onbuzz 4.8.3 → 4.9.0

package/scripts/memory-bench/results-baseline-2026-05-19T11-34-19-299Z.json

@@ -0,0 +1,1007 @@
+{
+  "timestamp": "2026-05-19T11:34:19.304Z",
+  "args": {
+    "models": [
+      "gpt-4.1-nano",
+      "gpt-4.1-mini",
+      "Kimi-K2.6"
+    ],
+    "trials": 2,
+    "scenarios": null,
+    "variant": "baseline",
+    "out": null
+  },
+  "results": [
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M1-catalog-recall",
+      "mechanism": "catalog-injection",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": false,
+      "reason": "expected memory.read, not called",
+      "callLog": [],
+      "totalPromptTokens": 420,
+      "totalOutputTokens": 55,
+      "turns": 1,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M1-catalog-recall",
+      "mechanism": "catalog-injection",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": false,
+      "reason": "expected memory.read, not called",
+      "callLog": [],
+      "totalPromptTokens": 420,
+      "totalOutputTokens": 79,
+      "turns": 1,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M2-transition-attached",
+      "mechanism": "transition-handler-attached",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "matched taskmanager.update",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "update",
+          "params": {
+            "action": "update",
+            "taskId": "task-auth",
+            "status": "in_progress"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-auth-plan"
+          }
+        }
+      ],
+      "totalPromptTokens": 2635,
+      "totalOutputTokens": 160,
+      "turns": 4,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M2-transition-attached",
+      "mechanism": "transition-handler-attached",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched taskmanager.update",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "update",
+          "params": {
+            "action": "update",
+            "taskId": "task-auth",
+            "status": "in_progress"
+          }
+        }
+      ],
+      "totalPromptTokens": 1731,
+      "totalOutputTokens": 77,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M3-transition-nudge",
+      "mechanism": "transition-handler-nudge",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": false,
+      "reason": "expected memory.list, not called",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Polish README: Improve Introduction Paragraph"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "add",
+          "params": {
+            "action": "add",
+            "title": "plan/polish_readme",
+            "content": "Rewrite the intro paragraph of the README to make it less robotic and more engaging."
+          }
+        }
+      ],
+      "totalPromptTokens": 2161,
+      "totalOutputTokens": 139,
+      "turns": 4,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M3-transition-nudge",
+      "mechanism": "transition-handler-nudge",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": false,
+      "reason": "expected memory.list, not called",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Polish README: Improve Introduction Paragraph"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "add",
+          "params": {
+            "action": "add",
+            "title": "plan/polish-readme",
+            "content": "Rewrite the intro paragraph of the README to make it less robotic and more engaging."
+          }
+        }
+      ],
+      "totalPromptTokens": 2161,
+      "totalOutputTokens": 138,
+      "turns": 4,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M4-post-compaction",
+      "mechanism": "post-compaction-nudge",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-decision"
+          }
+        }
+      ],
+      "totalPromptTokens": 979,
+      "totalOutputTokens": 30,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M4-post-compaction",
+      "mechanism": "post-compaction-nudge",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-decision"
+          }
+        }
+      ],
+      "totalPromptTokens": 979,
+      "totalOutputTokens": 30,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M5-append-evolution",
+      "mechanism": "append-action",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "matched memory.append",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "list",
+          "params": {
+            "action": "list",
+            "title": "plan/onboarding"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-plan"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "append",
+          "params": {
+            "action": "append",
+            "id": "mem-plan",
+            "content": "Each new hire also gets paired with a senior engineer for code reviews during the first month."
+          }
+        }
+      ],
+      "totalPromptTokens": 2112,
+      "totalOutputTokens": 117,
+      "turns": 4,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M5-append-evolution",
+      "mechanism": "append-action",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched memory.append",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "list",
+          "params": {
+            "action": "list",
+            "title": "plan/onboarding"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-plan"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "append",
+          "params": {
+            "action": "append",
+            "id": "mem-plan",
+            "content": "Each new hire also gets paired with a senior engineer for code reviews during the first month."
