diff --git a/.github/workflows/ci-nightly.yml b/.github/workflows/ci-nightly.yml
index 684083edb1..42226d1535 100644
--- a/.github/workflows/ci-nightly.yml
+++ b/.github/workflows/ci-nightly.yml
@@ -98,7 +98,7 @@ jobs:
       run-id: ${{ needs.find-wheels.outputs.RUN_ID }}
       sha: ${{ needs.find-wheels.outputs.HEAD_SHA }}
       test-mode: nightly-pytorch
-      matrix_filter: 'map(select(.MODE == "nightly-pytorch"))'
+      matrix_filter: 'map(select(.ENV.MODE == "nightly-pytorch"))'
 
   test-pytorch-linux-aarch64:
     name: "Nightly PyTorch (linux-aarch64)"
@@ -116,7 +116,7 @@ jobs:
       run-id: ${{ needs.find-wheels.outputs.RUN_ID }}
       sha: ${{ needs.find-wheels.outputs.HEAD_SHA }}
       test-mode: nightly-pytorch
-      matrix_filter: 'map(select(.MODE == "nightly-pytorch"))'
+      matrix_filter: 'map(select(.ENV.MODE == "nightly-pytorch"))'
 
   test-pytorch-windows:
     name: "Nightly PyTorch (win-64)"
@@ -134,7 +134,7 @@ jobs:
       run-id: ${{ needs.find-wheels.outputs.RUN_ID }}
       sha: ${{ needs.find-wheels.outputs.HEAD_SHA }}
       test-mode: nightly-pytorch
-      matrix_filter: 'map(select(.MODE == "nightly-pytorch"))'
+      matrix_filter: 'map(select(.ENV.MODE == "nightly-pytorch"))'
 
   # ── numba-cuda tests ──
 
@@ -154,7 +154,7 @@ jobs:
       run-id: ${{ needs.find-wheels.outputs.RUN_ID }}
       sha: ${{ needs.find-wheels.outputs.HEAD_SHA }}
       test-mode: nightly-numba-cuda
-      matrix_filter: 'map(select(.MODE == "nightly-numba-cuda"))'
+      matrix_filter: 'map(select(.ENV.MODE == "nightly-numba-cuda"))'
 
   test-numba-cuda-linux-aarch64:
     name: "Nightly numba-cuda (linux-aarch64)"
@@ -172,7 +172,7 @@ jobs:
       run-id: ${{ needs.find-wheels.outputs.RUN_ID }}
       sha: ${{ needs.find-wheels.outputs.HEAD_SHA }}
       test-mode: nightly-numba-cuda
-      matrix_filter: 'map(select(.MODE == "nightly-numba-cuda"))'
+      matrix_filter: 'map(select(.ENV.MODE == "nightly-numba-cuda"))'
 
   test-numba-cuda-windows:
     name: "Nightly numba-cuda (win-64)"
@@ -190,7 +190,7 @@ jobs:
       run-id: ${{ needs.find-wheels.outputs.RUN_ID }}
       sha: ${{ needs.find-wheels.outputs.HEAD_SHA }}
       test-mode: nightly-numba-cuda
-      matrix_filter: 'map(select(.MODE == "nightly-numba-cuda"))'
+      matrix_filter: 'map(select(.ENV.MODE == "nightly-numba-cuda"))'
 
   # ── Standard tests on nightly-only runners ──
 
@@ -210,7 +210,7 @@ jobs:
       run-id: ${{ needs.find-wheels.outputs.RUN_ID }}
       sha: ${{ needs.find-wheels.outputs.HEAD_SHA }}
       test-mode: standard
-      matrix_filter: 'map(select(.MODE == "nightly-standard"))'
+      matrix_filter: 'map(select(.ENV.MODE == "nightly-standard"))'
 
   # ── Status check ──
 
diff --git a/.github/workflows/test-wheel-linux.yml b/.github/workflows/test-wheel-linux.yml
index 72ee298f77..f72ebacb3e 100644
--- a/.github/workflows/test-wheel-linux.yml
+++ b/.github/workflows/test-wheel-linux.yml
@@ -100,7 +100,7 @@ jobs:
           echo "OLD_BRANCH=${OLD_BRANCH}" >> "$GITHUB_OUTPUT"
 
