Files
foxhunt/crates
jgrusewski 3a8a756332 feat: adaptive CV readiness — relative improvement from initial CV
readiness = (cv_initial - cv_current) / cv_initial

No fixed threshold. Captures initial CV at step 1, measures fractional
improvement. Fully domain-adaptive: works for any batch size, reward
scale, or Q-value magnitude.

Result: fold 2 Sharpe 2.47 → 16.15, peak epoch 18 → 42/50.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-13 20:02:02 +02:00
..