Enhanced Log-GP

Agnostic methods are better suited than physics-only TAS-AI for unknown global response mapping in blind reconstruction benchmarks. Enhanced Log-GP was competitive with grid sampling in the PySpinW gapless case and best in the gapped case, while physics-only TAS-AI performed poorly for blind whole-window mapping. Enhanced Log-GP has higher benchmark-harness planner-side costs than TAS-AI physics planning. The agnostic phase uses Gaussian-process regression in log-intensity space for neutron spectroscopy reconstruction. The enhanced Log-GP policy adds initialization, variance weighting, consumed-area exclusion, and an energy taper to the base approach.