feat: complete Panel Tier-1 Stage C covariance

.github/workflows/benchmark-frontend.yml

@@ -16,10 +16,12 @@ on:
       - 'dev/tests/test_frontend_domain_coverage.py'
       - 'dev/tests/test_panel_stage_b_frontend_source.py'
       - 'dev/tests/test_panel_stage_b_applicable_hausman_parser.py'
+      - 'dev/tests/test_panel_stage_c_frontend_source.py'
       - 'dev/tests/fixtures/benchmark_frontend/**'
       - 'frontend/**'
       - 'docs/assets/benchmarks/**'
       - 'results/benchmark_frontend_sources/**'
+      - 'results/pr126_p100/**'
       - 'pyproject.toml'
   pull_request:
     paths:
@@ -35,10 +37,12 @@ on:
       - 'dev/tests/test_frontend_domain_coverage.py'
       - 'dev/tests/test_panel_stage_b_frontend_source.py'
       - 'dev/tests/test_panel_stage_b_applicable_hausman_parser.py'
+      - 'dev/tests/test_panel_stage_c_frontend_source.py'
       - 'dev/tests/fixtures/benchmark_frontend/**'
       - 'frontend/**'
       - 'docs/assets/benchmarks/**'
       - 'results/benchmark_frontend_sources/**'
+      - 'results/pr126_p100/**'
       - 'pyproject.toml'
 
 concurrency:
@@ -70,7 +74,8 @@ jobs:
             dev/tests/test_frontend_cv_determinism.py \
             dev/tests/test_frontend_domain_coverage.py \
             dev/tests/test_panel_stage_b_frontend_source.py \
-            dev/tests/test_panel_stage_b_applicable_hausman_parser.py -v
+            dev/tests/test_panel_stage_b_applicable_hausman_parser.py \
+            dev/tests/test_panel_stage_c_frontend_source.py -v
       - name: Validate generator output
         run: python dev/benchmarks/generate_benchmark_data.py --check --strict-sources
 

.github/workflows/panel-stage-c-external.yml

@@ -0,0 +1,52 @@
+name: Panel Stage C external covariance
+
+on:
+  pull_request:
+    branches: [master]
+    paths:
+      - 'statgpu/panel/**'
+      - 'dev/tests/test_panel_stage_c_external.py'
+      - 'dev/tests/test_panel_stage_c_external_defaults.py'
+      - 'dev/tests/test_panel_stage_c_linearmodels_estimators.py'
+      - '.github/workflows/panel-stage-c-external.yml'
+  push:
+    branches: [master]
+    paths:
+      - 'statgpu/panel/**'
+      - 'dev/tests/test_panel_stage_c_external.py'
+      - 'dev/tests/test_panel_stage_c_external_defaults.py'
+      - 'dev/tests/test_panel_stage_c_linearmodels_estimators.py'
+      - '.github/workflows/panel-stage-c-external.yml'
+
+permissions:
+  contents: read
+
+jobs:
+  external-definitions:
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+      - name: Install pinned references
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install -e . 'linearmodels==7.0' 'statsmodels==0.14.6' pytest
+      - name: Confirm reference versions
+        run: |
+          python - <<'PY'
+          import linearmodels, statsmodels
+          assert linearmodels.__version__ == '7.0'
+          assert statsmodels.__version__ == '0.14.6'
+          print('linearmodels', linearmodels.__version__)
+          print('statsmodels', statsmodels.__version__)
+          PY
+      - name: Run Stage C external covariance alignment
+        run: |
+          python -m pytest \
+            dev/tests/test_panel_stage_c_external.py \
+            dev/tests/test_panel_stage_c_external_defaults.py \
+            dev/tests/test_panel_stage_c_linearmodels_estimators.py \
+            -q --tb=short

