Traversal-as-Policy: Log-Distilled Gated Behavior Trees as Externalized, Verifiable Policies for Safe, Robust, and Efficient Agents

Collection of new machine-learning research (arXiv cs.LG) covering methods for safer, more efficient autonomous LLM agents (Traversal-as-Policy: log-distilled Gated Behavior Trees that externalize policies and dramatically reduce safety violations), stability and regularization for operator/surrogate models (JAWS: spatially-adaptive Jacobian regularization), a continuously-updating video data construction system with provenance (VDCook), integer-sequence modeling (IntSeqBERT using magnitude + modulo-spectrum embeddings), a stochastic-process analysis of time-series decision making, continuous‑