Higher-order theories (HOT) of consciousness claim a mental state is conscious only when a second, higher-order state represents it. This paper argues HOT is not a genuine theory but a labeling scheme: it specifies a bare relation between two states, calls one 'conscious,' and supplies physical details and evidence only informally as needed. Because HOT refuses to commit to any specific physical implementation, it cannot legitimately use empirical findings as confirmation. The regress problem—the oldest objection—reveals structural failures: the theory provides no resources to block regress, license empirical citations, or distinguish conscious from non-conscious pairs. Defenders add ad hoc fixes (asymmetry stipulations, signal detection apparatus) that protect the bare biconditional, achieving unfalsifiability through variant-switching rather than absorbing anomalies.
A preregistered adversarial collaboration testing Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT) had a soft-adversarial design that allowed both theories to fail predictions they themselves chose. The collaboration tested biological implementations the consortium flagged as separable from each theory's central claim, named a retreat path before data collection, and accepted predictions overlapping with other theories. Both theories cleared some tests and failed others, with failures concentrated on the imports the theories had committed to in advance.