+          }
+        }
+      ],
+      "totalPromptTokens": 2112,
+      "totalOutputTokens": 117,
+      "turns": 4,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M6-negative-control",
+      "mechanism": "negative-control",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "no write — correct",
+      "callLog": [],
+      "totalPromptTokens": 346,
+      "totalOutputTokens": 16,
+      "turns": 1,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M6-negative-control",
+      "mechanism": "negative-control",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "no write — correct",
+      "callLog": [],
+      "totalPromptTokens": 346,
+      "totalOutputTokens": 16,
+      "turns": 1,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M1-catalog-recall",
+      "mechanism": "catalog-injection",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-rate-limits"
+          }
+        }
+      ],
+      "totalPromptTokens": 1025,
+      "totalOutputTokens": 165,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M1-catalog-recall",
+      "mechanism": "catalog-injection",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-rate-limits"
+          }
+        }
+      ],
+      "totalPromptTokens": 1025,
+      "totalOutputTokens": 114,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M2-transition-attached",
+      "mechanism": "transition-handler-attached",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "matched orAction memory.read on mem-auth-plan",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-auth-plan"
+          }
+        }
+      ],
+      "totalPromptTokens": 1012,
+      "totalOutputTokens": 88,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M2-transition-attached",
+      "mechanism": "transition-handler-attached",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched orAction memory.read on mem-auth-plan",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-auth-plan"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Implement Step 1 of auth migration: dual-issue tokens for 7 days"
+          }
+        }
+      ],
+      "totalPromptTokens": 1735,
+      "totalOutputTokens": 257,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M3-transition-nudge",
+      "mechanism": "transition-handler-nudge",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": false,
+      "reason": "expected memory.list, not called",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Polish the README: Rewrite the intro paragraph to feel less robotic"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "update",
+          "params": {
+            "action": "update",
+            "taskId": "task-readme",
+            "status": "in_progress"
+          }
+        }
+      ],
+      "totalPromptTokens": 2196,
+      "totalOutputTokens": 95,
+      "turns": 4,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M3-transition-nudge",
+      "mechanism": "transition-handler-nudge",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": false,
+      "reason": "expected memory.list, not called",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Polish the README: Rewrite the intro paragraph to feel less robotic"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "update",
+          "params": {
+            "action": "update",
+            "taskId": "task-readme",
+            "status": "in_progress"
+          }
+        }
+      ],
+      "totalPromptTokens": 2196,
+      "totalOutputTokens": 95,
+      "turns": 4,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M4-post-compaction",
+      "mechanism": "post-compaction-nudge",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": false,
+      "reason": "memory.read on wrong id: mem-decision  finding/retry-policy (wanted mem-decision)",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-decision  finding/retry-policy"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-decision"
+          }
+        }
+      ],
+      "totalPromptTokens": 2149,
+      "totalOutputTokens": 102,
+      "turns": 4,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M4-post-compaction",
+      "mechanism": "post-compaction-nudge",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-decision"
+          }
+        }
+      ],
+      "totalPromptTokens": 979,
+      "totalOutputTokens": 58,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M5-append-evolution",
+      "mechanism": "append-action",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "matched memory.append",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "append",
+          "params": {
+            "action": "append",
+            "id": "mem-plan",
+            "content": "Add step: Each new hire gets paired with a senior engineer for code reviews during the first month."
+          }
+        }
+      ],
+      "totalPromptTokens": 882,
+      "totalOutputTokens": 76,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M5-append-evolution",
+      "mechanism": "append-action",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched memory.append",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "append",
+          "params": {
+            "action": "append",
+            "id": "mem-plan",
+            "content": "Add step: Each new hire gets paired with a senior engineer for code reviews during the first month."
