Measuring Behavioral Drift in LLMs: 22 Signals, 5 Dimensions, and the Calcification Effect

How we borrowed LIWC, OCEAN, and VAD from behavioral science to build a reproducible framework for quantifying personality shift in AI agents - and what the early data revealed that we didn’t expect. In Article 1 of this series, we established what LLM Drift is and why it matters. In Article 2, we walked through the LangGraph architecture that powers our adversarial debate engine.