diff --git a/config/training/dqn-smoketest.toml b/config/training/dqn-smoketest.toml
index ec7924ab0..3e5705ed5 100644
--- a/config/training/dqn-smoketest.toml
+++ b/config/training/dqn-smoketest.toml
@@ -21,6 +21,7 @@ hidden_dim_base = 64
 reward_scale = 1.0
 huber_delta = 1.0
 symbol = "ES.FUT"
+max_bars = 60000
 imbalance_bar_threshold = 1.0
 imbalance_bar_ewma_alpha = 0.1
 
diff --git a/crates/ml/src/trainers/dqn/config.rs b/crates/ml/src/trainers/dqn/config.rs
index 022f310b2..99ac0858c 100644
--- a/crates/ml/src/trainers/dqn/config.rs
+++ b/crates/ml/src/trainers/dqn/config.rs
@@ -1218,6 +1218,10 @@ pub struct DQNHyperparameters {
     /// one instrument's DBN files are loaded. Default: "ES.FUT".
     pub symbol: String,
 
+    /// Maximum bars to load for training. 0 = unlimited.
+    /// Truncates dataset after loading, before walk-forward split.
+    pub max_bars: usize,
+
     /// Imbalance bar threshold for MBP-10 to bar conversion.
     /// Higher = fewer bars (less noise), lower = more bars (more signal).
     /// Only used when `data_source = "mbp10"`. Default: 100.0.
@@ -1705,6 +1709,7 @@ impl DQNHyperparameters {
             // Data source: "ohlcv" (backward compatible) or "mbp10" (imbalance bars)
             data_source: "ohlcv".to_string(),
             symbol: "ES.FUT".to_string(),
+            max_bars: 0, // 0 = unlimited
             imbalance_bar_threshold: 100.0,
             imbalance_bar_ewma_alpha: 0.1,
 
diff --git a/crates/ml/src/trainers/dqn/data_loading.rs b/crates/ml/src/trainers/dqn/data_loading.rs
index 2f3db094b..1b57fcb61 100644
--- a/crates/ml/src/trainers/dqn/data_loading.rs
+++ b/crates/ml/src/trainers/dqn/data_loading.rs
@@ -126,7 +126,7 @@ impl DQNTrainer {
                 self.ofi_features = Some(Arc::from(cached.ofi));
             }
 
-            let all_data: Vec<(FeatureVector, Vec<f64>)> = cached.features
+            let mut all_data: Vec<(FeatureVector, Vec<f64>)> = cached.features
                 .into_iter()
                 .zip(cached.targets.into_iter())
                 .map(|(f, t)| {
@@ -134,6 +134,12 @@ impl DQNTrainer {
                 })
                 .collect();
 
+            // Truncate to max_bars if configured (0 = unlimited)
+            if self.hyperparams.max_bars > 0 && all_data.len() > self.hyperparams.max_bars {
+                tracing::info!("max_bars={}: truncating {} → {} bars", self.hyperparams.max_bars, all_data.len(), self.hyperparams.max_bars);
+                all_data.truncate(self.hyperparams.max_bars);
+            }
+
             let split = (all_data.len() * 80) / 100;
             let train = all_data[..split].to_vec();
             let val = all_data[split..].to_vec();
@@ -419,6 +425,12 @@ impl DQNTrainer {
             feature_dim
         );
 
+        // Truncate to max_bars if configured (0 = unlimited)
+        if self.hyperparams.max_bars > 0 && training_data.len() > self.hyperparams.max_bars {
+            tracing::info!("max_bars={}: truncating {} → {} bars", self.hyperparams.max_bars, training_data.len(), self.hyperparams.max_bars);
+            training_data.truncate(self.hyperparams.max_bars);
+        }
+
         // Split training data 80/20 for train/validation
         let split_idx = (training_data.len() * 80) / 100;
         let train_data = training_data[..split_idx].to_vec(); // cpu-side split
diff --git a/crates/ml/src/training_profile.rs b/crates/ml/src/training_profile.rs
index 3109a6c61..3298e8843 100644
--- a/crates/ml/src/training_profile.rs
+++ b/crates/ml/src/training_profile.rs
@@ -92,6 +92,10 @@ pub struct TrainingSection {
 
     /// EWMA alpha for adaptive imbalance threshold. Default: 0.1.
     pub imbalance_bar_ewma_alpha: Option<f64>,
+
+    /// Maximum bars to load for training. 0 = unlimited (load all data).
+    /// Useful for fast smoketests on large datasets.
+    pub max_bars: Option<usize>,
 }
 
 /// Epsilon-greedy exploration parameters (DQN-specific).
@@ -741,6 +745,9 @@ impl DqnTrainingProfile {
             if let Some(ref v) = t.symbol {
                 hp.symbol = v.clone();
             }
+            if let Some(v) = t.max_bars {
+                hp.max_bars = v;
+            }
             if let Some(v) = t.imbalance_bar_threshold {
                 hp.imbalance_bar_threshold = v;
             }