Bridge Paper M: Chemical and Neural Normalization vs. WCM-PL / LLM Normalization — Two Closed Negative Results and the Retirement of LLM-11

This Bridge Paper closes two lines of inquiry. Part I resolves the AUDIT-4 carryover concerning LLM-11 ("the grokking training-dynamics timeline is structurally isomorphic to the clinical course of schizophrenia"): connection_audit finds it REFUTED, inheriting the falsification independently established for LLM-23 this session. Part II audits whether biological divisive normalization shares structural grounding with WCM-PL/LLM normalization (RMSNorm, LayerNorm) beyond surface algebra. Three sub-claims are tested. Brain-Norm-2 (an "outlier-preservation" analogy) is REFUTED on mechanism-level inspection. Brain-Norm-1 (a general gain-control template match) is resolved via an original derivation: applying Bucher and Brandenburger's (2022, PNAS) efficient-coding logic to RMSNorm yields a necessary and sufficient condition (isotropic input direction) that is directly falsified by the well-documented anisotropy of transformer representations. A mid-session methodological error (an overclaimed "independent replication" via GDN layers) is also reported and corrected. No STRUCTURAL cross-domain finding is reported; all three sub-claims resolve to REFUTED, recorded here with the same care as positive results.