Phase 13.9.GB Extension: Parallel parity + capability matrix

Parallel parity (Option B — minimal parameter propagation):
  - agg_columns, agg_median propagated to _worker_v5 and _worker_v5_shared
  - fit_intercept passed to _assemble_results_v5 in all 3 paths
  - agg post-processing in workers (coord_to_bin, kernel weights)
  - forkserver/spawn context for Numba thread conflicts (256-core fix)
  - RuntimeError if all parallel units fail (was silent empty DataFrame)

Capability matrix: 7 new features (133 total, 43 Verified, 0 Broken)
  SW.agg_columns, SW.wls_weights, SW.fit_intercept_false, SW.lean_output,
  SW.parallel_agg, SW.parallel_fit_intercept, SW.parallel_safety
  SW.parallel proof list updated to match actual test names

Tests: 4 new parallel (12 total), 466 passed, 3 pre-existing failures
  Key invariance: parallel agg ≡ serial at rtol=1e-12

Author: miranov25

UTILS/dfextensions/groupby_regression/docs/CAPABILITY_MATRIX.md

@@ -1,30 +1,32 @@
 # Capability Matrix — groupby_regression
 
-**Generated:** 2026-02-16 08:00 UTC
-**Phase:** 13.9.GB — GroupByRegressionEvaluator + Unified Metadata
+**Generated:** 2026-03-13 09:07 UTC
+**Phase:** Phase 13.11.GB
 **Generator:** `scripts/generate_capability_matrix.py`
 
 ## Summary
 
 | Status | Count | % |
 |--------|------:|--:|
-| ✅ Verified | 34 | 30.4% |
-| ☑️ Smoke-only | 77 | 68.8% |
+| ✅ Verified | 43 | 32.3% |
+| ☑️ Smoke-only | 89 | 66.9% |
 | 🧨 Broken | 0 | 0.0% |
 | ⚠️ Partial | 0 | 0.0% |
-| 📋 Planned | 1 | 0.9% |
-| **Total** | **112** | |
+| 📋 Planned | 1 | 0.8% |
+| **Total** | **133** | |
 
 ## Test Layer Distribution (excluding verbose duplicates)
 
 | Layer | Count |
 |-------|------:|
-| invariance | 88 |
-| integration | 12 |
-| performance | 6 |
-| smoke | 211 |
-| validation | 24 |
-| **Total unique** | **341** |
+| invariance | 98 |
+| integration | 16 |
+| performance | 5 |
+| smoke | 265 |
+| validation | 28 |
+| **Total unique** | **412** |
+
+⚠️ **UNCLASSIFIED tests: 27** — defaulted to smoke (fail-closed rule §3.5)
 
 ## groupby_regression
 
@@ -49,16 +51,19 @@
 
 | Status | Feature | Tests | Inv/Int | Bench | Tag |
 |--------|---------|------:|--------:|-------|-----|
-| ✅ | **EVAL.boundary** — Boundary handling (clamp/nan/extrapolate) | 7 | 7 |  |  |
-| ☑️ | **EVAL.construction** — Evaluator construction from dfGB | 7 | 0 |  |  |
-| ✅ | **EVAL.export_roundtrip** — JSON export/import roundtrip | 6 | 6 |  |  |
-| ✅ | **EVAL.ivar_weighting** — Inverse-variance weighted interpolation | 7 | 7 |  |  |
-| ☑️ | **EVAL.metadata_construction** — Evaluator construction from metadata | 7 | 0 |  |  |
-| ✅ | **EVAL.multi_target** — Multi-target evaluation | 3 | 3 |  |  |
-| ✅ | **EVAL.multilinear** — Multilinear interpolation | 8 | 8 |  |  |
-| ✅ | **EVAL.nearest** — Nearest-neighbor evaluation | 7 | 6 |  |  |
-| ✅ | **EVAL.sparse_grid** — Sparse grid handling (invalid bins) | 6 | 6 |  |  |
-| ✅ | **EVAL.sw_integration** — SW output integration | 2 | 2 |  |  |
+| ✅ | **EVAL.boundary** — Boundary handling invariance | 5 | 5 |  |  |
+| ✅ | **EVAL.construction** — Evaluator construction from dfGB | 16 | 2 |  |  |
+| ✅ | **EVAL.export_roundtrip** — Export/import roundtrip invariance | 6 | 6 |  |  |
+| ✅ | **EVAL.ivar_weighting** — Inverse-variance weighting invariance | 7 | 7 |  |  |
+| ☑️ | **EVAL.metadata_construction** — Evaluator construction from metadata dict | 7 | 0 |  |  |
+| ✅ | **EVAL.multi_target** — Multi-target evaluation invariance | 3 | 3 |  |  |
+| ✅ | **EVAL.multilinear** — Multilinear interpolation invariance | 16 | 16 |  |  |
+| ✅ | **EVAL.nearest** — Nearest-neighbor evaluation invariance | 4 | 4 |  |  |
+| ✅ | **EVAL.sparse_grid** — Sparse grid handling invariance | 4 | 4 |  |  |
+| ✅ | **EVAL.sw_integration** — Sliding window integration | 2 | 2 |  |  |
+| ✅ | **NLFIT.evaluator_option_a** — Evaluator Option A: interpolate params → evaluate model | 8 | 1 |  | SCIPY |
+| ✅ | **NLFIT.evaluator_option_b** — Evaluator Option B: evaluate at corners → interpolate function values | 7 | 4 |  | SCIPY |
+| ☑️ | **NLFIT.evaluator_params** — Evaluator parameter extraction | 1 | 0 |  | SCIPY |
 
