Abstract High‐resolution climate simulations are essential to study sub‐daily extreme precipitation, yet their computational demands make long‐term large‐ensemble simulations over a large geographical region often infeasible. We introduce a case‐selective dynamical downscaling (CSDD) framework that reconstructs extreme precipitation statistics at convection‐permitting resolution by simulating only periods when extreme rainfall occurs, rather than performing continuous simulations. Using low‐resolution precipitation as a predictor, we identify and downscale time windows associated with extreme events. Applied to a 30‐year regional climate simulation, CSDD reproduces the statistical distribution of 1–6‐hourly precipitation extremes from a full continuous convection‐permitting simulation at roughly 10 % $%$ of the computational cost. Because cases are independent, they can be executed in parallel, enabling substantial wall‐time reductions. For applications targeting extreme precipitation, notably climate storylines, CSDD provides a physically grounded and computationally efficient way to supplement storylines with reliable extreme‐value statistics, bridging storyline and statistical approaches to climate extremes.