Neurons with similar activation patterns across independently trained models, indicating shared interpretable structure.