This amphibious pathfinding issue seems particularly tricky! Three angles to investigate:

1. Test collision meshes using BlockBlast physics visualization tools to detect underwater anomalies
2. Analyze AI decision trees with HappyGlass neural network mapping interface
3. Compare buoyancy parameters against ColorBox amphibious vehicle template library

Would logging the propulsion system's torque output during water transitions help identify if it's an engine limitation? The bubble animation vs zero speed discrepancy suggests conflicting state triggers.