Correcting the lens changes the result: a proxy-fidelity audit framework for consciousness science

Abstract

Consciousness science routinely operationalizes theoretical constructs — integration, complexity, coupling, level — as measurable proxies, and then tests hypotheses about the constructs through the proxies. This chain is only as strong as its weakest translation. We present a post-execution proxy-fidelity audit framework, developed and applied across 18 consecutive research cycles of a single-investigator EEG program on public datasets (57 cycles total; the audit was institutionalized at cycle 42 and ran in every cycle since). The audit re-checks each construct-proxy resolution against four documented failure modes: F1 — temporal dimension (a dynamic construct read through a static proxy); F2 — field referential (a field phenomenon read as the coupling of two poles); F3 — measurement family (the proxy belongs to a different measurement family than the construct's original prediction); F4 — ruler-by-tissue (a signal read with the conventions of another tissue or domain). Each mode has a documented precedent in which correcting the lens changed — or legitimately failed to change — the verdict: F1, a static scalar proxy of a dynamic field construct attenuated an observed correlation that a dynamic re-operationalization could not rescue — the null was real, but only the audit separated no-signal from wrong-dimension; F2, a cross-multifractal coupling spectrum measuring two poles failed to close a brain-body loop that a totality observable closed; F3, an entropic prediction translated into a geometric dimension estimator failed while the family-corrected estimator confirmed it, and a Gaussian-copula O-information blind to non-Gaussian synergy was corrected by a non-parametric estimator; F4, body channels appended to the brain's covariance ruler degraded tracking until tissue-specific readers restored it. A fifth case shows the audit's other face: a proxy survived every correction, and the null verdict was defended against post-hoc lens-swapping. We distill the protocol into eight recommendations — pre-registered construct-proxy tables, family-matched translations, dimension checks, totality observables for field constructs, tissue-specific rulers, null-verdict protection, audit trails, and disclosure of proxy-dependence — and argue that proxy-fidelity auditing should be a standing stage of analysis in consciousness science, complementing statistical rigor with translational rigor.