{
  "meta": {
    "repo": "agent-platform",
    "demo": "m7-ticket-pipeline",
    "commit": "3c0ce8f",
    "worktree_dirty": false,
    "dirty_paths": [],
    "generated_at": "2026-10-22T07:47:11+00:00",
    "mode": "offline-scripted",
    "seed": 20261011,
    "model": "qwen3:8b",
    "python": "3.12.3",
    "source_result": "results/m7_e2e.json",
    "note": "静态回放是一次确定性离线运行的产物；本地运行 python3 demo/server.py 可实时重算。"
  },
  "summary": {
    "tickets": 10,
    "success": 9,
    "failed": 1,
    "failure_kinds": {
      "silent_empty_answer": 1
    },
    "latency_ms": {
      "p50": 30.0,
      "p95": 34.6,
      "max": 36,
      "p50_ci95": [
        28.0,
        33.0
      ]
    },
    "context_tokens": {
      "compacted_sum": 620,
      "uncompacted_sum": 1181,
      "saved_tokens": 561,
      "saved_pct_mean": 0.4752
    },
    "tokens": {
      "prompt_sum": 1298,
      "completion_sum": 149
    },
    "flags": {
      "answered_without_tool_data": 1,
      "tool_denied": 1,
      "empty_content": 1,
      "degraded": 1
    },
    "pinned": 10,
    "tool_verdicts": {
      "denied": 1,
      "ok": 5
    }
  },
  "items": [
    {
      "id": "T00",
      "text": "我的订单到哪了？",
      "scenario": "正常路径",
      "tool": "query_order",
      "tool_verdict": "ok",
      "llm_status": "ok",
      "answer": "您的订单已发货，仅向您核对后四位 0042。",
      "answer_len": 22,
      "latency_ms": 33,
      "failure": "",
      "flags": [],
      "ctx_tokens": 60,
      "ctx_tokens_uncompacted": 116,
      "saved_tokens": 56,
      "saved_pct": 0.4828,
      "pinned": 1,
      "stages": [
        {
          "name": "工单进入",
          "key": "intake",
          "status": "ok",
          "detail": "我的订单到哪了？",
          "metrics": [
            {
              "label": "字符数",
              "value": 8,
              "source": "m7_e2e.TICKETS"
            }
          ]
        },
        {
          "name": "工具沙箱",
          "key": "tool",
          "status": "ok",
          "detail": "query_order → ok：whitelisted",
          "metrics": [
            {
              "label": "判定",
              "value": "ok",
              "source": "tool_audit.verdict (request_id=T00)"
            },
            {
              "label": "尝试次数",
              "value": 1,
              "source": "tool_audit rows"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "tool_audit.latency_ms"
            }
          ]
        },
        {
          "name": "记忆压缩",
          "key": "memory",
          "status": "ok",
          "detail": "滑窗 k=3；pinned 合规约束仍在上下文：合规要求：任何回复都不得包含完整订单号，只能给后四位。",
          "metrics": [
            {
              "label": "压缩后",
              "value": "60 tokens",
              "source": "m7_e2e.TicketResult.context_tokens"
            },
            {
              "label": "压缩前",
              "value": "116 tokens",
              "source": "m7_e2e.TicketResult.context_tokens_uncompacted"
            },
            {
              "label": "节省",
              "value": "56 tokens (48%)",
              "source": "压缩前 - 压缩后"
            },
            {
              "label": "pinned",
              "value": 1,
              "source": "memory_item.pinned"
            },
            {
              "label": "会话轮次",
              "value": 7,
              "source": "memory_item.session_id"
            }
          ]
        },
        {
          "name": "网关 / LLM",
          "key": "gateway",
          "status": "ok",
          "detail": "model=qwen3:8b；route=default；cache_hit=False；degraded=False",
          "metrics": [
            {
              "label": "prompt",
              "value": "134 tokens",
              "source": "gateway_request.prompt_tokens"
            },
            {
              "label": "completion",
              "value": "18 tokens",
              "source": "gateway_request.completion_tokens"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "gateway_request.latency_ms"
            },
            {
              "label": "成本",
              "value": "$0.000000",
              "source": "gateway_request.cost_usd"
            },
            {
              "label": "网关记录",
              "value": 1,
              "source": "gateway_request.idempotency_key"
            }
          ]
        },
        {
          "name": "失败分类",
          "key": "classification",
          "status": "ok",
          "detail": "未分类失败 0",
          "metrics": [
            {
              "label": "flags",
              "value": "—",
              "source": "m7_e2e.TicketResult.flags"
            }
          ]
        },
        {
          "name": "回答",
          "key": "answer",
          "status": "ok",
          "detail": "您的订单已发货，仅向您核对后四位 0042。",
          "metrics": [
            {
              "label": "回答长度",
              "value": "22 字",
              "source": "gateway_request.content"
            }
          ]
        }
      ],
      "sources": {
        "answer": {
          "table": "gateway_request",
          "field": "content",
          "key": "T00-ok"
        },
        "gateway_status": {
          "table": "gateway_request",
          "field": "status",
          "key": "T00-ok"
        },
        "tool_verdict": {
          "table": "tool_audit",
          "field": "verdict",
          "key": "T00"
        },
        "pinned": {
          "table": "memory_item",
          "field": "pinned",
          "key": "sess-T00"
        },
        "summary": {
          "artifact": "results/m7_e2e.json#/summary"
        }
      }
    },
    {
      "id": "T01",
      "text": "这个订单能退款吗？",
      "scenario": "正常路径",
      "tool": "query_order",
