Expand description
The synthetic user (the reference: Milestones, the M3 gate).
A ground-truth taste (θ*, τ*, cuts*) that generates noisy feedback exactly per the observation model. The gate tests assert the posterior recovers θ* and predicts held-out feedback — making the taste core falsifiable with no UI and no human. Later this doubles as demo mode (“watch it learn a fake user in fast-forward”).
Synthetic φ live directly on the model’s scale (unit normals), so the
observations these emit carry no feature names and are fed to the model
through crate::FitSet::as_is rather than being re-standardized.
Structs§
- Ideal
Point User - A simulated user with an ideal point: there is a sound they are looking for, and both too little and too much of any quality is worse.
- Mixture
Synthetic User - A simulated user whose taste has several islands: true utility is the max over component tastes (“I love a great drone OR a great pluck”). A single linear θ provably cannot represent this — it is the ground truth for the K > 1 mixture gate.
- Synthetic
User - A simulated user with fixed ground-truth taste.
Functions§
- cosine
- Cosine similarity between two vectors (θ-recovery metric).