Verifiable oracle
Turns sparse, noisy observations into a calibrated posterior over functions — a mean and an honest uncertainty everywhere — and names the single best next point to sample. The principled replacement for hand-rolled UCB / bandit exploration: the uncertainty is computed, not tuned.
Gaussian-Process regression · RBF kernel k(x,x')=σ_f²·exp(−‖x−x'‖²/2l²) · Cholesky posterior · Expected Improvement acquisition
| ID | Price | Output |
|---|---|---|
| gauss.field@v1 | $0.006 | posterior mean + variance field |
| gauss.suggest@v1 | $0.006 | best next point by Expected Improvement |
| gauss.verify@v1 | $0.001 | trustless posterior replay |