+          }
+        }
+      ],
+      "totalPromptTokens": 882,
+      "totalOutputTokens": 76,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M6-negative-control",
+      "mechanism": "negative-control",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "no write — correct",
+      "callLog": [],
+      "totalPromptTokens": 346,
+      "totalOutputTokens": 21,
+      "turns": 1,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M6-negative-control",
+      "mechanism": "negative-control",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "no write — correct",
+      "callLog": [],
+      "totalPromptTokens": 346,
+      "totalOutputTokens": 21,
+      "turns": 1,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M1-catalog-recall",
+      "mechanism": "catalog-injection",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-rate-limits"
+          }
+        }
+      ],
+      "totalPromptTokens": 1066,
+      "totalOutputTokens": 556,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M1-catalog-recall",
+      "mechanism": "catalog-injection",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-rate-limits"
+          }
+        }
+      ],
+      "totalPromptTokens": 1066,
+      "totalOutputTokens": 547,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M2-transition-attached",
+      "mechanism": "transition-handler-attached",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "matched taskmanager.update",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-auth-plan"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "update",
+          "params": {
+            "action": "update",
+            "taskId": "task-auth",
+            "status": "in_progress"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Step 1: Implement dual-issue tokens (7-day window)",
+            "status": "in_progress"
+          }
+        }
+      ],
+      "totalPromptTokens": 2312,
+      "totalOutputTokens": 702,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M2-transition-attached",
+      "mechanism": "transition-handler-attached",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched taskmanager.update",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-auth-plan"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "update",
+          "params": {
+            "action": "update",
+            "taskId": "task-auth",
+            "status": "in_progress"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Step 1a: Implement dual-token issuance in auth service"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Step 1b: Add v2 token validation (side-by-side with v1)"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Step 1c: Deploy and monitor for 7 days"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "append",
+          "params": {
+            "action": "append",
+            "id": "mem-auth-plan",
+            "content": "\n---\nActive work: Step 1 (dual-issue tokens). Actions: (a) update issuance to generate v1+v2, (b) add v2 verification alongside existing v1, (c) deploy and monitor 7 days. v1 verification must NOT be removed until Step 3 per hard requirement."
+          }
+        }
+      ],
+      "totalPromptTokens": 3851,
+      "totalOutputTokens": 1182,
+      "turns": 4,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M3-transition-nudge",
+      "mechanism": "transition-handler-nudge",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": false,
+      "reason": "expected memory.list, not called",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "update",
+          "params": {
+            "action": "update",
+            "taskId": "task-readme",
+            "status": "in_progress"
+          }
+        }
+      ],
+      "totalPromptTokens": 994,
+      "totalOutputTokens": 503,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M3-transition-nudge",
+      "mechanism": "transition-handler-nudge",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched orAction memory.read",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "update",
+          "params": {
+            "action": "update",
+            "taskId": "task-readme",
+            "status": "in_progress"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-style"
+          }
+        }
+      ],
+      "totalPromptTokens": 1655,
+      "totalOutputTokens": 493,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M4-post-compaction",
+      "mechanism": "post-compaction-nudge",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-decision"
+          }
+        }
+      ],
+      "totalPromptTokens": 1016,
+      "totalOutputTokens": 189,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M4-post-compaction",
+      "mechanism": "post-compaction-nudge",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-decision"
+          }
+        }
+      ],
+      "totalPromptTokens": 1016,
+      "totalOutputTokens": 189,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M5-append-evolution",
+      "mechanism": "append-action",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "matched memory.append",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-plan"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "append",
+          "params": {
+            "action": "append",
+            "id": "mem-plan",
+            "content": "First month: paired with senior engineer for code reviews."
+          }
+        }
+      ],
+      "totalPromptTokens": 1535,
+      "totalOutputTokens": 214,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M5-append-evolution",
+      "mechanism": "append-action",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched memory.append",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-plan"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "append",
+          "params": {
+            "action": "append",
+            "id": "mem-plan",
+            "content": " First month: paired with senior engineer for code reviews."
+          }
+        }
+      ],
+      "totalPromptTokens": 1535,
+      "totalOutputTokens": 238,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M6-negative-control",
+      "mechanism": "negative-control",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "no write — correct",
+      "callLog": [],
+      "totalPromptTokens": 367,
+      "totalOutputTokens": 91,
+      "turns": 1,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M6-negative-control",
+      "mechanism": "negative-control",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "no write — correct",
+      "callLog": [],
+      "totalPromptTokens": 367,
+      "totalOutputTokens": 68,
+      "turns": 1,
+      "error": null
+    }
+  ]
+}