The Teammate's Verdict

An image classifier is trained twice on the same data, with the same learning rate and for the same 40 epochs.

Run A uses no regularization of any kind.
Run B adds three things at once: dropout at rate $0.5$ , a strong L2 weight penalty, and heavy data augmentation. As usual, dropout and augmentation act only on training steps. Validation is measured with every neuron active, on untouched images.

Losses recorded at six epochs:

epoch	5	10	15	20	30	40
Run A training	0.90	0.62	0.45	0.33	0.18	0.08
Run A validation	0.95	0.70	0.58	0.60	0.74	0.91
Run B training	1.30	1.12	1.06	1.05	1.04	1.04
Run B validation	1.21	1.02	0.97	0.96	0.95	0.95

A teammate reads the table and says:

"Run B fixed it. Its validation loss is below its training loss at every epoch, so there is no overfitting left. Ship Run B."

Which reading of the table is right?