   test:
-    name: Python ${{ matrix.PY_VER }}, CUDA ${{ matrix.CUDA_VER }} (${{ (matrix.LOCAL_CTK == '1' && 'local') || 'wheels' }}), GPU ${{ matrix.GPU }}${{ matrix.GPU_COUNT != '1' && format(' (x{0})', matrix.GPU_COUNT) || '' }}${{ matrix.FLAVOR && format(', {0}', matrix.FLAVOR) || '' }}${{ matrix.TORCH_VER && format(', {0}', matrix.TORCH_VER) || '' }}${{ matrix.MODE == 'nightly-numba-cuda' && ', latest' || '' }}
+    name: Python ${{ matrix.PY_VER }}, CUDA ${{ matrix.CUDA_VER }} (${{ (matrix.LOCAL_CTK == '1' && 'local') || 'wheels' }}), GPU ${{ matrix.GPU }}${{ matrix.GPU_COUNT != '1' && format(' (x{0})', matrix.GPU_COUNT) || '' }}${{ matrix.FLAVOR && format(', {0}', matrix.FLAVOR) || '' }}${{ matrix.ENV.TORCH_VER && format(', {0}+{1}', matrix.ENV.TORCH_VER, matrix.ENV.TORCH_CUDA) || '' }}${{ matrix.ENV.MODE == 'nightly-numba-cuda' && ', latest' || '' }}
     timeout-minutes: 60
     needs: compute-matrix
     strategy:
@@ -400,8 +400,6 @@ jobs:
         env:
           CUDA_VER: ${{ matrix.CUDA_VER }}
           LOCAL_CTK: ${{ matrix.LOCAL_CTK }}
-          TORCH_VER: ${{ matrix.TORCH_VER }}
-          TORCH_CUDA: ${{ matrix.TORCH_CUDA }}
         run: run-tests nightly-pytorch
 
       - name: Install cuda-python wheels + numba-cuda
diff --git a/.github/workflows/test-wheel-windows.yml b/.github/workflows/test-wheel-windows.yml
index 6db50b89da..bf50581527 100644
--- a/.github/workflows/test-wheel-windows.yml
+++ b/.github/workflows/test-wheel-windows.yml
@@ -90,7 +90,7 @@ jobs:
           echo "MATRIX=${MATRIX}" | tee --append "${GITHUB_OUTPUT}"
 
   test:
-    name: Python ${{ matrix.PY_VER }}, CUDA ${{ matrix.CUDA_VER }} (${{ (matrix.LOCAL_CTK == '1' && 'local') || 'wheels' }}), GPU ${{ matrix.GPU }}${{ matrix.GPU_COUNT != '1' && format(' (x{0})', matrix.GPU_COUNT) || '' }} (${{ matrix.DRIVER_MODE }})${{ matrix.TORCH_VER && format(', {0}', matrix.TORCH_VER) || '' }}${{ matrix.MODE == 'nightly-numba-cuda' && ', latest' || '' }}
+    name: Python ${{ matrix.PY_VER }}, CUDA ${{ matrix.CUDA_VER }} (${{ (matrix.LOCAL_CTK == '1' && 'local') || 'wheels' }}), GPU ${{ matrix.GPU }}${{ matrix.GPU_COUNT != '1' && format(' (x{0})', matrix.GPU_COUNT) || '' }} (${{ matrix.DRIVER_MODE }})${{ matrix.ENV.TORCH_VER && format(', {0}+{1}', matrix.ENV.TORCH_VER, matrix.ENV.TORCH_CUDA) || '' }}${{ matrix.ENV.MODE == 'nightly-numba-cuda' && ', latest' || '' }}
     timeout-minutes: 60
     # The build stage could fail but we want the CI to keep moving.
     needs: compute-matrix
@@ -377,8 +377,6 @@ jobs:
         env:
           CUDA_VER: ${{ matrix.CUDA_VER }}
           LOCAL_CTK: ${{ matrix.LOCAL_CTK }}
-          TORCH_VER: ${{ matrix.TORCH_VER }}
-          TORCH_CUDA: ${{ matrix.TORCH_CUDA }}
         shell: bash --noprofile --norc -xeuo pipefail {0}
         run: run-tests nightly-pytorch
 
