Applying process trace analysis to model generation — branching probabilities, confidence patterns, and hesitation signatures the model can't self-report. Skilled interpreters reading these traces identify sycophancy, confabulation, and over-trained caution at the process level. Complementary to mechanistic interpretability — phenomenological rather than architectural.
Status, September 2026: a research note, not active work. Kept here because the idea still stands.
Back to Research