deployment-memory

This paper introduces a behavioral evaluation framework for calibrating claims about deployment-time memory in LLM test-time training, proposing an evidence ladder and explicit baselines to bridge proxy metrics and behavioral evidence.