Pseudo-Observation Batch Selection

Constant Liar and Kriging Believer perform well empirically only when the surrogate is a Gaussian Process. Constant Liar and Kriging Believer build batches by adding synthetic observations after each selected point. Fantasy models use the same pseudo-observation principle but sample synthetic observations from the predictive distribution. Kriging Believer leaves the posterior mean unchanged because its lie value equals the current predictive mean. Constant Liar variants add mean shifts that help explain differences between CL-min, CL-max, and Kriging Believer.