onbuzz 4.8.2 → 4.9.0

package/scripts/memory-bench/results-baseline-2026-05-19T11-30-28-763Z.json

@@ -0,0 +1,1024 @@
+{
+  "timestamp": "2026-05-19T11:30:28.765Z",
+  "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": 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": 181,
+      "turns": 3,
+      "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": 54,
+      "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": 2643,
+      "totalOutputTokens": 150,
+      "turns": 4,
+      "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": 54,
+      "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 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": 2193,
+      "totalOutputTokens": 104,
+      "turns": 4,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-nano",
+      "scenario": "M3-transition-nudge",
+      "mechanism": "transition-handler-nudge",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": true,
+      "reason": "matched orAction memory.read",
+      "callLog": [
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Polish README"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "list",
+          "params": {
+            "action": "list"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "readme"
+          }
+        }
+      ],
+      "totalPromptTokens": 2080,
+      "totalOutputTokens": 114,
+      "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",
+            "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",
+            "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": 157,
+      "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": 123,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M2-transition-attached",
+      "mechanism": "transition-handler-attached",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": false,
+      "reason": "expected taskmanager.update, not called",
+      "callLog": [
+        {
+          "tool": "memory",
+          "action": "read",
+          "params": {
+            "action": "read",
+            "id": "mem-auth-plan"
+          }
+        }
+      ],
+      "totalPromptTokens": 1012,
+      "totalOutputTokens": 89,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "gpt-4.1-mini",
+      "scenario": "M2-transition-attached",
+      "mechanism": "transition-handler-attached",
+      "variant": "baseline",
+      "trial": 1,
+      "hit": false,
+      "reason": "expected taskmanager.update, not called",
+      "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"
+          }
+        },
+        {
+          "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": 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": 1,
+      "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": 96,
+      "turns": 4,
+      "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": "M6-negative-control",
+      "mechanism": "negative-control",
+      "variant": "baseline",
+      "trial": 0,
+      "hit": true,
+      "reason": "no write — correct",
+      "callLog": [],
+      "totalPromptTokens": 346,
+      "totalOutputTokens": 24,
+      "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": 554,
+      "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": 640,
+      "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 1a: Implement dual-token issuance in auth service"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Step 1b: Verify v1 token validation remains intact"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Step 1c: Monitor dual-issue rollout for 7 days"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "append",
+          "params": {
+            "action": "append",
+            "id": "mem-auth-plan",
+            "content": "\n[Progress] Step 1 initiated. Plan: (a) update issuance to dual-issue v1+v2, (b) confirm v1 validation untouched, (c) monitor for 7-day window before proceeding to Step 2."
+          }
+        }
+      ],
+      "totalPromptTokens": 3806,
+      "totalOutputTokens": 1117,
+      "turns": 4,
+      "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 token issuance (7-day window)",
+            "status": "in_progress"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Step 2: Switch read paths to new token verification",
+            "status": "pending"
+          }
+        },
+        {
+          "tool": "taskmanager",
+          "action": "create",
+          "params": {
+            "action": "create",
+            "title": "Step 3: Deprecate v1 token verification",
+            "status": "pending"
+          }
+        },
+        {
+          "tool": "memory",
+          "action": "add",
+          "params": {
+            "action": "add",
+            "title": "plan/auth-migration-step1",
+            "content": "Step 1: Dual-issue tokens for 7 days\n\nObjectives:\n- Modify token issuance endpoint to generate BOTH v1 and v2 tokens\n- Return both tokens in response (or v2 primary + v1 backward-compat header)\n- Ensure v1 verification remains fully operational (hard requirement)\n- Log/metric v2 token usage for observability\n- Set 7-day calendar reminder before proceeding to Step 2\n\nKey risks to mitigate:\n- Payload size increase (two tokens instead of one)\n- Client confusion over which token to use\n- Secret/key management for v2 signing\n\nAcceptance criteria:\n- 100% of new sessions receive v2 token\n- 100% of new sessions still receive usable v1 token\n- Zero errors in v1 verification path\n- Monitoring dashboard shows v2 issuance rate"
+          }
+        }
+      ],
+      "totalPromptTokens": 3912,
+      "totalOutputTokens": 1177,
+      "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": 399,
+      "turns": 2,
+      "error": null
+    },
+    {
+      "model": "Kimi-K2.6",
+      "scenario": "M3-transition-nudge",
+      "mechanism": "transition-handler-nudge",
+      "variant": "baseline",
+      "trial": 1,
+      "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": 596,
+      "turns": 2,
+      "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": 192,
+      "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": "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": 254,
+      "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": 247,
+      "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": 94,
+      "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": 49,
+      "turns": 1,
+      "error": null
+    }
+  ]
+}