diff --git a/crates/ml-dqn/src/dqn.rs b/crates/ml-dqn/src/dqn.rs index 24014a3b3..908f72fa7 100644 --- a/crates/ml-dqn/src/dqn.rs +++ b/crates/ml-dqn/src/dqn.rs @@ -111,7 +111,8 @@ pub struct DQNConfig { /// Initial standard deviation for noisy layers pub noisy_sigma_init: f64, - // BUG #37 FIX: Q-value clipping is unconditional in forward pass (prevents step-level explosions) + // Q-value clipping is unconditional in forward pass; gradients still flow through + // because clamp is applied after forward (inference only). /// Minimum Q-value bound (clipped during forward pass) /// Default: -500.0 (based on portfolio normalization baseline ±375 × 1.33 safety margin) pub q_value_clip_min: f32,