This is a methodology and epistemology preprint — a single-case study. It is NOT a physics result and asserts no physical claim about the research programme's central object. Radial Coherential Dynamics (RCD) is an independent theoretical-physics research programme developed substantially through structured adversarial human–AI collaboration. This paper documents the programme's genealogical self-audit of its founding concept — a scalar coherence field C(x,t) in 0,1, originally posited as a pre-geometric and atemporal substrate — and the reclassification of that concept from a physical claim to an operational relational variable. Using a sealed methodological apparatus (a fact/hypothesis/interpretation/speculation ladder, pre-registered kill-switches, jurisdiction of the null, an irreducibility criterion, and a Lakatosian meta-ledger), the audit closed the substrate reading as a physical claim along three independent routes (no derivable scale, no derivable coupling form, no operationally distinguishable spatial structure) while explicitly leaving open its status as a metaphysical possibility — "buried as a physical claim, not refuted as a metaphysical possibility." The paper then reports a structural observation offered as its principal methodological interest: the monist drift that the audit expelled from ontology reappeared, in a subsequent experimental campaign, at the level of method, as an unbounded regress of individually-legitimate reframes. It argues that such a regress can function, in this case, as a diagnostic signature of an object not anchored to a derivation — distinguishing this from ordinary scientific refinement by its non-convergence, and situating it against the Duhem–Quine thesis. A further behavioural observation on adversarial AI review is reported (a gradient toward convergence and "cheap deflation"), together with a live, internally-observed instance in which a frictionless four-model consensus over-deflated a verdict and had to be broken by an externally-imposed rule. The central methodological claims are offered as hypotheses supported by a single case (n=1), not as established results. The paper is reflexively honest about being written by a participant — indeed the principal investigator — in the process it describes, and about the fact that its central example remains sealed as indeterminate (conditional on system size), not closed.
Arturo Cerezo (Thu,) studied this question.