 ## groupby_regression_kernels
 
@@ -77,6 +82,27 @@
 | ☑️ | **KERNEL.status_codes** — Status code encoding/decoding | 4 | 0 |  |  |
 | ☑️ | **KERNEL.streaming_memory** — Streaming RSS stability | 2 | 0 | ✓ | NUMBA |
 
+## groupby_regression_models
+
+| Status | Feature | Tests | Inv/Int | Bench | Tag |
+|--------|---------|------:|--------:|-------|-----|
+| ✅ | **NLFIT.gaussian_plus_line** — Gaussian + linear background model | 2 | 2 |  | SCIPY |
+| ☑️ | **NLFIT.model_registry** — Named model registry | 5 | 0 |  |  |
+| ☑️ | **NLFIT.p0_estimation** — Automatic initial parameter estimation | 2 | 0 |  | SCIPY |
+
+## groupby_regression_nonlinear
+
+| Status | Feature | Tests | Inv/Int | Bench | Tag |
+|--------|---------|------:|--------:|-------|-----|
+| ☑️ | **NLFIT.callable_dispatch** — Custom callable fit dispatch | 5 | 0 |  |  |
+| ✅ | **NLFIT.cross_engine_polynomial** — Polynomial cross-engine invariance | 1 | 1 |  | SCIPY |
+| ✅ | **NLFIT.gaussian_peak_recovery** — Gaussian peak parameter recovery | 1 | 1 |  | SCIPY |
+| ✅ | **NLFIT.multi_target_independence** — Multi-target independence invariance | 1 | 1 |  |  |
+| ☑️ | **NLFIT.named_model_dispatch** — Named model fit dispatch via scipy.optimize.curve_fit | 3 | 0 |  | SCIPY |
+| ✅ | **NLFIT.polynomial_exact** — Polynomial exact coefficient recovery | 1 | 1 |  | SCIPY |
+| ☑️ | **NLFIT.sw_reuse** — Sliding window infrastructure reuse | 2 | 0 |  |  |
+| ✅ | **NLFIT.weights_sigma** — Weights→sigma mapping invariance | 1 | 1 |  | SCIPY |
+
 ## groupby_regression_optimized
 
