diff --git a/crates/ml/src/trainers/dqn/trainer/training_loop.rs b/crates/ml/src/trainers/dqn/trainer/training_loop.rs
index 665dc085e..a1813db8e 100644
--- a/crates/ml/src/trainers/dqn/trainer/training_loop.rs
+++ b/crates/ml/src/trainers/dqn/trainer/training_loop.rs
@@ -1872,9 +1872,9 @@ impl DQNTrainer {
         let active_factored = active_dirmag * active_ord * active_urg;
         let diversity_pct = (active_factored as f64 / total_factored_space as f64) * 100.0;
 
-        // Max reachable dir*mag = (b0-1)*b1 + 1: Short and Long each have b1 mag options.
-        // Flat forces mag=Half, so only 1 of the 3 Flat cells is reachable.
-        let max_dirmag = (b0.saturating_sub(1)) * b1 + 1; // 2*3+1 = 7 (of 9 total)
+        // Max dir*mag = b0 * b1 = 9 (full combinatorial space).
+        // Epsilon exploration can hit any combo including Flat×{Small,Full}.
+        let max_dirmag = b0 * b1;
         info!(
             "Epoch {}/{}: Action diversity={}/{} ({:.1}%) — dir*mag={}/{} order={}/{} urgency={}/{}",
             epoch + 1, self.hyperparams.epochs,