Contemporary theories of consciousness share a structural commitment whose cost is now visible: they lack a fully explicit diagnostic separation between formal organization and realization adequacy. The consequences are already on the table. Large language models cross behavioral thresholds that these frameworks have limited principled resources to resist counting as evidence of consciousness, and the most common rejoinder, that biological substrate is required, is usually offered without supporting diagnostic criteria. This paper advances modal consciousness theory as a revised framework that separates the two questions. Formally, conscious experience partly consists in an agent's capacity to navigate intervention-sensitive counterfactual possibilities; distinguishing prediction from intervention (P(Y|X) vs. P(Y|do(X))), I propose a five-level hierarchy of modal access, a bridge from causal models to modal frames, and an epsilon-bounded modal coherence criterion that explains unity across self-relevant submodules. Realizationally, the revision refuses to treat formal adequacy as a consciousness certificate: counterfactual competence is necessary for candidate status without being automatically sufficient under every substrate description. I contrast the account with Predictive Processing/Active Inference, Global Neuronal Workspace, and Integrated Information Theory, arguing that each faces consequences that are often treated as theoretical costs, and that intervention-sensitive structure supplies resources these frameworks lack. Behavioral assays, neural signatures, and AI benchmarks operationalize the claims; an asymptotic sixth-level bound is treated as a limit concept in the appendix rather than a load-bearing part of the hierarchy, and artificial-system claims are framed as graded candidacy rather than settled attribution. The result is an empirically tractable framework connecting counterfactual competence to phenomenal character while remaining answerable to realization-sensitive constraints.