagnose or mitigate coherence loss. Even under highly sceptical priors (1% credence), the expected benefit of this pilot exceeds 1000× its cost due to avoidable mis-allocation in frontier physics and other existential-risk domains. The study provides a transparent, falsifiable, and scalable test of whether functional-state-space modelling can offer an early-warning system for fragmentation in fundamental science.