      "tool_verdict": "ok",
      "llm_status": "ok",
      "answer": "您的订单已发货，仅向您核对后四位 0042。",
      "answer_len": 22,
      "latency_ms": 30,
      "failure": "",
      "flags": [],
      "ctx_tokens": 61,
      "ctx_tokens_uncompacted": 117,
      "saved_tokens": 56,
      "saved_pct": 0.4786,
      "pinned": 1,
      "stages": [
        {
          "name": "工单进入",
          "key": "intake",
          "status": "ok",
          "detail": "这个订单能退款吗？",
          "metrics": [
            {
              "label": "字符数",
              "value": 9,
              "source": "m7_e2e.TICKETS"
            }
          ]
        },
        {
          "name": "工具沙箱",
          "key": "tool",
          "status": "ok",
          "detail": "query_order → ok：whitelisted",
          "metrics": [
            {
              "label": "判定",
              "value": "ok",
              "source": "tool_audit.verdict (request_id=T01)"
            },
            {
              "label": "尝试次数",
              "value": 1,
              "source": "tool_audit rows"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "tool_audit.latency_ms"
            }
          ]
        },
        {
          "name": "记忆压缩",
          "key": "memory",
          "status": "ok",
          "detail": "滑窗 k=3；pinned 合规约束仍在上下文：合规要求：任何回复都不得包含完整订单号，只能给后四位。",
          "metrics": [
            {
              "label": "压缩后",
              "value": "61 tokens",
              "source": "m7_e2e.TicketResult.context_tokens"
            },
            {
              "label": "压缩前",
              "value": "117 tokens",
              "source": "m7_e2e.TicketResult.context_tokens_uncompacted"
            },
            {
              "label": "节省",
              "value": "56 tokens (48%)",
              "source": "压缩前 - 压缩后"
            },
            {
              "label": "pinned",
              "value": 1,
              "source": "memory_item.pinned"
            },
            {
              "label": "会话轮次",
              "value": 7,
              "source": "memory_item.session_id"
            }
          ]
        },
        {
          "name": "网关 / LLM",
          "key": "gateway",
          "status": "ok",
          "detail": "model=qwen3:8b；route=default；cache_hit=False；degraded=False",
          "metrics": [
            {
              "label": "prompt",
              "value": "135 tokens",
              "source": "gateway_request.prompt_tokens"
            },
            {
              "label": "completion",
              "value": "18 tokens",
              "source": "gateway_request.completion_tokens"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "gateway_request.latency_ms"
            },
            {
              "label": "成本",
              "value": "$0.000000",
              "source": "gateway_request.cost_usd"
            },
            {
              "label": "网关记录",
              "value": 1,
              "source": "gateway_request.idempotency_key"
            }
          ]
        },
        {
          "name": "失败分类",
          "key": "classification",
          "status": "ok",
          "detail": "未分类失败 0",
          "metrics": [
            {
              "label": "flags",
              "value": "—",
              "source": "m7_e2e.TicketResult.flags"
            }
          ]
        },
        {
          "name": "回答",
          "key": "answer",
          "status": "ok",
          "detail": "您的订单已发货，仅向您核对后四位 0042。",
          "metrics": [
            {
              "label": "回答长度",
              "value": "22 字",
              "source": "gateway_request.content"
            }
          ]
        }
      ],
      "sources": {
        "answer": {
          "table": "gateway_request",
          "field": "content",
          "key": "T01-ok"
        },
        "gateway_status": {
          "table": "gateway_request",
          "field": "status",
          "key": "T01-ok"
        },
        "tool_verdict": {
          "table": "tool_audit",
          "field": "verdict",
          "key": "T01"
        },
        "pinned": {
          "table": "memory_item",
          "field": "pinned",
          "key": "sess-T01"
        },
        "summary": {
          "artifact": "results/m7_e2e.json#/summary"
        }
      }
    },
    {
      "id": "T02",
      "text": "你们的退货政策是什么？",
      "scenario": "脚本注入 timeout：primary 失败，网关走 fallback，结果标记 degraded。",
      "tool": "",
      "tool_verdict": "",
      "llm_status": "degraded",
      "answer": "退货政策为签收后七天无理由。",
      "answer_len": 14,
      "latency_ms": 29,
      "failure": "",
      "flags": [
        "degraded"
      ],
      "ctx_tokens": 63,
      "ctx_tokens_uncompacted": 119,
      "saved_tokens": 56,
      "saved_pct": 0.4706,
      "pinned": 1,
      "stages": [
        {
          "name": "工单进入",
          "key": "intake",
          "status": "ok",
          "detail": "你们的退货政策是什么？",
          "metrics": [
            {
              "label": "字符数",
              "value": 11,
              "source": "m7_e2e.TICKETS"
            }
          ]
        },
        {
          "name": "工具沙箱",
          "key": "tool",
          "status": "skipped",
          "detail": "无需工具：策略/知识类问题，未进入工具调用路径。",
          "metrics": [
            {
              "label": "判定",
              "value": "skipped",
              "source": "m7_e2e.tool_for_ticket"
            }
          ]
        },
        {
          "name": "记忆压缩",
          "key": "memory",
          "status": "ok",
          "detail": "滑窗 k=3；pinned 合规约束仍在上下文：合规要求：任何回复都不得包含完整订单号，只能给后四位。",
          "metrics": [
            {
              "label": "压缩后",
