diff --git a/crates/ml/src/hyperopt/adapters/dqn.rs b/crates/ml/src/hyperopt/adapters/dqn.rs
index ae13fa3f3..113b64e29 100644
--- a/crates/ml/src/hyperopt/adapters/dqn.rs
+++ b/crates/ml/src/hyperopt/adapters/dqn.rs
@@ -2084,7 +2084,7 @@ fn calculate_diversity_penalty(action_distribution: &[f64; 3]) -> f64 {
         .collect();
 
     let entropy = calculate_action_entropy(&*action_counts);
-    let max_entropy = (9.0_f64).log2(); // ~3.170
+    let max_entropy = (3.0_f64).log2(); // ~1.585 (3 buckets: BUY/SELL/HOLD)
 
     // Smooth quadratic penalty: -5.0 * (1 - entropy/max_entropy)²
     // At entropy=0: -5.0, at entropy=max: 0.0
@@ -4237,12 +4237,12 @@ mod tests {
         assert!(penalty_mid > -1.0, "Medium entropy mild: {}", penalty_mid);
 
         let penalty_uniform = calculate_diversity_penalty(&[0.33, 0.33, 0.34]);
-        assert!(penalty_uniform.abs() < 0.1, "Uniform ~0: {}", penalty_uniform);
+        assert!(penalty_uniform.abs() < 0.1, "Uniform 3-bucket ~0: {}", penalty_uniform);
 
-        // Monotonic
-        assert!(penalty_uniform > penalty_mid);
-        assert!(penalty_mid > penalty_low);
-        assert!(penalty_low > penalty_zero);
+        // Monotonic: uniform > mid > low > zero (less negative = better)
+        assert!(penalty_uniform > penalty_mid, "uniform > mid: {} > {}", penalty_uniform, penalty_mid);
+        assert!(penalty_mid > penalty_low, "mid > low: {} > {}", penalty_mid, penalty_low);
+        assert!(penalty_low > penalty_zero, "low > zero: {} > {}", penalty_low, penalty_zero);
     }
 
     #[test]