bench: promote PR126 identifiability P100 evidence v4

Author: github-actions[bot]

.github/workflows/pr126-v4-promotion.yml

@@ -92,7 +92,9 @@
         "panel-stage-c-rank-policy-validation-pr126-20260811-c67ada7ec59f",
         "panel-stage-c-rank-policy-performance-pr126-20260811-f27bef0b7c55",
         "panel-stage-c-rank-df-validation-pr126-20260812-0b4eb5810ad0",
-        "panel-stage-c-rank-df-performance-pr126-20260812-09337cc62c94"
+        "panel-stage-c-rank-df-performance-pr126-20260812-09337cc62c94",
+        "panel-stage-c-identifiability-validation-pr126-20260812-2d929bccf1c7",
+        "panel-stage-c-identifiability-performance-pr126-20260812-2238002d491f"
       ],
       "representative_dimensions": [
         "estimator",
@@ -103,10 +105,13 @@
         "inference_regression",
         "covariance",
         "n_times",
-        "timing_protocol"
+        "timing_protocol",
+        "coefficient_inference_applicability",
+        "prediction_backend",
+        "two_way_fe_timing"
       ],
       "issue": "#108",
-      "disposition": "June timing rows cover aligned PanelOLS and RandomEffects; PR #122 provides canonical Stage-B diagnostic/physical validation. PR #126 current Stage-C rank-policy acceptance is the exact-clean 3dc7df19 P100 source: CuPy and Torch each pass 27 estimator integrations plus 12 direct public primitives (39/39), including the numerical-rank boundary, with requested/executed backend identity. The paired current performance source contributes 58 synchronized rows over three base scales plus bounded N=10,000, k=2, T=200 QS all-lag cases. The older ec511f53, 5ed763be, aad53587, c151550a, and 9c0b3050 artifacts remain immutable historical evidence. No Stage-C speedup claim is made."
+      "disposition": "June timing rows cover aligned PanelOLS and RandomEffects; PR #122 provides canonical Stage-B diagnostic/physical validation. PR #126 current Stage-C acceptance is the exact-clean a99726e1 P100 identifiability source: CuPy and Torch each pass 35 estimator integrations plus 12 direct public primitives (47/47), including nine rank-deficient acceptance cases with ordinary coordinate inference explicitly unavailable and representative PanelOLS/RandomEffects prediction backend provenance. The paired current performance source contributes 60 synchronized rows: 54 base rows, four bounded N=10,000, k=2, T=200 QS rows, and two N=10,000, k=2, T=20 unbalanced two-way-FE rows. Historical v1/v2/v3 Stage-C sources remain immutable audit evidence and are not overwritten. No Stage-C speedup claim is made."
     },
     {
       "capability_id": "gam-nonparametric",

dev/benchmarks/benchmark_coverage_matrix.json

@@ -31,6 +31,10 @@
     parse_panel_stage_c_rank_df_physical_validation,
     parse_panel_stage_c_rank_df_performance,
 )
+from .panel_stage_c_identifiability import (
+    parse_panel_stage_c_identifiability_physical_validation,
+    parse_panel_stage_c_identifiability_performance,
+)
 
 __all__ = [
     "parse_penalized_glm_bench_perf",
@@ -56,4 +60,6 @@
     "parse_panel_stage_c_rank_policy_performance",
     "parse_panel_stage_c_rank_df_physical_validation",
     "parse_panel_stage_c_rank_df_performance",
+    "parse_panel_stage_c_identifiability_physical_validation",
+    "parse_panel_stage_c_identifiability_performance",
 ]