Quotienting Markov chains by their peripheral invariant subspace separates persistent regime profiles from transient dynamics for stable policy evaluation.
Truncated signature inversion is reframed as learning conditional path distributions via signature-conditioned flow matching, with derived Bayes error baselines and validated reconstruction on real data.
AgentForesight introduces online trajectory auditing that predicts multi-agent failures during execution, with a 7B model outperforming GPT-4.1 and DeepSeek-V4-Pro by up to 19.9% with 3x lower step error.