1\section{Robustness}2\label{sec:robustness}3Held-out prompt families give hit rates within 0.01 of the fit families at every capacity (v1 traces), and the transition4predictor's recall drops from 0.53 (fit) to 0.50 (held-out), so the locality statistics are not an artefact of the fitting5prompts. Throughput measurements on the laptop vary by less than 1\,\% across repeats for the all-resident configurations and by6up to 40\,\% across prompt families for demand loading (code prompts route more repetitively than general prose, hence higher7hit rates). Thermal state and competing processes are not yet controlled in the smoke tests reported here; the final sweeps8sample \code{powermetrics} thermal pressure every second and discard runs under throttling.9