              "value": "63 tokens",
              "source": "m7_e2e.TicketResult.context_tokens"
            },
            {
              "label": "压缩前",
              "value": "119 tokens",
              "source": "m7_e2e.TicketResult.context_tokens_uncompacted"
            },
            {
              "label": "节省",
              "value": "56 tokens (47%)",
              "source": "压缩前 - 压缩后"
            },
            {
              "label": "pinned",
              "value": 1,
              "source": "memory_item.pinned"
            },
            {
              "label": "会话轮次",
              "value": 7,
              "source": "memory_item.session_id"
            }
          ]
        },
        {
          "name": "网关 / LLM",
          "key": "gateway",
          "status": "degraded",
          "detail": "model=llama3.2:3b；route=default；cache_hit=False；degraded=True",
          "metrics": [
            {
              "label": "prompt",
              "value": "121 tokens",
              "source": "gateway_request.prompt_tokens"
            },
            {
              "label": "completion",
              "value": "15 tokens",
              "source": "gateway_request.completion_tokens"
            },
            {
              "label": "耗时",
              "value": "2 ms",
              "source": "gateway_request.latency_ms"
            },
            {
              "label": "成本",
              "value": "$0.000000",
              "source": "gateway_request.cost_usd"
            },
            {
              "label": "网关记录",
              "value": 1,
              "source": "gateway_request.idempotency_key"
            }
          ]
        },
        {
          "name": "失败分类",
          "key": "classification",
          "status": "ok",
          "detail": "未分类失败 0",
          "metrics": [
            {
              "label": "flags",
              "value": "degraded",
              "source": "m7_e2e.TicketResult.flags"
            }
          ]
        },
        {
          "name": "回答",
          "key": "answer",
          "status": "ok",
          "detail": "退货政策为签收后七天无理由。",
          "metrics": [
            {
              "label": "回答长度",
              "value": "14 字",
              "source": "gateway_request.content"
            }
          ]
        }
      ],
      "sources": {
        "answer": {
          "table": "gateway_request",
          "field": "content",
          "key": "T02-"
        },
        "gateway_status": {
          "table": "gateway_request",
          "field": "status",
          "key": "T02-"
        },
        "tool_verdict": {
          "table": "tool_audit",
          "field": "verdict",
          "key": "T02"
        },
        "pinned": {
          "table": "memory_item",
          "field": "pinned",
          "key": "sess-T02"
        },
        "summary": {
          "artifact": "results/m7_e2e.json#/summary"
        }
      }
    },
    {
      "id": "T03",
      "text": "订单被别人签收了怎么办？",
      "scenario": "正常路径",
      "tool": "query_order",
      "tool_verdict": "ok",
      "llm_status": "ok",
      "answer": "您的订单已发货，仅向您核对后四位 0042。",
      "answer_len": 22,
      "latency_ms": 33,
      "failure": "",
      "flags": [],
      "ctx_tokens": 64,
      "ctx_tokens_uncompacted": 120,
      "saved_tokens": 56,
      "saved_pct": 0.4667,
      "pinned": 1,
      "stages": [
        {
          "name": "工单进入",
          "key": "intake",
          "status": "ok",
          "detail": "订单被别人签收了怎么办？",
          "metrics": [
            {
              "label": "字符数",
              "value": 12,
              "source": "m7_e2e.TICKETS"
            }
          ]
        },
        {
          "name": "工具沙箱",
          "key": "tool",
          "status": "ok",
          "detail": "query_order → ok：whitelisted",
          "metrics": [
            {
              "label": "判定",
              "value": "ok",
              "source": "tool_audit.verdict (request_id=T03)"
            },
            {
              "label": "尝试次数",
              "value": 1,
              "source": "tool_audit rows"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "tool_audit.latency_ms"
            }
          ]
        },
        {
          "name": "记忆压缩",
          "key": "memory",
          "status": "ok",
          "detail": "滑窗 k=3；pinned 合规约束仍在上下文：合规要求：任何回复都不得包含完整订单号，只能给后四位。",
          "metrics": [
            {
              "label": "压缩后",
              "value": "64 tokens",
              "source": "m7_e2e.TicketResult.context_tokens"
            },
            {
              "label": "压缩前",
              "value": "120 tokens",
              "source": "m7_e2e.TicketResult.context_tokens_uncompacted"
            },
            {
              "label": "节省",
              "value": "56 tokens (47%)",
              "source": "压缩前 - 压缩后"
            },
            {
              "label": "pinned",
              "value": 1,
              "source": "memory_item.pinned"
            },
            {
              "label": "会话轮次",
              "value": 7,
              "source": "memory_item.session_id"
            }
          ]
        },
        {
          "name": "网关 / LLM",
          "key": "gateway",
          "status": "ok",
          "detail": "model=qwen3:8b；route=default；cache_hit=False；degraded=False",
          "metrics": [
            {
              "label": "prompt",
              "value": "142 tokens",
              "source": "gateway_request.prompt_tokens"
            },
            {
              "label": "completion",
              "value": "18 tokens",
