Learning engine
Learning engine
Section titled “Learning engine”The app and optional cloud backend share a single deterministic TypeScript learning engine. It implements Bayesian Rasch/1PL ability estimation, written-exposure weighting, FSRS-6 per-word memory, adaptive Practice selection, a short-term challenge controller, rank hysteresis, and alignment-aware spelling correction.
The same ordered evidence and model version must produce the same result on mobile and server. Short-term session challenge state must not be mistaken for durable ability.
Complete reference: Learning engine specification · Mobile components · API components.
Status: planned, not implemented.