This paper presents a coherence-based diagnostic framework for backward reconstruction of deep-time Earth system histories. The approach integrates heterogeneous paleoclimate records into a minimal latent state and evaluates reconstruction consistency using coverage-weighted coherence metrics, smoothness regularization, and weak admissibility constraints. A fully reproducible Cenozoic demonstration (0–5.3 Ma) recovers canonical climate features while maintaining interpretable uncertainty under proxy sparsity. The framework is positioned as a synthesis and diagnostic layer that complements, rather than replaces, mechanistic Earth system models, and is designed to scale to deeper Phanerozoic timescales.
Peter Brunzelle (Sun,) studied this question.