diff --git a/ci/test-matrix.yml b/ci/test-matrix.yml
index ad12bb243a..c9eafd4f52 100644
--- a/ci/test-matrix.yml
+++ b/ci/test-matrix.yml
@@ -27,8 +27,10 @@
 #         job. Every key/value pair is written to GITHUB_ENV after the standard
 #         environment-variable setup step, so the variables are visible to all
 #         subsequent steps (including the cuda.bindings and cuda.core test
-#         steps). Example:
+#         steps). Nightly rows also use ENV.MODE as a matrix-filter tag (see
+#         ci-nightly.yml). Examples:
 #           ENV: { CUDA_PYTHON_PER_THREAD_DEFAULT_STREAM: '1' }
+#           ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.12.1', TORCH_CUDA: 'cu126' }
 
 linux:
   pull-request:
@@ -81,22 +83,25 @@ linux:
     - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '13.3.0', LOCAL_CTK: '0', GPU: 'rtx4090',    GPU_COUNT: '1', DRIVER: 'latest', FLAVOR: 'wsl' }
   nightly:
     # nightly-pytorch
-    - { MODE: 'nightly-pytorch',    ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '12.6.3', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', TORCH_VER: '2.11.0', TORCH_CUDA: 'cu126' }
-    - { MODE: 'nightly-pytorch',    ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '13.0.2', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', TORCH_VER: '2.11.0', TORCH_CUDA: 'cu130' }
-    - { MODE: 'nightly-pytorch',    ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '12.6.3', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', TORCH_VER: '2.9.1',  TORCH_CUDA: 'cu126' }
-    - { MODE: 'nightly-pytorch',    ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '13.0.2', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', TORCH_VER: '2.9.1',  TORCH_CUDA: 'cu130' }
-    - { MODE: 'nightly-pytorch',    ARCH: 'arm64', PY_VER: '3.12', CUDA_VER: '12.6.3', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', TORCH_VER: '2.11.0', TORCH_CUDA: 'cu126' }
-    - { MODE: 'nightly-pytorch',    ARCH: 'arm64', PY_VER: '3.12', CUDA_VER: '13.0.2', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', TORCH_VER: '2.11.0', TORCH_CUDA: 'cu130' }
-    - { MODE: 'nightly-pytorch',    ARCH: 'arm64', PY_VER: '3.12', CUDA_VER: '12.6.3', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', TORCH_VER: '2.9.1',  TORCH_CUDA: 'cu126' }
-    - { MODE: 'nightly-pytorch',    ARCH: 'arm64', PY_VER: '3.12', CUDA_VER: '13.0.2', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', TORCH_VER: '2.9.1',  TORCH_CUDA: 'cu130' }
+    - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '12.6.3', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',     ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.12.1', TORCH_CUDA: 'cu126' } }
+    - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '13.0.2', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',     ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.12.1', TORCH_CUDA: 'cu130' } }
+    - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '12.6.3', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',     ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.9.1',  TORCH_CUDA: 'cu126' } }
+    - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '13.0.2', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',     ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.9.1',  TORCH_CUDA: 'cu130' } }
+    - { ARCH: 'arm64', PY_VER: '3.12',  CUDA_VER: '12.6.3', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',     ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.12.1', TORCH_CUDA: 'cu126' } }
+    - { ARCH: 'arm64', PY_VER: '3.12',  CUDA_VER: '13.0.2', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',     ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.12.1', TORCH_CUDA: 'cu130' } }
+    - { ARCH: 'arm64', PY_VER: '3.12',  CUDA_VER: '12.6.3', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',     ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.9.1',  TORCH_CUDA: 'cu126' } }
+    - { ARCH: 'arm64', PY_VER: '3.12',  CUDA_VER: '13.0.2', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',     ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.9.1',  TORCH_CUDA: 'cu130' } }
     # nightly-numba-cuda
-    - { MODE: 'nightly-numba-cuda', ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '12.9.1', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest' }
-    - { MODE: 'nightly-numba-cuda', ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '13.3.0', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: '580.65.06' }
-    - { MODE: 'nightly-numba-cuda', ARCH: 'arm64', PY_VER: '3.12', CUDA_VER: '12.9.1', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest' }
-    - { MODE: 'nightly-numba-cuda', ARCH: 'arm64', PY_VER: '3.12', CUDA_VER: '13.3.0', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest' }
-    # nightly-standard (arm64 l4×2 — nightly-only per runner team request)
-    - { MODE: 'nightly-standard', ARCH: 'arm64', PY_VER: '3.14',  CUDA_VER: '13.3.0', LOCAL_CTK: '1', GPU: 'l4', GPU_COUNT: '2', DRIVER: 'latest' }
-    - { MODE: 'nightly-standard', ARCH: 'arm64', PY_VER: '3.14t', CUDA_VER: '13.3.0', LOCAL_CTK: '1', GPU: 'l4', GPU_COUNT: '2', DRIVER: 'latest' }
+    - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '12.9.1', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',     ENV: { MODE: 'nightly-numba-cuda' } }
+    - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '13.3.0', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: '580.65.06', ENV: { MODE: 'nightly-numba-cuda' } }
+    - { ARCH: 'arm64', PY_VER: '3.12',  CUDA_VER: '12.9.1', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',     ENV: { MODE: 'nightly-numba-cuda' } }
+    - { ARCH: 'arm64', PY_VER: '3.12',  CUDA_VER: '13.3.0', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',     ENV: { MODE: 'nightly-numba-cuda' } }
+    # nightly-standard (arm64 nightly-only runners — per runner team request)
+    # TODO: gh200 row disabled — currently hangs on stream-ordered memory
+    #       allocator (cudaMallocAsync); runner pool needs fixing first.
+    # - { ARCH: 'arm64', PY_VER: '3.14',  CUDA_VER: '13.3.0', LOCAL_CTK: '1', GPU: 'gh200',      GPU_COUNT: '1', DRIVER: 'latest',     ENV: { MODE: 'nightly-standard' } }
+    - { ARCH: 'arm64', PY_VER: '3.14',  CUDA_VER: '13.3.0', LOCAL_CTK: '1', GPU: 'l4',         GPU_COUNT: '2', DRIVER: 'latest',     ENV: { MODE: 'nightly-standard' } }
+    - { ARCH: 'arm64', PY_VER: '3.14t', CUDA_VER: '13.3.0', LOCAL_CTK: '1', GPU: 'l4',         GPU_COUNT: '2', DRIVER: 'latest',     ENV: { MODE: 'nightly-standard' } }
 
