This page tests whether protein-level AR(2) eigenvalue |λ| is confounded by protein half-life. AR(2) is fitted directly from the Otobe-Yoshitane 2026 mouse liver whole-cell proteomics time-series (106 genes, CT2/CT6/CT10/CT14/CT18/CT22). Spearman ρ between protein half-life and protein |λ| = −0.404 (10,000-shuffle permutation p < 0.0001; bootstrap 95% CI [−0.58, −0.25]). The correlation is negative: longer-lived proteins have lower |λ|. This is the opposite direction to a stability confound, which would predict ρ > 0. The most stable proteins — ribosomes, glycolytic enzymes (half-life 50–220h) — are constitutively expressed with flat abundance profiles and lower |λ|. The most rapidly degraded proteins — clock components (PER1 t½=1.5h), cell-cycle regulators (CCNB1 t½=0.7h) — are rhythmically expressed and show high |λ|. The causal driver is transcriptional regulation, not protein stability. Limitation: 6 timepoints per gene compress all rhythmic proteins to |λ| ≈ 0.999, preventing resolution of fine-grained differences within the rhythmic class. The clock-target hierarchy gap (−0.014, label-permutation p = 0.932) is not significant at this series length — replication requires ≥12 timepoints. The stability objection is refuted by the direction of the correlation, but the protein-level hierarchy result awaits a longer time-series dataset.