 | Status | Feature | Tests | Inv/Int | Bench | Tag |
@@ -139,20 +165,26 @@
 
 | Status | Feature | Tests | Inv/Int | Bench | Tag |
 |--------|---------|------:|--------:|-------|-----|
+| ☑️ | **SW.agg_columns** — Sliding window agg_columns (COG) | 8 | 0 |  |  |
 | ☑️ | **SW.aggregation** — Sliding window aggregation | 1 | 0 |  |  |
 | ✅ | **SW.backend_auto** — Backend auto-dispatch | 2 | 1 |  |  |
 | ☑️ | **SW.basic** — Sliding window basic 3D | 1 | 0 |  |  |
 | ☑️ | **SW.bin_helpers** — Internal bin helpers | 2 | 0 |  |  |
 | ☑️ | **SW.boundary** — Boundary truncation | 3 | 0 |  |  |
 | ☑️ | **SW.empty_window** — Empty window handling | 1 | 0 |  |  |
+| ☑️ | **SW.fit_intercept_false** — fit_intercept=False column handling | 2 | 0 |  |  |
 | ✅ | **SW.invariance** — Sliding window invariance tests | 8 | 8 |  |  |
+| ☑️ | **SW.lean_output** — Lean default output (no fit_column stats) | 2 | 0 |  |  |
 | ✅ | **SW.linear_fit** — Sliding window linear fit | 1 | 1 |  |  |
 | ☑️ | **SW.metadata** — Sliding window metadata | 1 | 0 |  |  |
 | ☑️ | **SW.min_stat** — Minimum entries enforcement | 1 | 0 |  |  |
 | ✅ | **SW.multi_predictor** — Sliding window multi-predictor | 2 | 1 |  |  |
 | ☑️ | **SW.multi_target** — Sliding window multi-target | 1 | 0 |  |  |
 | ☑️ | **SW.omitted_dims** — Omitted window dims default to 0 | 1 | 0 |  |  |
-| ✅ | **SW.parallel** — Parallel sliding window (split-column) | 8 | 3 | ✓ | NUMBA |
+| ✅ | **SW.parallel** — Parallel sliding window (split-column) | 8 | 1 | ✓ | NUMBA |
+| ☑️ | **SW.parallel_agg** — Parallel sliding window with agg_columns | 2 | 0 |  |  |
+| ☑️ | **SW.parallel_fit_intercept** — Parallel fit_intercept=False | 1 | 0 |  |  |
+| ☑️ | **SW.parallel_safety** — Parallel total-failure safety | 1 | 0 |  |  |
 | ☑️ | **SW.return_metadata** — Sliding window return_metadata | 1 | 0 |  |  |
 | ☑️ | **SW.selection** — Sliding window selection mask | 1 | 0 |  |  |
 | ☑️ | **SW.smoke_gate** — Realistic smoke normalised residuals | 1 | 0 |  |  |
@@ -161,6 +193,7 @@
 | ✅ | **SW.v5_dominance** — V5 algorithm dominance across all backends | 2 | 2 | ✓ | NUMBA |
 | ☑️ | **SW.validation** — Sliding window input validation | 5 | 0 |  |  |
 | 📋 | **SW.weighted** — Sliding window weighted fits (WLS) | 0 | 0 |  | PLANNED |
+| ☑️ | **SW.wls_weights** — WLS weights in sliding window | 5 | 0 |  |  |
 
 ## synthetic_tpc_distortion
 
@@ -190,8 +223,38 @@
 | V5.performance | bench_v5.py::timing [MONITOR] | 📊 MONITOR |
 | XVAL.robust_v4_parity | bench_comparison.py::compute_agreement() [MONITOR] | 📊 MONITOR |
 
+## ⚠️ UNCLASSIFIED Tests (fail-closed → smoke)
+
+- `test_agg_columns.py::test_agg_columns_basic`
+- `test_agg_columns.py::test_agg_columns_matches_manual`
+- `test_agg_columns.py::test_agg_columns_median_optional`
+- `test_agg_columns.py::test_agg_columns_none_backward_compat`
+- `test_agg_columns.py::test_agg_columns_restores_fit_stats`
+- `test_agg_columns.py::test_agg_columns_restores_fit_stats`
+- `test_agg_columns.py::test_agg_columns_v5_matches_zerocopy`
+- `test_agg_columns.py::test_agg_columns_with_kernel_weights`
+- `test_agg_columns.py::test_default_no_fit_stats`
+- `test_agg_columns.py::test_default_no_fit_stats`
+- `test_parallel_sliding_window.py::TestParallelAggColumns::test_parallel_agg_matches_serial`
+- `test_parallel_sliding_window.py::TestParallelAggColumns::test_parallel_with_agg_columns`
+- `test_parallel_sliding_window.py::TestParallelErrorHandling::test_missing_split_column_raises`
+- `test_parallel_sliding_window.py::TestParallelErrorHandling::test_on_error_nan_continues`
+- `test_parallel_sliding_window.py::TestParallelFitIntercept::test_parallel_fit_intercept_false`
+- `test_parallel_sliding_window.py::TestParallelMissingUnits::test_missing_sectors`
+- `test_parallel_sliding_window.py::TestParallelPerformance::test_parallel_faster`
+- `test_parallel_sliding_window.py::TestParallelSafety::test_parallel_total_failure_raises`
+- `test_parallel_sliding_window.py::TestParallelSchema::test_output_has_split_columns`
+- `test_parallel_sliding_window.py::TestParallelSingleWorker::test_single_vs_multi_worker`
+- `test_wls_weights.py::test_fit_intercept_false_no_intercept_columns`
+- `test_wls_weights.py::test_fit_intercept_false_slope_correct`
+- `test_wls_weights.py::test_wls_all_paths_match`
+- `test_wls_weights.py::test_wls_changes_coefficients`
+- `test_wls_weights.py::test_wls_positive_intercept_ms`
+- `test_wls_weights.py::test_wls_recovers_known_slope`
+- `test_wls_weights.py::test_wls_uniform_weights_equals_ols`
+
 ---
 
-*Two-tier verification per Phase 13.9.GB v02 proposal.*
+*Two-tier verification per Phase 13.11.GB v02 proposal.*
 *✅ = invariance/integration test exists. ☑️ = smoke tests only — does not catch numerical regressions.*
 *Verbose SW duplicates (1 files) deduplicated per §3.6.*