GCA-BULF: A Bottom-Up Framework for Short-Term Load Forecasting Using Grouped Critical Appliances

This arXiv feed contains multiple 2026 papers across ML, signal processing, energy forecasting, quantum optimization, and applied physics. Notable security-relevant items: (1) “Architecture Determines Observability in Transformers” shows that certain transformer architectures/training recipes lose linear-readout signals used to detect model errors (observability collapse emergent during training), with direct implications for model monitoring, safety validation, and runtime error-flagging. (2) Nautile-370M describes a compact 371M reasoning model combining spectral sequence operators and self‑