A hybrid cognitive-state modeling layer maintains evolving representations of the user across sessions using two complementary signals.
Explicit interpretable indicators are derived from measurable interaction traces — engagement trajectory, repetition and uncertainty patterns, reflective-depth signals, and cognitive-overload markers — supporting transparency and trust calibration. Latent cross-session embeddings, inspired by persistent user-conditioning research, capture behavioral rhythms and progression that handcrafted features miss. Together they make conditioning both legible and expressive.
This is a short thread — a fuller write-up is coming soon.