**Adversarial Stress Test Suite (STS v1.0): Track‑C Audit Report and Failure Frontiers for Z48 Holonomy Inference** **Abstract.** We present an adversarial Stress Test Suite (STS v1.0) for Track‑C, a correlated‑structure inference protocol that models a hidden discrete clock \(K_t \in \mathbb{Z}_{48}\) with a carry‑coupled mixed‑radix decomposition \(K=b+6r\) (the \(M_6\) hypothesis). The primary detector is Bayesian model comparison via \(\Delta \log \mathcal{Z}\) against hard null generators (\(M_{\mathrm{split}}, M_{\mathrm{wrong}}, M_{\mathrm{small}}, M_{\mathrm{rand}}\)). Patch‑A quantized leakage is included as an emission mixture \(p=\sin^2(\pi/24)\) and a transition leakage \(\varepsilon=\tan^2(\pi/24)\). We report failure frontiers under three adversarial families: (S2) correlated noise (AR(1) temporal correlation \(\rho_t\) and common‑mode drift \(\rho_g\)), (S6) Faraday rotation with biased RM correction (single‑band and dual‑band), and (S4) model misspecification in the emission map (nonlinear warping and slow drift in gate couplings). The suite yields interpretable winner‑at‑failure modes (e.g., confusion with smooth‑walk or wrong‑mapping nulls) and quantifies robustness gains from Patch‑A via ablation. The STS is intended as an audit artifact to reduce inverse‑crime risk, expose failure mechanisms, and provide preregisterable falsifier hooks for future real‑data tests.