Using device: cuda:0

============================================================
Seed 456
============================================================

--- Credit Bridge ---
  [CB] Warmup phase: 6 epochs (DFA fallback + value net training)
  [CB] Epoch 1 (warmup): loss=1.9934, train=0.2765, test=0.3431, vloss=0.635836
  [CB] Epoch 10 (blend=0.67): loss=1.9165, train=0.3182, test=0.3451, vloss=0.059035
  [CB] Epoch 20 (blend=1.00): loss=1.8877, train=0.3323, test=0.3456, vloss=0.025170
  [CB] Epoch 30 (blend=1.00): loss=1.8695, train=0.3410, test=0.3562, vloss=0.015811
  Final test acc: 0.3562

All results saved to results/round38_cb_penalty_30ep_s456/results_cifar10.json