Empirical Validation of the Consciousness Potential Framework: Developmental Trajectories in Cultured Cortical Networks and Clinical Group Differences in Dementia

This paper reports empirical tests of the Consciousness Potential framework on neural recordings spanning in-vitro cortical development, resting clinical EEG, and within-subject anesthetic state transitions. Phi as computed here is a four-factor product (interaction density, organization, recurrent feedback, and spectral complexity), a rate rather than a dimensionless index; its fourth factor, spectral complexity, replaces the theory paper's provisionally-retired persistence factor. The framework's most defensible empirical signals are both carried by the recurrent-feedback channel, and both are immune to the interaction channel's proxy. Across three anesthetic mechanisms in two animals, recurrent feedback drops on every waking-to-anesthetized transition and returns to baseline in recovery; and within a single clinical cohort it recovers the established mild-to-moderate hypersynchrony decline, with a suggestive severe reversal (n = 3), that an earlier saturating estimator could not see. Both are sample-limited (two animals; one cohort with three severe cases), and neither is a cross-cohort clinical replication, but as freshly derived, proxy-immune signals they are the paper's firmest footing. In resting clinical EEG, the confirmed cross-cohort signature narrows to a two-component direction: interaction reduced and organization increased, at an inverse-variance pooled |d| of about 0.43 across the two retained cohorts (n = 2 cohorts; the 95% CI excludes zero but is wide). The organization direction is proxy-immune; the interaction direction is computed on an uncapped graph-weight proxy and is provisional pending recomputation under the capped definition. The remaining components do not generalize across cohorts. Stated plainly, the composite scalar carries no discriminative advantage over band power, and the framework's clinical content is a two-component direction rather than a scalar classifier. In cultured cortical networks, the consciousness index rises and then falls across development, a non-monotonic trajectory that a measure tracking activity or connectivity alone would not produce. The rise-and-fall itself is robust: it survives with the interaction channel removed, carried by the organization decline. The location of the peak, and therefore its alignment with the independently described shift from complex to stereotyped bursting, is driven by the interaction channel and is provisional pending recomputation under the capped definition of that channel. Recurrent feedback is defined as returning mutual information; the normalized spectral radius reported here is its linear sub-critical estimator, and every reported value lies strictly inside that estimator's licensed range.