onbuzz 4.8.3 → 4.9.1

package/scripts/memory-bench/results-baseline-2026-05-19T11-44-18-054Z.json

@@ -0,0 +1,1432 @@
+{
+  "timestamp": "2026-05-19T11:44:18.057Z",
+  "args": {
+    "models": [
+      "gpt-4.1-nano",
+      "gpt-4.1-mini",
+      "Kimi-K2.6"
+    ],
+    "trials": 3,
+    "scenarios": null,
+    "variant": "baseline",
+    "out": null
+  },
+  "results": [
+    {
+      "model": "gpt-4.1-nano",
+      "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": 1077,
+      "totalOutputTokens": 161,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M1-catalog-recall",
+      "mechanism": "catalog-injection",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-rate-limits"
+          }
+        }
+      ],
+      "totalPromptTokens": 1800,
+      "totalOutputTokens": 183,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M1-catalog-recall",
+      "mechanism": "catalog-injection",
+      "variant": "baseline",
+      "trial": 2,
+      "hit": false,
+      "reason": "expected memory.read, not called",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        }
+      ],
+      "totalPromptTokens": 1041,
+      "totalOutputTokens": 172,
+      "turns": 2,
+      "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"
+          }
+        }
+      ],
+      "totalPromptTokens": 1731,
+      "totalOutputTokens": 92,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M2-transition-attached",
+      "mechanism": "transition-handler-attached",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": false,
+      "reason": "expected taskmanager.update, not called",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        }
+      ],
+      "totalPromptTokens": 945,
+      "totalOutputTokens": 53,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M2-transition-attached",
+      "mechanism": "transition-handler-attached",
+      "variant": "baseline",
+      "trial": 2,
+      "hit": false,
+      "reason": "expected taskmanager.update, not called",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        }
+      ],
+      "totalPromptTokens": 945,
+      "totalOutputTokens": 57,
+      "turns": 2,
+      "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: Rewrite intro paragraph to be less robotic"
+          }
+        }
+      ],
+      "totalPromptTokens": 867,
+      "totalOutputTokens": 90,
+      "turns": 2,
+      "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: Rewrite intro paragraph to be less robotic"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "update",
+          "params": {
+            "action": "update",
+            "taskId": "task-readme",
+            "status": "pending"
+          }
+        }
+      ],
+      "totalPromptTokens": 1462,
+      "totalOutputTokens": 110,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M3-transition-nudge",
+      "mechanism": "transition-handler-nudge",
+      "variant": "baseline",
+      "trial": 2,
+      "hit": false,
+      "reason": "expected memory.list, not called",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Polish README: Rewrite intro paragraph to be less robotic"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        }
+      ],
+      "totalPromptTokens": 1485,
+      "totalOutputTokens": 117,
+      "turns": 3,
+      "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": 38,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M4-post-compaction",
+      "mechanism": "post-compaction-nudge",
+      "variant": "baseline",
+      "trial": 2,
+      "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",
+            "id": "mem-plan"
+          }
+        },
+        {
+          "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": 2109,
+      "totalOutputTokens": 91,
+      "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",
+            "id": "mem-plan"
+          }
+        },
+        {
+          "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": 2109,
+      "totalOutputTokens": 91,
+      "turns": 4,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M5-append-evolution",
+      "mechanism": "append-action",
+      "variant": "baseline",
+      "trial": 2,
+      "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-nano",
+      "scenario": "M6-negative-control",
+      "mechanism": "negative-control",
+      "variant": "baseline",
+      "trial": 2,
+      "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": 137,
+      "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": 117,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M1-catalog-recall",
+      "mechanism": "catalog-injection",
+      "variant": "baseline",
+      "trial": 2,
+      "hit": false,
+      "reason": "expected memory.read, not called",
+      "callLog": [],
+      "totalPromptTokens": 420,
+      "totalOutputTokens": 88,
+      "turns": 1,
+      "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"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Implement Step 1 of auth migration: dual-issue tokens for 7 days",
+            "status": "in_progress"
+          }
+        }
+      ],
+      "totalPromptTokens": 1736,
+      "totalOutputTokens": 209,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "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": "create",
+          "params": {
+            "action": "create",
+            "title": "Implement Step 1 of auth migration: dual-issue tokens for 7 days",
+            "status": "in_progress"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "update",
+          "params": {
+            "action": "update",
+            "taskId": "task-auth",
+            "status": "in_progress"
+          }
+        }
+      ],
+      "totalPromptTokens": 3747,
+      "totalOutputTokens": 324,
+      "turns": 5,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M2-transition-attached",
+      "mechanism": "transition-handler-attached",
+      "variant": "baseline",
+      "trial": 2,
+      "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": 86,
+      "turns": 2,
+      "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"
+          }
+        }
+      ],
+      "totalPromptTokens": 869,
+      "totalOutputTokens": 77,
+      "turns": 2,
+      "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": 106,
+      "turns": 4,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M3-transition-nudge",
+      "mechanism": "transition-handler-nudge",
+      "variant": "baseline",
+      "trial": 2,
+      "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"
+          }
+        }
+      ],
+      "totalPromptTokens": 869,
+      "totalOutputTokens": 81,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "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": 54,
+      "turns": 2,
+      "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": 50,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M4-post-compaction",
+      "mechanism": "post-compaction-nudge",
+      "variant": "baseline",
+      "trial": 2,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-decision"
+          }
+        }
+      ],
+      "totalPromptTokens": 979,
+      "totalOutputTokens": 54,
+      "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": "Each new hire also gets paired with a senior engineer for code reviews for the first month."
