Covariance identifiability from noise-selected structures: code and retained data

Code and retained data accompanying Covariance identifiability from noise-selected structures by Dekun Yang. The archive contains simulation sources, scalar observations, complete run summaries, symbolic and Lyapunov identity checks, prediction-geometry checks, dimension benchmarks, and figure-generation code. It retains the homogeneous and two-layer outcomes, all bootstrap draws, and the 320-history correlation scan. It also includes the manuscript source and proof-verification guide. Full per-update weight trajectories were not stored. Exact inverse results concern the stated effective law; the learning weak-noise limit is conditional on localization and moment assumptions. OpenAI ChatGPT/Codex assisted mathematical development, writing, coding and computational checks. The author has not yet completed personal step-by-step verification of all core proofs; computational checks are not independent human peer review. Internal historical version labels and predeposit status notes preserve development provenance. Current public-deposit status is supplied by this Zenodo record.