Mesoscale eddies are the ocean’s primary reservoir of kinetic energy, yet most climate models cannot fully resolve them due to computational limits. In this study led by David Kamm, two machine-learning–based eddy parameterizations, Zanna and Bolton (2020) parameterization (ZB20) and Guillaumin and Zanna (2021) parameterization (GZ21), are implemented in the NEMO ocean model and evaluated against high-resolution simulations. While GZ21 shows systematic biases linked to grid spacing and does not improve coarse-resolution performance, ZB20 successfully captures subgrid energy transfers, leading to improved kinetic energy spectra and large-scale circulation. The results highlight that carefully designed, resolution-aware training data are essential for developing robust and generalizable data-driven eddy parameterizations.