+          }
+        }
+      ],
+      "totalPromptTokens": 880,
+      "totalOutputTokens": 74,
+      "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": "Each new hire also gets paired with a senior engineer for code reviews for the first month."
+          }
+        }
+      ],
+      "totalPromptTokens": 880,
+      "totalOutputTokens": 74,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M5-append-evolution",
+      "mechanism": "append-action",
+      "variant": "baseline",
+      "trial": 2,
+      "hit": true,
+      "reason": "matched memory.append",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "append",
+          "params": {
+            "action": "append",
+            "id": "mem-plan",
+            "content": "Each new hire also gets paired with a senior engineer for code reviews for the first month."
+          }
+        }
+      ],
+      "totalPromptTokens": 880,
+      "totalOutputTokens": 74,
+      "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": "gpt-4.1-mini",
+      "scenario": "M6-negative-control",
+      "mechanism": "negative-control",
+      "variant": "baseline",
+      "trial": 2,
+      "hit": true,
+      "reason": "no write — correct",
+      "callLog": [],
+      "totalPromptTokens": 346,
+      "totalOutputTokens": 24,
+      "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": 551,
+      "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": 503,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M1-catalog-recall",
+      "mechanism": "catalog-injection",
+      "variant": "baseline",
+      "trial": 2,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-rate-limits"
+          }
+        }
+      ],
+      "totalPromptTokens": 1066,
+      "totalOutputTokens": 536,
+      "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": "Auth Step 1: Implement dual-issue tokens (7-day window)"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Auth Step 2: Switch token reads to new path"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Auth Step 3: Deprecate v1 verification"
+          }
+        }
+      ],
+      "totalPromptTokens": 2464,
+      "totalOutputTokens": 946,
+      "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 1: Implement dual-issue tokens (7 days)",
+            "status": "in_progress"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Step 2: Switch reads to new auth path",
+            "status": "pending"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Step 3: Deprecate v1 token verification",
+            "status": "pending"
+          }
+        }
+      ],
+      "totalPromptTokens": 2475,
+      "totalOutputTokens": 946,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M2-transition-attached",
+      "mechanism": "transition-handler-attached",
+      "variant": "baseline",
+      "trial": 2,
+      "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": 2329,
+      "totalOutputTokens": 781,
+      "turns": 3,
+      "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": 469,
+      "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": 657,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M3-transition-nudge",
+      "mechanism": "transition-handler-nudge",
+      "variant": "baseline",
+      "trial": 2,
+      "hit": true,
+      "reason": "matched memory.list",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "update",
+          "params": {
+            "action": "update",
+            "taskId": "task-readme",
+            "status": "in_progress"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        }
+      ],
+      "totalPromptTokens": 1633,
+      "totalOutputTokens": 972,
+      "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": 182,
+      "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": 183,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M4-post-compaction",
+      "mechanism": "post-compaction-nudge",
+      "variant": "baseline",
+      "trial": 2,
+      "hit": true,
+      "reason": "matched memory.read",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-decision"
+          }
+        }
+      ],
+      "totalPromptTokens": 1016,
+      "totalOutputTokens": 184,
+      "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": 230,
+      "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 a senior engineer for code reviews."
+          }
+        }
+      ],
+      "totalPromptTokens": 1536,
+      "totalOutputTokens": 254,
+      "turns": 3,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M5-append-evolution",
+      "mechanism": "append-action",
+      "variant": "baseline",
+      "trial": 2,
+      "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 a senior engineer for code reviews."
+          }
+        }
+      ],
+      "totalPromptTokens": 1536,
+      "totalOutputTokens": 285,
+      "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": 49,
+      "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": 76,
+      "turns": 1,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M6-negative-control",
+      "mechanism": "negative-control",
+      "variant": "baseline",
+      "trial": 2,
+      "hit": true,
+      "reason": "no write — correct",
+      "callLog": [],
+      "totalPromptTokens": 367,
+      "totalOutputTokens": 51,
+      "turns": 1,
+      "error": null
+    }
+  ]
+}