diff --git a/docs/figures/example05/fig06_hybrid_decoding_summary.png b/docs/figures/example05/fig06_hybrid_decoding_summary.png
index 4d0bea4..519f66d 100644
Binary files a/docs/figures/example05/fig06_hybrid_decoding_summary.png and b/docs/figures/example05/fig06_hybrid_decoding_summary.png differ
diff --git a/nstat/decoding_algorithms.py b/nstat/decoding_algorithms.py
index 8c7b783..15bcd51 100644
--- a/nstat/decoding_algorithms.py
+++ b/nstat/decoding_algorithms.py
@@ -1993,6 +1993,26 @@ def PPHybridFilterLinear(
         pNGivenS = np.zeros((n_models, num_steps), dtype=float)
         S_est = np.zeros(num_steps, dtype=int)
 
+        # Fuse initial state prior with terminal constraint for each model
+        # (Srinivasan et al. Eq. 2.23).  Matches PPDecodeFilterLinear and
+        # the corrected MATLAB PPHybridFilterLinear.  Without this step the
+        # goal-directed filter starts from the raw prior, never incorporating
+        # target information into x0/Pi0.
+        for s in range(n_models):
+            if _has_target[s] and estimateTarget == 0:
+                dim = state_dims[s]
+                Pi0s = Pi0_models[s]
+                x0s = x0_models[s]
+                det_Pi0 = np.linalg.det(Pi0s)
+                if det_Pi0 != 0.0:
+                    invPi0s = np.linalg.pinv(Pi0s)
+                    invPitTs = np.linalg.pinv(PitT_m[s][:, :, 0])
+                    Pi0New = np.linalg.pinv(invPi0s + invPitTs)
+                    Pi0New = np.where(np.isnan(Pi0New), 0.0, Pi0New)
+                    x0New = Pi0New @ (invPi0s @ x0s + invPitTs @ PhitT_m[s][:, :, 0] @ yT_models[s])
+                    x0_models[s] = x0New
+                    Pi0_models[s] = Pi0New
+
         fit_type = str(fitType)
 
         for time_index in range(num_steps):
@@ -2031,8 +2051,13 @@ def PPHybridFilterLinear(
             for model_index in range(n_models):
                 dim = state_dims[model_index]
                 if _has_target[model_index]:
-                    A_t = B_m[model_index][:, :, time_index]
-                    Q_t = QT_m[model_index][:, :, time_index]
+                    # Use B(:,:,0) and QT(:,:,-1) — matches the original
+                    # Srinivasan et al. implementation where the prediction
+                    # uses the initial modified dynamics B_1 with the terminal
+                    # modified noise covariance QT_N for consistent goal
+                    # correction across all time steps.
+                    A_t = B_m[model_index][:, :, 0]
+                    Q_t = QT_m[model_index][:, :, -1]
                 else:
                     A_t = _select_time_matrix(A_models[model_index], time_index, dim)
                     Q_t = _select_time_matrix(Q_models[model_index], time_index, dim)