A framework-agnostic Python library that watches your model training loop and gives plain-English coaching advice — overfitting, plateaus, bad learning rates, class imbalance, and divergence, all detected from simple metric history. Zero ML dependencies, 280+ tests.