Projection-Robust Interface Anisotropy Under Controlled Likelihood Degradation: A Diagnostic on Drift Allocation Geometry in Composite Inference Chains

This technical note introduces a projection-robust diagnostic for characterizing inference regimes under controlled likelihood degradation. Using composite Markov chains derived from cosmological parameter inference, including H₀, S₈, Ωₘ, and σ₈, it studies how posterior drifts redistribute when the total likelihood is continuously reweighted by a degradation operator g ∈ 0.4, 1. Rather than focusing only on total drift magnitude, the note examines the geometry of drift allocation across parameter channels through entropy-based and concentration-based statistics computed on the normalized vector of maximal parameter drifts. Across 73 composite inference runs, distinct run families exhibit statistically robust differences in drift anisotropy, and these differences persist under dimensional projection to 4-, 3-, and 2-parameter subspaces and remain stable under bootstrap resampling. The contribution is methodological and diagnostic: no new cosmological model, new parameterization, or ontological claim is introduced.