              "source": "gateway_request.completion_tokens"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "gateway_request.latency_ms"
            },
            {
              "label": "成本",
              "value": "$0.000000",
              "source": "gateway_request.cost_usd"
            },
            {
              "label": "网关记录",
              "value": 1,
              "source": "gateway_request.idempotency_key"
            }
          ]
        },
        {
          "name": "失败分类",
          "key": "classification",
          "status": "ok",
          "detail": "未分类失败 0",
          "metrics": [
            {
              "label": "flags",
              "value": "—",
              "source": "m7_e2e.TicketResult.flags"
            }
          ]
        },
        {
          "name": "回答",
          "key": "answer",
          "status": "ok",
          "detail": "您的订单已发货，仅向您核对后四位 0042。",
          "metrics": [
            {
              "label": "回答长度",
              "value": "22 字",
              "source": "gateway_request.content"
            }
          ]
        }
      ],
      "sources": {
        "answer": {
          "table": "gateway_request",
          "field": "content",
          "key": "T03-ok"
        },
        "gateway_status": {
          "table": "gateway_request",
          "field": "status",
          "key": "T03-ok"
        },
        "tool_verdict": {
          "table": "tool_audit",
          "field": "verdict",
          "key": "T03"
        },
        "pinned": {
          "table": "memory_item",
          "field": "pinned",
          "key": "sess-T03"
        },
        "summary": {
          "artifact": "results/m7_e2e.json#/summary"
        }
      }
    },
    {
      "id": "T04",
      "text": "发票能重开吗？",
      "scenario": "脚本注入空 content + done_reason=length：HTTP 200 但回答为空，端到端记为 silent_empty_answer。",
      "tool": "",
      "tool_verdict": "",
      "llm_status": "empty",
      "answer": "",
      "answer_len": 0,
      "latency_ms": 28,
      "failure": "silent_empty_answer",
      "flags": [
        "empty_content"
      ],
      "ctx_tokens": 59,
      "ctx_tokens_uncompacted": 115,
      "saved_tokens": 56,
      "saved_pct": 0.487,
      "pinned": 1,
      "stages": [
        {
          "name": "工单进入",
          "key": "intake",
          "status": "ok",
          "detail": "发票能重开吗？",
          "metrics": [
            {
              "label": "字符数",
              "value": 7,
              "source": "m7_e2e.TICKETS"
            }
          ]
        },
        {
          "name": "工具沙箱",
          "key": "tool",
          "status": "skipped",
          "detail": "无需工具：策略/知识类问题，未进入工具调用路径。",
          "metrics": [
            {
              "label": "判定",
              "value": "skipped",
              "source": "m7_e2e.tool_for_ticket"
            }
          ]
        },
        {
          "name": "记忆压缩",
          "key": "memory",
          "status": "ok",
          "detail": "滑窗 k=3；pinned 合规约束仍在上下文：合规要求：任何回复都不得包含完整订单号，只能给后四位。",
          "metrics": [
            {
              "label": "压缩后",
              "value": "59 tokens",
              "source": "m7_e2e.TicketResult.context_tokens"
            },
            {
              "label": "压缩前",
              "value": "115 tokens",
              "source": "m7_e2e.TicketResult.context_tokens_uncompacted"
            },
            {
              "label": "节省",
              "value": "56 tokens (49%)",
              "source": "压缩前 - 压缩后"
            },
            {
              "label": "pinned",
              "value": 1,
              "source": "memory_item.pinned"
            },
            {
              "label": "会话轮次",
              "value": 7,
              "source": "memory_item.session_id"
            }
          ]
        },
        {
          "name": "网关 / LLM",
          "key": "gateway",
          "status": "empty",
          "detail": "model=qwen3:8b；route=default；cache_hit=False；degraded=False；error_kind=empty_content",
          "metrics": [
            {
              "label": "prompt",
              "value": "113 tokens",
              "source": "gateway_request.prompt_tokens"
            },
            {
              "label": "completion",
              "value": "0 tokens",
              "source": "gateway_request.completion_tokens"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "gateway_request.latency_ms"
            },
            {
              "label": "成本",
              "value": "$0.000000",
              "source": "gateway_request.cost_usd"
            },
            {
              "label": "网关记录",
              "value": 1,
              "source": "gateway_request.idempotency_key"
            }
          ]
        },
        {
          "name": "失败分类",
          "key": "classification",
          "status": "error",
          "detail": "silent_empty_answer",
          "metrics": [
            {
              "label": "flags",
              "value": "empty_content",
              "source": "m7_e2e.TicketResult.flags"
            }
          ]
        },
        {
          "name": "回答",
          "key": "answer",
          "status": "error",
          "detail": "(空回答：HTTP 200 但 content 为空)",
          "metrics": [
            {
              "label": "回答长度",
              "value": "0 字",
              "source": "gateway_request.content"
            }
          ]
        }
      ],
      "sources": {
        "answer": {
          "table": "gateway_request",
          "field": "content",
          "key": "T04-"
        },
        "gateway_status": {
          "table": "gateway_request",
          "field": "status",
