Add certified BenchCheck evaluation logs: summary.json

results/benchcheck_agentic_20260918/summary.json

@@ -0,0 +1,39 @@
+{
+  "dataset": "GMLRVigil/BenchCheck-MetaBenchmark",
+  "dataset_revision": "05e1acb6d6f6db9a6cf239795edfa21a1f75543a",
+  "evaluation_date_utc": "2026-09-18",
+  "condition": "VIDEO",
+  "models": [
+    {
+      "model": "gpt-6-astra",
+      "total": 840,
+      "correct": 656,
+      "accuracy": 0.780952380952381,
+      "valid_answers": 839,
+      "failed_attempts": 1,
+      "timeouts": 0,
+      "technical_errors": 1,
+      "interrupted": 0
+    },
+    {
+      "model": "claude-fable-5-1",
+      "total": 840,
+      "correct": 583,
+      "accuracy": 0.694047619047619,
+      "valid_answers": 802,
+      "failed_attempts": 38,
+      "timeouts": 21,
+      "technical_errors": 9,
+      "interrupted": 8
+    }
+  ],
+  "protocol": {
+    "reasoning_effort": "max",
+    "workers_per_model": 8,
+    "cli_timeout_seconds": 1800,
+    "sampling": "model-directed",
+    "preset_fps_or_frame_cap": null,
+    "blind_condition": false
+  },
+  "scoring": "Exact answer-letter match; all failures count as incorrect; no retry or salvage of formal failures."
+}