--- /dev/null	2025-08-20 16:29:02.164000374 +0000
+++ b/select_k_smoke_test.cu.cc	2025-08-20 20:54:24.000000000 +0000
@@ -0,0 +1,41 @@
+#include <cuda_runtime.h>
+#include <gtest/gtest.h>
+
+#include "select_k_runner.hpp"
+
+TEST(SelectKWrapperTest, BasicTopK) {
+  raft::resources res;
+
+  const uint32_t batch = 8;
+  const uint32_t n = 8192;
+  const uint32_t k = 16;
+
+  float* d_data_in = nullptr;
+  float* d_data_out = nullptr;
+  uint32_t* d_indices_out = nullptr;
+
+  ASSERT_EQ(cudaMalloc(&d_data_in, batch * n * sizeof(float)), cudaSuccess);
+  ASSERT_EQ(cudaMalloc(&d_data_out, batch * k * sizeof(float)), cudaSuccess);
+  ASSERT_EQ(cudaMalloc(&d_indices_out, batch * k * sizeof(uint32_t)),
+            cudaSuccess);
+
+  // host input: 0..n-1
+  std::vector<float> h_data_in(batch * n);
+  for (uint32_t row = 0; row < batch; ++row) {
+    for (uint32_t col = 0; col < n; ++col) {
+      h_data_in[row * n + col] = static_cast<float>(col);
+    }
+  }
+  ASSERT_EQ(
+      cudaMemcpy(d_data_in, h_data_in.data(), h_data_in.size() * sizeof(float),
+                 cudaMemcpyHostToDevice),
+      cudaSuccess);
+
+  // call wrapper
+  myraft::run_select_k(res, d_data_in, d_data_out, d_indices_out, batch, n, k,
+                       /*select_min=*/false, /*sorted=*/true);
+
+  cudaFree(d_data_in);
+  cudaFree(d_data_out);
+  cudaFree(d_indices_out);
+}