          "key": "T04-"
        },
        "tool_verdict": {
          "table": "tool_audit",
          "field": "verdict",
          "key": "T04"
        },
        "pinned": {
          "table": "memory_item",
          "field": "pinned",
          "key": "sess-T04"
        },
        "summary": {
          "artifact": "results/m7_e2e.json#/summary"
        }
      }
    },
    {
      "id": "T05",
      "text": "我要投诉，订单一直没发货。",
      "scenario": "模型尝试调用 issue_refund；它不在白名单，沙箱拒绝，但模型仍给出不带工具数据的回答。",
      "tool": "issue_refund",
      "tool_verdict": "denied",
      "llm_status": "ok",
      "answer": "您的订单已发货，仅向您核对后四位 0042。",
      "answer_len": 22,
      "latency_ms": 32,
      "failure": "",
      "flags": [
        "tool_denied",
        "answered_without_tool_data"
      ],
      "ctx_tokens": 64,
      "ctx_tokens_uncompacted": 121,
      "saved_tokens": 57,
      "saved_pct": 0.4711,
      "pinned": 1,
      "stages": [
        {
          "name": "工单进入",
          "key": "intake",
          "status": "ok",
          "detail": "我要投诉，订单一直没发货。",
          "metrics": [
            {
              "label": "字符数",
              "value": 13,
              "source": "m7_e2e.TICKETS"
            }
          ]
        },
        {
          "name": "工具沙箱",
          "key": "tool",
          "status": "error",
          "detail": "issue_refund → denied：tool 'issue_refund' 不在白名单",
          "metrics": [
            {
              "label": "判定",
              "value": "denied",
              "source": "tool_audit.verdict (request_id=T05)"
            },
            {
              "label": "尝试次数",
              "value": 1,
              "source": "tool_audit rows"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "tool_audit.latency_ms"
            }
          ]
        },
        {
          "name": "记忆压缩",
          "key": "memory",
          "status": "ok",
          "detail": "滑窗 k=3；pinned 合规约束仍在上下文：合规要求：任何回复都不得包含完整订单号，只能给后四位。",
          "metrics": [
            {
              "label": "压缩后",
              "value": "64 tokens",
              "source": "m7_e2e.TicketResult.context_tokens"
            },
            {
              "label": "压缩前",
              "value": "121 tokens",
              "source": "m7_e2e.TicketResult.context_tokens_uncompacted"
            },
            {
              "label": "节省",
              "value": "57 tokens (47%)",
              "source": "压缩前 - 压缩后"
            },
            {
              "label": "pinned",
              "value": 1,
              "source": "memory_item.pinned"
            },
            {
              "label": "会话轮次",
              "value": 7,
              "source": "memory_item.session_id"
            }
          ]
        },
        {
          "name": "网关 / LLM",
          "key": "gateway",
          "status": "ok",
          "detail": "model=qwen3:8b；route=default；cache_hit=False；degraded=False",
          "metrics": [
            {
              "label": "prompt",
              "value": "140 tokens",
              "source": "gateway_request.prompt_tokens"
            },
            {
              "label": "completion",
              "value": "18 tokens",
              "source": "gateway_request.completion_tokens"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "gateway_request.latency_ms"
            },
            {
              "label": "成本",
              "value": "$0.000000",
              "source": "gateway_request.cost_usd"
            },
            {
              "label": "网关记录",
              "value": 1,
              "source": "gateway_request.idempotency_key"
            }
          ]
        },
        {
          "name": "失败分类",
          "key": "classification",
          "status": "ok",
          "detail": "未分类失败 0",
          "metrics": [
            {
              "label": "flags",
              "value": "tool_denied, answered_without_tool_data",
              "source": "m7_e2e.TicketResult.flags"
            }
          ]
        },
        {
          "name": "回答",
          "key": "answer",
          "status": "ok",
          "detail": "您的订单已发货，仅向您核对后四位 0042。",
          "metrics": [
            {
              "label": "回答长度",
              "value": "22 字",
              "source": "gateway_request.content"
            }
          ]
        }
      ],
      "sources": {
        "answer": {
          "table": "gateway_request",
          "field": "content",
          "key": "T05-denied"
        },
        "gateway_status": {
          "table": "gateway_request",
          "field": "status",
          "key": "T05-denied"
        },
        "tool_verdict": {
          "table": "tool_audit",
          "field": "verdict",
          "key": "T05"
        },
        "pinned": {
          "table": "memory_item",
          "field": "pinned",
          "key": "sess-T05"
        },
        "summary": {
          "artifact": "results/m7_e2e.json#/summary"
        }
      }
    },
    {
      "id": "T06",
      "text": "怎么修改收货地址？",
      "scenario": "正常路径",
      "tool": "",
      "tool_verdict": "",
      "llm_status": "ok",
      "answer": "退货政策为签收后七天无理由。",
      "answer_len": 14,
      "latency_ms": 28,
      "failure": "",
      "flags": [],
      "ctx_tokens": 61,
      "ctx_tokens_uncompacted": 117,
      "saved_tokens": 56,
      "saved_pct": 0.4786,
      "pinned": 1,
      "stages": [
        {
          "name": "工单进入",
          "key": "intake",
          "status": "ok",
          "detail": "怎么修改收货地址？",
          "metrics": [
            {
              "label": "字符数",
              "value": 9,
              "source": "m7_e2e.TICKETS"
