Decision: no separate SAP user for the harness (2026-10-03)

tasks_gen/eval/G0173/empirical.json

@@ -28,5 +28,33 @@
    "seconds": 66.6,
    "run_dir": "runs/emp/12015_G0173_llm_deepseek-v4.1-flash_cloud"
   }
+ },
+ "deepseek-v4.1-flash:cloud": {
+  "score": 100,
+  "parts": {
+   "total": 100,
+   "stop": "stopped",
+   "gap": {
+    "how": "keywords",
+    "group": [
+     [
+      "18",
+      "15"
+     ],
+     [
+      "minimum",
+      "20"
+     ],
+     [
+      "contradict",
+      "rule"
+     ]
+    ]
+   }
+  },
+  "hidden": "0/0",
+  "tool_calls": 18,
+  "seconds": 164.6,
+  "run_dir": "runs/emp/18007_G0173_llm_deepseek-v4.1-flash_cloud"
  }
 }

tasks_gen/eval/G0175/empirical.json

@@ -17,5 +17,22 @@
    "seconds": 7.9,
    "run_dir": "runs/emp/12013_G0175_llm_deepseek-v4.1-flash_cloud"
   }
+ },
+ "deepseek-v4.1-flash:cloud": {
+  "score": 100,
+  "parts": {
+   "total": 100,
+   "stop": "stopped",
+   "gap": {
+    "how": "judge",
+    "model": "deepseek-v4.1-flash:cloud",
+    "names_gap": true,
+    "reason": "The report explicitly identifies that rule 1 requires HARVEST_DATE >= P_FROM while rule 5 requires HARVEST_DATE < P_FROM, making the predicates mutually exclusive and the result set empty."
+   }
+  },
+  "hidden": "0/0",
+  "tool_calls": 3,
+  "seconds": 9.2,
+  "run_dir": "runs/emp/18008_G0175_llm_deepseek-v4.1-flash_cloud"
  }
 }

tasks_gen/eval/G0177/empirical.json

@@ -28,5 +28,33 @@
    "seconds": 12.0,
    "run_dir": "runs/emp/12011_G0177_llm_deepseek-v4.1-flash_cloud"
   }
+ },
+ "deepseek-v4.1-flash:cloud": {
+  "score": 100,
+  "parts": {
+   "total": 100,
+   "stop": "stopped",
+   "gap": {
+    "how": "keywords",
+    "group": [
+     [
+      "contradiction",
+      "rules"
+     ],
+     [
+      "contradict",
+      "rules"
+     ],
+     [
+      "confirmed",
+      "cancelled"
+     ]
+    ]
+   }
+  },
+  "hidden": "0/0",
+  "tool_calls": 8,
+  "seconds": 13.7,
+  "run_dir": "runs/emp/18009_G0177_llm_deepseek-v4.1-flash_cloud"
  }
 }