{"id":"75303973-06ac-4890-a433-2889b9803024","arxiv_id":"2504.11982","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A unified prediction-error-minimization framework separates deterministic dynamics from innovation noise and estimates both with L-BFGS-B and automatic differentiation, with consistency guarantees under classical assumptions.","lead":"This paper presents a unified method to identify linear, parameter-varying, and nonlinear state-space models together with noise models using efficient optimization. It reports order-of-magnitude faster training than deep-learning baselines while keeping one-step-ahead prediction accuracy.","discovery_kind":"new_method","skeptic_critique":null,"referee_report":null,"author_rebuttal":null,"desk_editor":null,"rs_alignment":null,"lean_confirmation":null,"pith_extraction":null,"created_at":"2026-08-16T12:41:16.751342+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":null,"supporting_citations":[],"review_version":1}