diff --git a/crates/ml/src/cuda_pipeline/c51_grad_kernel.cu b/crates/ml/src/cuda_pipeline/c51_grad_kernel.cu
index 15a5b03fa..4dab245f9 100644
--- a/crates/ml/src/cuda_pipeline/c51_grad_kernel.cu
+++ b/crates/ml/src/cuda_pipeline/c51_grad_kernel.cu
@@ -26,7 +26,8 @@ extern "C" __global__ void c51_grad_kernel(
     const float* __restrict__ branch_scales,    /* [B, 4] per-sample per-branch gradient scale */
     const float* __restrict__ per_sample_support, /* [B, 3] per-sample [v_min, v_max, delta_z] */
     const float* __restrict__ liquid_mod,        /* [4] pinned device-mapped per-branch modulators */
-    const float* __restrict__ atom_positions)    /* [4, num_atoms] adaptive positions. NULL = linear. */
+    const float* __restrict__ atom_positions,    /* [4, num_atoms] adaptive positions. NULL = linear. */
+    const float* __restrict__ q_mean_ema_ptr)   /* [1] pinned — unused but preserves 19-param graph node on Hopper */
 {
     int tid = blockIdx.x * blockDim.x + threadIdx.x;
     int total_elems = batch_size * num_atoms;
diff --git a/crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs b/crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs
index 5a2d25824..ca68e4da7 100644
--- a/crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs
+++ b/crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs
@@ -3415,7 +3415,7 @@ impl GpuDqnTrainer {
 
         let launch_cfg = cudarc::driver::LaunchConfig {
             grid_dim: (her_batch_size as u32, 1, 1),
-            block_dim: (1, 1, 1),
+            block_dim: (32, 1, 1),  // must match working SHA — graph_mega node structure on Hopper
             shared_mem_bytes: 0,
         };
 
@@ -8690,6 +8690,7 @@ impl GpuDqnTrainer {
                 .arg(&self.per_sample_support_ptr)
                 .arg(&self.liquid_mod_buf.raw_ptr())
                 .arg(&self.atom_positions_buf)
+                .arg(&self.q_mean_ema_dev_ptr) // unused but preserves 19-param graph node structure on Hopper
                 .launch(LaunchConfig {
                     grid_dim: (blocks, 1, 1),
                     block_dim: (256, 1, 1),