diff --git a/emulmps/emulmps_emul/emulmps_w0wa.py b/emulmps/emulmps_emul/emulmps_w0wa.py
index 836e0a8..420c8d6 100644
--- a/emulmps/emulmps_emul/emulmps_w0wa.py
+++ b/emulmps/emulmps_emul/emulmps_w0wa.py
@@ -457,7 +457,7 @@ def _compute_boost_approximation(self, Pk_lin, params: np.ndarray) -> np.ndarray
             return_boost=True,
             Plin_in=Pk_lin * h**3,
         )
-        return boost / h**3
+        return boost
 
     def _predict_fracs_all_z(self, params_norm: np.ndarray) -> np.ndarray:
         """Linear model: normalised params -> log-frac for all z."""
@@ -592,6 +592,15 @@ def get_boost(
 
         boost = (np.exp(self._predict_nl_fracs_all_z(x_norm)) * syren_boost).astype(np.float32)
 
+        k_t = 0.005  # [1/Mpc]
+        n   = 2.0   # 1 = pure exponential, larger = sharper transition
+        self._lin_to_nl_weight = (
+            1.0 - np.exp(-(self.K_MODES / k_t)**n)
+        ).astype(np.float32)
+
+        self._k_lin_mask = self.K_MODES < k_t
+        boost = 1.0 + (boost - 1.0) * self._lin_to_nl_weight
+
         return self.K_MODES, self.Z_MODES, boost
 
     def has_nl_model(self) -> bool: