Recent large-scale studies over the Great Lakes have demonstrated that long short-term memory (LSTM) models can provide highly skillful streamflow simulations, particularly when coupled with explicit river–lake routing to account for spatial heterogeneity. The Spatially Recursive LSTM (SR-LSTM) framework has shown that applying trained LSTM models at subbasin scales and routing their outputs can improve runoff representation in large and complex basins. Building on these established findings, this contribution focuses on scaling and extending the SR-LSTM framework to a basin-wide Great Lakes application, including the Ottawa River, as an initial implementation using a consistent basin discretization and routing strategy. The emphasis is on developing a reproducible, spatially distributed machine-learning runoff modeling framework for large, interconnected lake–river systems, rather than on re-evaluating previously published model intercomparisons. The objective of this work is to improve basin-wide runoff estimates as a key input to synthesizing and reducing uncertainty in net basin supply and lake water balance components, thereby supporting coordinated Great Lakes water-level monitoring and reporting. This work enforces a consistent basin representation and routing scheme across the domain, outlining a practical way to extend spatially distributed machine learning hydrological models to large transboundary watersheds, with relevance for both research and operational hydrology in the Great Lakes–St. Lawrence River system and beyond.