We develop an empirical d-a posteriori approach to estimations with uniformly minimal d-risk, when the a priori distribution is completely unknown. For a scalar parameter of a discrete exponential family we construct ...