Resolution Cost of Structured Inquiry States Under a SAT Kernel

This preregistration governs Run B of a three-phase experiment measuring the physical energy cost of resolving structured question-states encoded as boolean satisfiability instances. A question-state over the six interrogative dimensions (WHO, WHAT, WHERE, WHEN, HOW, WHY) is encoded in CNF so that interrogative entropy equals the free-variable count, making the solver's work the resolution of the question rather than a stand-in for it. Run A (July 8, 2026) executed the full protocol and is disclosed here without omission: its negative control detected GPU-channel contamination, a CPU-only recomputation quantified the contamination at roughly two thirds of the shipped combined energy, and the clean channel revealed one duration-independent physical signal, a significant negative correlation between solver conflicts and mean CPU package power, registered here as prediction P-W. Run B is a confirmatory replication under corrected instrument conditions, bound to a machine-state technical annex and a sha256 code manifest. All predictions, falsification conditions, analysis procedures, and the claim ceiling are fixed in advance.