diff --git a/crates/ml/src/cuda_pipeline/backtest_env_kernel.cu b/crates/ml/src/cuda_pipeline/backtest_env_kernel.cu
index 62ea60d83..6d5998338 100644
--- a/crates/ml/src/cuda_pipeline/backtest_env_kernel.cu
+++ b/crates/ml/src/cuda_pipeline/backtest_env_kernel.cu
@@ -261,14 +261,22 @@ extern "C" __global__ void backtest_env_step(
     // Re-encode position -> exposure_idx -> factored action to ensure metrics
     // kernel counts real position changes, not model-requested actions.
     {
-        float f_actual_frac = position / fmaxf(max_position, 0.01f);
-        int actual_exp_idx = (int)roundf((f_actual_frac + 1.0f) * 0.5f * (float)(b0_size - 1));
-        if (actual_exp_idx < 0) actual_exp_idx = 0;
-        if (actual_exp_idx >= b0_size) actual_exp_idx = b0_size - 1;
-        // Preserve original order/urgency from model action
-        int orig_order = (b1_size > 0 && b2_size > 0) ? (action_val / b2_size) % b1_size : 0;
-        int orig_urgency = (b2_size > 0) ? action_val % b2_size : 0;
-        int actual_action = actual_exp_idx * b1_size * b2_size + orig_order * b2_size + orig_urgency;
+        /* Re-encode actual position as 4-branch factored action.
+         * Recover direction and magnitude from position, preserve order/urgency. */
+        float pos_sign = position;
+        int actual_dir = (pos_sign > 0.001f) ? 2 : (pos_sign < -0.001f) ? 0 : 1;
+        float abs_pos = fabsf(position) / fmaxf(max_position, 0.01f);
+        int actual_mag;
+        if (abs_pos < 0.375f) actual_mag = 0;       /* ~0.25 -> Quarter */
+        else if (abs_pos < 0.75f) actual_mag = 1;   /* ~0.50 -> Half    */
+        else actual_mag = 2;                         /* ~1.00 -> Full    */
+        if (actual_dir == 1) actual_mag = 0;         /* Flat: magnitude irrelevant */
+        int orig_order   = decode_order_4b(action_val, b2_size, b3_size);
+        int orig_urgency = decode_urgency_4b(action_val, b3_size);
+        int actual_action = actual_dir * b1_size * b2_size * b3_size
+                          + actual_mag * b2_size * b3_size
+                          + orig_order * b3_size
+                          + orig_urgency;
         actions_history[w * max_len + current_step] = actual_action;
     }
 }