Identifiability Without Gaussianity: Symbolic World Models and Near-Infinite Temporal Consistency

Collection of new ML/statistics arXiv submissions (June 12, 2026) covering theoretical and applied advances: a Physics-Grounded Symbolic Architecture (PGSA) proving identifiability and near-infinite temporal consistency (with Lean4/Mathlib4 formalization); a critique and reclassification of epistemic uncertainty into aleatoric, sample-reducible and mechanism-reducible types; semi-supervised prediction-powered causal inference (DML-PPCI) with Riesz regression; robust feature-weighted jump models for temporal clustering; ProtoX-AD, a prototype-based self-explainable time-series anomaly detector;