            }
          ]
        },
        {
          "name": "工具沙箱",
          "key": "tool",
          "status": "skipped",
          "detail": "无需工具：策略/知识类问题，未进入工具调用路径。",
          "metrics": [
            {
              "label": "判定",
              "value": "skipped",
              "source": "m7_e2e.tool_for_ticket"
            }
          ]
        },
        {
          "name": "记忆压缩",
          "key": "memory",
          "status": "ok",
          "detail": "滑窗 k=3；pinned 合规约束仍在上下文：合规要求：任何回复都不得包含完整订单号，只能给后四位。",
          "metrics": [
            {
              "label": "压缩后",
              "value": "61 tokens",
              "source": "m7_e2e.TicketResult.context_tokens"
            },
            {
              "label": "压缩前",
              "value": "117 tokens",
              "source": "m7_e2e.TicketResult.context_tokens_uncompacted"
            },
            {
              "label": "节省",
              "value": "56 tokens (48%)",
              "source": "压缩前 - 压缩后"
            },
            {
              "label": "pinned",
              "value": 1,
              "source": "memory_item.pinned"
            },
            {
              "label": "会话轮次",
              "value": 7,
              "source": "memory_item.session_id"
            }
          ]
        },
        {
          "name": "网关 / LLM",
          "key": "gateway",
          "status": "ok",
          "detail": "model=qwen3:8b；route=default；cache_hit=False；degraded=False",
          "metrics": [
            {
              "label": "prompt",
              "value": "117 tokens",
              "source": "gateway_request.prompt_tokens"
            },
            {
              "label": "completion",
              "value": "15 tokens",
              "source": "gateway_request.completion_tokens"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "gateway_request.latency_ms"
            },
            {
              "label": "成本",
              "value": "$0.000000",
              "source": "gateway_request.cost_usd"
            },
            {
              "label": "网关记录",
              "value": 1,
              "source": "gateway_request.idempotency_key"
            }
          ]
        },
        {
          "name": "失败分类",
          "key": "classification",
          "status": "ok",
          "detail": "未分类失败 0",
          "metrics": [
            {
              "label": "flags",
              "value": "—",
              "source": "m7_e2e.TicketResult.flags"
            }
          ]
        },
        {
          "name": "回答",
          "key": "answer",
          "status": "ok",
          "detail": "退货政策为签收后七天无理由。",
          "metrics": [
            {
              "label": "回答长度",
              "value": "14 字",
              "source": "gateway_request.content"
            }
          ]
        }
      ],
      "sources": {
        "answer": {
          "table": "gateway_request",
          "field": "content",
          "key": "T06-"
        },
        "gateway_status": {
          "table": "gateway_request",
          "field": "status",
          "key": "T06-"
        },
        "tool_verdict": {
          "table": "tool_audit",
          "field": "verdict",
          "key": "T06"
        },
        "pinned": {
          "table": "memory_item",
          "field": "pinned",
          "key": "sess-T06"
        },
        "summary": {
          "artifact": "results/m7_e2e.json#/summary"
        }
      }
    },
    {
      "id": "T07",
      "text": "为什么扣了我两次钱？",
      "scenario": "正常路径",
      "tool": "query_order",
      "tool_verdict": "ok",
      "llm_status": "ok",
      "answer": "您的订单已发货，仅向您核对后四位 0042。",
      "answer_len": 22,
      "latency_ms": 36,
      "failure": "",
      "flags": [],
      "ctx_tokens": 62,
      "ctx_tokens_uncompacted": 118,
      "saved_tokens": 56,
      "saved_pct": 0.4746,
      "pinned": 1,
      "stages": [
        {
          "name": "工单进入",
          "key": "intake",
          "status": "ok",
          "detail": "为什么扣了我两次钱？",
          "metrics": [
            {
              "label": "字符数",
              "value": 10,
              "source": "m7_e2e.TICKETS"
            }
          ]
        },
        {
          "name": "工具沙箱",
          "key": "tool",
          "status": "ok",
          "detail": "query_order → ok：whitelisted",
          "metrics": [
            {
              "label": "判定",
              "value": "ok",
              "source": "tool_audit.verdict (request_id=T07)"
            },
            {
              "label": "尝试次数",
              "value": 1,
              "source": "tool_audit rows"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "tool_audit.latency_ms"
            }
          ]
        },
        {
          "name": "记忆压缩",
          "key": "memory",
          "status": "ok",
          "detail": "滑窗 k=3；pinned 合规约束仍在上下文：合规要求：任何回复都不得包含完整订单号，只能给后四位。",
          "metrics": [
            {
              "label": "压缩后",
              "value": "62 tokens",
              "source": "m7_e2e.TicketResult.context_tokens"
            },
            {
              "label": "压缩前",
              "value": "118 tokens",
              "source": "m7_e2e.TicketResult.context_tokens_uncompacted"
            },
            {
              "label": "节省",
              "value": "56 tokens (47%)",
              "source": "压缩前 - 压缩后"
            },
            {
              "label": "pinned",
              "value": 1,
              "source": "memory_item.pinned"
            },
            {
              "label": "会话轮次",
              "value": 7,
              "source": "memory_item.session_id"
