diff --git a/crates/ml/src/trainers/dqn/trainer/enrichment.rs b/crates/ml/src/trainers/dqn/trainer/enrichment.rs
index 5505566d2..f704ec399 100644
--- a/crates/ml/src/trainers/dqn/trainer/enrichment.rs
+++ b/crates/ml/src/trainers/dqn/trainer/enrichment.rs
@@ -48,7 +48,7 @@ pub(crate) struct HindsightExperience {
 /// Output of one enrichment cycle.
 #[derive(Debug, Clone)]
 pub(crate) struct EnrichmentResult {
-    /// E1: additive correction to Q-targets (clipped ±2.0).
+    /// E1: additive correction to Q-targets (clipped ±10.0).
     pub q_correction: f32,
     /// E2: next-epoch epsilon.
     pub epsilon: f32,
@@ -148,7 +148,7 @@ fn compute_q_correction(trades: &[EvalTrade]) -> f32 {
         .map(|t| t.predicted_q - t.pnl)
         .sum::<f32>()
         / trades.len() as f32;
-    (-bias).clamp(-2.0, 2.0)
+    (-bias).clamp(-10.0, 10.0)
 }
 
 /// E2: Adaptive epsilon — performance-driven exploration rate.