AI planners usually require an accurate description of the planning task to obtain a solution that achieves the goals of the problem. However, generating such descriptions can be time-consuming and error-prone, often resulting in unsolvable planning tasks. Planners lack the ability to identify semantic errors or explain how to solve them. Previous works offer repaired initial states/domains to make the planning task solvable. Still, they are limited to fixing things for a single problem instance or rely on input plan traces. In this work, we address the reparation of a flawed domain only based on a set of planning instances viewed holistically. By obtaining individual repairs for each problem, we search in the space of models for a repaired one that covers the full problem set. Our experimental results show the effective application of this approach in repairing a lifted planning domain and establish metrics that quantify the effort required to obtain the ground truth repair through an anytime algorithm.