<p>Resource estimation in the mining context is primarily achieved using kriging. Kriging is a linear unbiased estimator that minimizes the error variance using a model of the spatial correlation. Recoverable resources are sensitive to the variability of the estimated attributes at a given volume that depends on mining selectivity. The expected variability of a resource variable for a selective mining unit (SMU) can be calculated based on the point scale variability and a change of support (COS) model. Kriging estimates are smooth; in particular, they are smoother than they should be with widely spaced data. The smoothness depends on many factors such as data spacing and configuration, SMU size, and spatial continuity model. Commonly, the kriging plan calls for a restricted search near each estimation location to avoid over smoothing. The estimates are more variable when the search restricts the number of data used in kriging. Restricting the kriging search plan to achieve the expected variability at the SMU scale is the main method of generating estimates that can be used for recoverable resources. Note that no restrictions are required for final grade control estimates. Defining the search plan is a laborious task that usually involves testing multiple scenarios including multiple estimation passes, each with different search restrictions. The local search optimization (LSO) method proposed in this article is aimed at dynamically restricting the search at the time of estimation for each estimation location while targeting the expected variance of the estimates. This method greatly simplifies the calibration process and often results in better estimation performance in densely sampled areas as it does not over-penalize the estimation by using too few data.</p>