            }
          ]
        },
        {
          "name": "网关 / LLM",
          "key": "gateway",
          "status": "ok",
          "detail": "model=qwen3:8b；route=default；cache_hit=False；degraded=False",
          "metrics": [
            {
              "label": "prompt",
              "value": "138 tokens",
              "source": "gateway_request.prompt_tokens"
            },
            {
              "label": "completion",
              "value": "18 tokens",
              "source": "gateway_request.completion_tokens"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "gateway_request.latency_ms"
            },
            {
              "label": "成本",
              "value": "$0.000000",
              "source": "gateway_request.cost_usd"
            },
            {
              "label": "网关记录",
              "value": 1,
              "source": "gateway_request.idempotency_key"
            }
          ]
        },
        {
          "name": "失败分类",
          "key": "classification",
          "status": "ok",
          "detail": "未分类失败 0",
          "metrics": [
            {
              "label": "flags",
              "value": "—",
              "source": "m7_e2e.TicketResult.flags"
            }
          ]
        },
        {
          "name": "回答",
          "key": "answer",
          "status": "ok",
          "detail": "您的订单已发货，仅向您核对后四位 0042。",
          "metrics": [
            {
              "label": "回答长度",
              "value": "22 字",
              "source": "gateway_request.content"
            }
          ]
        }
      ],
      "sources": {
        "answer": {
          "table": "gateway_request",
          "field": "content",
          "key": "T07-ok"
        },
        "gateway_status": {
          "table": "gateway_request",
          "field": "status",
          "key": "T07-ok"
        },
        "tool_verdict": {
          "table": "tool_audit",
          "field": "verdict",
          "key": "T07"
        },
        "pinned": {
          "table": "memory_item",
          "field": "pinned",
          "key": "sess-T07"
        },
        "summary": {
          "artifact": "results/m7_e2e.json#/summary"
        }
      }
    },
    {
      "id": "T08",
      "text": "会员积分怎么算？",
      "scenario": "正常路径",
      "tool": "",
      "tool_verdict": "",
      "llm_status": "ok",
      "answer": "退货政策为签收后七天无理由。",
      "answer_len": 14,
      "latency_ms": 28,
      "failure": "",
      "flags": [],
      "ctx_tokens": 60,
      "ctx_tokens_uncompacted": 116,
      "saved_tokens": 56,
      "saved_pct": 0.4828,
      "pinned": 1,
      "stages": [
        {
          "name": "工单进入",
          "key": "intake",
          "status": "ok",
          "detail": "会员积分怎么算？",
          "metrics": [
            {
              "label": "字符数",
              "value": 8,
              "source": "m7_e2e.TICKETS"
            }
          ]
        },
        {
          "name": "工具沙箱",
          "key": "tool",
          "status": "skipped",
          "detail": "无需工具：策略/知识类问题，未进入工具调用路径。",
          "metrics": [
            {
              "label": "判定",
              "value": "skipped",
              "source": "m7_e2e.tool_for_ticket"
            }
          ]
        },
        {
          "name": "记忆压缩",
          "key": "memory",
          "status": "ok",
          "detail": "滑窗 k=3；pinned 合规约束仍在上下文：合规要求：任何回复都不得包含完整订单号，只能给后四位。",
          "metrics": [
            {
              "label": "压缩后",
              "value": "60 tokens",
              "source": "m7_e2e.TicketResult.context_tokens"
            },
            {
              "label": "压缩前",
              "value": "116 tokens",
              "source": "m7_e2e.TicketResult.context_tokens_uncompacted"
            },
            {
              "label": "节省",
              "value": "56 tokens (48%)",
              "source": "压缩前 - 压缩后"
            },
            {
              "label": "pinned",
              "value": 1,
              "source": "memory_item.pinned"
            },
            {
              "label": "会话轮次",
              "value": 7,
              "source": "memory_item.session_id"
            }
          ]
        },
        {
          "name": "网关 / LLM",
          "key": "gateway",
          "status": "ok",
          "detail": "model=qwen3:8b；route=default；cache_hit=False；degraded=False",
          "metrics": [
            {
              "label": "prompt",
              "value": "116 tokens",
              "source": "gateway_request.prompt_tokens"
            },
            {
              "label": "completion",
              "value": "15 tokens",
              "source": "gateway_request.completion_tokens"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "gateway_request.latency_ms"
            },
            {
              "label": "成本",
              "value": "$0.000000",
              "source": "gateway_request.cost_usd"
            },
            {
              "label": "网关记录",
              "value": 1,
              "source": "gateway_request.idempotency_key"
            }
          ]
        },
        {
          "name": "失败分类",
          "key": "classification",
          "status": "ok",
          "detail": "未分类失败 0",
          "metrics": [
            {
              "label": "flags",
              "value": "—",
              "source": "m7_e2e.TicketResult.flags"
            }
          ]
        },
        {
          "name": "回答",
          "key": "answer",
          "status": "ok",
          "detail": "退货政策为签收后七天无理由。",
          "metrics": [
            {
              "label": "回答长度",
              "value": "14 字",