.github/workflows/panel-stage-c-torch-cpu.yml

@@ -0,0 +1,44 @@
+name: Panel Stage C Torch CPU
+
+on:
+  pull_request:
+    branches: [master]
+    paths:
+      - 'statgpu/panel/**'
+      - 'dev/tests/test_panel_stage_c_torch_cpu.py'
+      - '.github/workflows/panel-stage-c-torch-cpu.yml'
+  push:
+    branches: [master]
+    paths:
+      - 'statgpu/panel/**'
+      - 'dev/tests/test_panel_stage_c_torch_cpu.py'
+      - '.github/workflows/panel-stage-c-torch-cpu.yml'
+
+permissions:
+  contents: read
+
+jobs:
+  torch-2-0-cpu:
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.9'
+      - name: Install maintained Torch CPU environment
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install "numpy<2" pytest
+          python -m pip install "torch==2.0.1+cpu" --extra-index-url https://download.pytorch.org/whl/cpu
+          python -m pip install -e .
+      - name: Confirm maintained Torch version
+        run: |
+          python - <<'PY'
+          import torch
+          assert torch.__version__.startswith('2.0.1')
+          assert not torch.cuda.is_available()
+          print('torch', torch.__version__, 'device=cpu')
+          PY
+      - name: Run Stage C Torch CPU covariance parity
+        run: python -m pytest dev/tests/test_panel_stage_c_torch_cpu.py -q --tb=short

CHANGELOG.md

@@ -2,6 +2,13 @@
 
 All notable changes to statgpu are documented here, organized by release and date.
 
+## 2026-08-09
+
+### PR #126 — Complete Panel Tier-1 Stage C covariance
+- Added HC0/HC2/HC3, robust RandomEffects inference, cluster group debiasing, and Driscoll-Kraay covariance with NumPy/CuPy/Torch-native accumulation.
+- Preserved historical HC1 (`robust`), Pooled row-HAC, default clustered covariance, coefficient estimates, and Stage-B diagnostics.
+- Added pinned statsmodels/linearmodels covariance checks plus exact-head physical GPU and performance validators. After the Ready-triggered ordered-categorical Driscoll-Kraay chronology fix, Tesla P100 acceptance was rerun and passed on exact clean implementation head `c151550ab17bd9533a51599f86b6a4ea12a292e9`: both CuPy and Torch passed all 26 estimator covariance cases plus two direct public primitives, and synchronized performance evidence includes the bounded `N=10,000`, `k=2`, `T=200` QS all-lag scenario without making a speedup claim. The earlier `9c0b3050dd143c43a06bb6393d69f4f83e861637` artifacts remain immutable historical evidence.
+
 ## 2026-08-08
 
 ### PR #122 — Panel Tier-1 diagnostics Stage B

dev/benchmarks/benchmark_coverage_matrix.json

@@ -86,18 +86,23 @@
       "status": "partial_canonical",
       "source_ids": [
         "new-modules-20260624-bcbdb676223b",
-        "panel-stage-b-pr122-20260809-2056f836bfe2"
+        "panel-stage-b-pr122-20260809-2056f836bfe2",
+        "panel-stage-c-validation-pr126-20260810-91e9ec53b6ee",
+        "panel-stage-c-performance-pr126-20260810-1861fc4f1817"
       ],
       "representative_dimensions": [
         "estimator",
         "backend",
         "aligned_scale",
         "physical_validation",
         "diagnostics",
-        "inference_regression"
+        "inference_regression",
+        "covariance",
+        "n_times",
+        "timing_protocol"
       ],
       "issue": "#108",
-      "disposition": "June timing rows cover aligned PanelOLS and RandomEffects. PR #122 adds canonical validation-only CuPy/Torch evidence for the 17-case Stage-B estimator matrix, five Hausman diagnostics per backend (including a physically applicable nonzero-effect statistic/p-value/df path), backend provenance, Stage-A inference regression, and the disconnected two-way FE df=1 physical boundary; broader performance/covariance timing remains open."
+      "disposition": "June timing rows cover aligned PanelOLS and RandomEffects; PR #122 provides canonical Stage-B diagnostic/physical validation. PR #126 current post-review P100 evidence at c151550a adds canonical Stage-C CuPy/Torch correctness for 26 estimator covariance integrations plus two direct public primitives per backend, and synchronized timing rows for three base scales plus a bounded N=10,000, k=2, T=200 QS all-lag scenario. The pre-review 9c0b3050 artifacts are retained only as historical evidence. No Stage-C speedup claim is made."
     },
     {
       "capability_id": "gam-nonparametric",