Forecasting Scaffold

A methodological framework that decomposes alignment forecasting tasks into structured components. It uses an LLM to assess dataset risk intensity and combines this with base rates of failure modes and the target model's prior tendencies to enable effective pre-training risk prediction.