 windows:
   pull-request:
@@ -124,10 +129,10 @@ windows:
     - { ARCH: 'amd64', PY_VER: '3.14',  CUDA_VER: '13.3.0', LOCAL_CTK: '0', GPU: 'h100',       GPU_COUNT: '2', DRIVER: 'latest', DRIVER_MODE: 'MCDM' }
   nightly:
     # nightly-pytorch
-    - { MODE: 'nightly-pytorch',    ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '12.6.3', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', DRIVER_MODE: 'TCC', TORCH_VER: '2.11.0',   TORCH_CUDA: 'cu126' }
-    - { MODE: 'nightly-pytorch',    ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '13.0.2', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', DRIVER_MODE: 'TCC', TORCH_VER: '2.11.0',   TORCH_CUDA: 'cu130' }
-    - { MODE: 'nightly-pytorch',    ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '12.6.3', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', DRIVER_MODE: 'TCC', TORCH_VER: '2.9.1',  TORCH_CUDA: 'cu126' }
-    - { MODE: 'nightly-pytorch',    ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '13.0.2', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', DRIVER_MODE: 'TCC', TORCH_VER: '2.9.1',  TORCH_CUDA: 'cu130' }
+    - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '12.6.3', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',  DRIVER_MODE: 'TCC', ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.12.1', TORCH_CUDA: 'cu126' } }
+    - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '13.0.2', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',  DRIVER_MODE: 'TCC', ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.12.1', TORCH_CUDA: 'cu130' } }
+    - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '12.6.3', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',  DRIVER_MODE: 'TCC', ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.9.1',  TORCH_CUDA: 'cu126' } }
+    - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '13.0.2', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',  DRIVER_MODE: 'TCC', ENV: { MODE: 'nightly-pytorch', TORCH_VER: '2.9.1',  TORCH_CUDA: 'cu130' } }
     # nightly-numba-cuda
-    - { MODE: 'nightly-numba-cuda', ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '12.9.1', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: 'latest', DRIVER_MODE: 'TCC' }
-    - { MODE: 'nightly-numba-cuda', ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '13.3.0', LOCAL_CTK: '0', GPU: 'l4', GPU_COUNT: '1', DRIVER: '596.36',  DRIVER_MODE: 'TCC' }
+    - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '12.9.1', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: 'latest',  DRIVER_MODE: 'TCC', ENV: { MODE: 'nightly-numba-cuda' } }
+    - { ARCH: 'amd64', PY_VER: '3.12',  CUDA_VER: '13.3.0', LOCAL_CTK: '0', GPU: 'l4',         GPU_COUNT: '1', DRIVER: '596.36',  DRIVER_MODE: 'TCC', ENV: { MODE: 'nightly-numba-cuda' } }