              "source": "gateway_request.content"
            }
          ]
        }
      ],
      "sources": {
        "answer": {
          "table": "gateway_request",
          "field": "content",
          "key": "T08-"
        },
        "gateway_status": {
          "table": "gateway_request",
          "field": "status",
          "key": "T08-"
        },
        "tool_verdict": {
          "table": "tool_audit",
          "field": "verdict",
          "key": "T08"
        },
        "pinned": {
          "table": "memory_item",
          "field": "pinned",
          "key": "sess-T08"
        },
        "summary": {
          "artifact": "results/m7_e2e.json#/summary"
        }
      }
    },
    {
      "id": "T09",
      "text": "我要查一个不存在的订单号 00000000。",
      "scenario": "订单号 00000000 查不到；工具如实返回空，模型回答查无此单。",
      "tool": "query_order",
      "tool_verdict": "ok",
      "llm_status": "ok",
      "answer": "未查到该订单，请核对后四位。",
      "answer_len": 14,
      "latency_ms": 30,
      "failure": "",
      "flags": [],
      "ctx_tokens": 66,
      "ctx_tokens_uncompacted": 122,
      "saved_tokens": 56,
      "saved_pct": 0.459,
      "pinned": 1,
      "stages": [
        {
          "name": "工单进入",
          "key": "intake",
          "status": "ok",
          "detail": "我要查一个不存在的订单号 00000000。",
          "metrics": [
            {
              "label": "字符数",
              "value": 22,
              "source": "m7_e2e.TICKETS"
            }
          ]
        },
        {
          "name": "工具沙箱",
          "key": "tool",
          "status": "ok",
          "detail": "query_order → ok：whitelisted",
          "metrics": [
            {
              "label": "判定",
              "value": "ok",
              "source": "tool_audit.verdict (request_id=T09)"
            },
            {
              "label": "尝试次数",
              "value": 1,
              "source": "tool_audit rows"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "tool_audit.latency_ms"
            }
          ]
        },
        {
          "name": "记忆压缩",
          "key": "memory",
          "status": "ok",
          "detail": "滑窗 k=3；pinned 合规约束仍在上下文：合规要求：任何回复都不得包含完整订单号，只能给后四位。",
          "metrics": [
            {
              "label": "压缩后",
              "value": "66 tokens",
              "source": "m7_e2e.TicketResult.context_tokens"
            },
            {
              "label": "压缩前",
              "value": "122 tokens",
              "source": "m7_e2e.TicketResult.context_tokens_uncompacted"
            },
            {
              "label": "节省",
              "value": "56 tokens (46%)",
              "source": "压缩前 - 压缩后"
            },
            {
              "label": "pinned",
              "value": 1,
              "source": "memory_item.pinned"
            },
            {
              "label": "会话轮次",
              "value": 7,
              "source": "memory_item.session_id"
            }
          ]
        },
        {
          "name": "网关 / LLM",
          "key": "gateway",
          "status": "ok",
          "detail": "model=qwen3:8b；route=default；cache_hit=False；degraded=False",
          "metrics": [
            {
              "label": "prompt",
              "value": "142 tokens",
              "source": "gateway_request.prompt_tokens"
            },
            {
              "label": "completion",
              "value": "14 tokens",
              "source": "gateway_request.completion_tokens"
            },
            {
              "label": "耗时",
              "value": "0 ms",
              "source": "gateway_request.latency_ms"
            },
            {
              "label": "成本",
              "value": "$0.000000",
              "source": "gateway_request.cost_usd"
            },
            {
              "label": "网关记录",
              "value": 1,
              "source": "gateway_request.idempotency_key"
            }
          ]
        },
        {
          "name": "失败分类",
          "key": "classification",
          "status": "ok",
          "detail": "未分类失败 0",
          "metrics": [
            {
              "label": "flags",
              "value": "—",
              "source": "m7_e2e.TicketResult.flags"
            }
          ]
        },
        {
          "name": "回答",
          "key": "answer",
          "status": "ok",
          "detail": "未查到该订单，请核对后四位。",
          "metrics": [
            {
              "label": "回答长度",
              "value": "14 字",
              "source": "gateway_request.content"
            }
          ]
        }
      ],
      "sources": {
        "answer": {
          "table": "gateway_request",
          "field": "content",
          "key": "T09-ok"
        },
        "gateway_status": {
          "table": "gateway_request",
          "field": "status",
          "key": "T09-ok"
        },
        "tool_verdict": {
          "table": "tool_audit",
          "field": "verdict",
          "key": "T09"
        },
        "pinned": {
          "table": "memory_item",
          "field": "pinned",
          "key": "sess-T09"
        },
        "summary": {
          "artifact": "results/m7_e2e.json#/summary"
        }
      }
    }
  ],
  "boundaries": [
    "工单只有 10 条、且是手写的，不是真实客服流量分布；端到端 p95 是描述统计，不构成 SLO。",
    "离线模式的失败率（timeout 5% + 空回答 5%）是**脚本注入**的，不是 ollama 的真实可用率。",
    "工具只做 order_id 精确查询，没有模糊匹配/多轮澄清；真实客服场景里「查不到」的比例会高得多。",
    "没有用户侧反馈回路（工单是否解决），所以「成功」定义是「模型返回了非空内容」，不等于用户满意。",
    "M6 的订单库是合成 2000 行，不是真实交易数据。"
  ],
  "verification": {
    "status": "MATCH",
    "checks": 36,
    "mismatches": [],
    "source": "results/m7_e2e.json"
  }
}
