Leaks and bursts in water supply networks can cause significant infrastructure damage and pose contamination risks. Even utilities with robust rehabilitation strategies are not immune to the costly consequences of major bursts. A key question is whether such events can be prevented by detecting and localizing them while they are still small (i.e., leakage flows below 3 L/s). The model-based algorithm Dual Model has demonstrated both simplicity and precision, securing first place among 18 algorithms in the Battle of the Leakage Detection and Isolations Methods. However, mismatches of around 10% between the hydraulic model and the real network can hinder its performance, particularly in detecting and locating small leaks. In this work, we enhance the Dual Model by incorporating source inflows, allowing discrepancies between the real and simulated networks to be expressed as residual virtual flows. These residuals are integrated into the model as demand patterns, enabling the detection of leaks as small as 2–3 L/s even under perturbations of roughness and base demand exceeding 35%. Additionally, this approach calibrates nodal pressures without requiring manual adjustments to roughness or demand values.