Base Ari
Held-out reward rose from 0.188 to 0.228 across 8 training runs
Best so far
No improvement
Not evaluated
Base Ari
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Held-out eval reward
1
2
3
4
5
6
7
8
Training run
Baseline 0.188
Baseline
Move check
FEN board
16k
Longer run
SFT→GRPO
Smoke test
SFT stage