The NCAA model improved after I split the problem into seed zones

The NCAA model improved after I split the problem into seed zones

The model got more stable after seed-zone routing replaced one-size-fits-all estimation.

This setup uses top, mid, and low band ensembles, routing by rank and bid type, then assignment optimization for final slots.

What mattered most was consistency in data and process, not only model flavor. Stable scripts helped more than aggressive tweaks.

Current status is synced repo updates with a clearer model narrative. GitHub: https://github.com/pbathuri/Kory_The_Cat-NCAA