The paper develops a method to assess drift between target and model distributions using noisy measurements at a few points.
problem Evaluating the difference between a target and model distribution when only a few noisy measurements are available.
method The paper introduces a method based on finite-probe total-variation certificates for drifting models, using antisymmetric interactions and absolutely continuous laws in a finite density basis.
result The method provides a nonpositive observability margin that returns the trivial TV bound and abstains when the margin is nonpositive, otherwise it computes a TV upper confidence bound.