feat(v2): T7 网关扩展 + T9 token 计量与账单（/chat v2,/chat/legacy,/runs,/review,/metrics）

router_system/v2stats.py

@@ -0,0 +1,80 @@
+"""v2 统计聚合器（V2Stats）—— token 计量与账单（纯标准库）。
+
+T9：每请求 API token 记账；聚合快路径命中率、回合数分布、熔断次数、累计 token/成本。
+配合 /metrics 对外透出（论文 E1 token 经济学数据来源之一）。
+"""
+from __future__ import annotations
+
+import threading
+from typing import Any, Dict, List, Optional
+
+
+class V2Stats:
+    """线程安全的 v2 运行统计。"""
+
+    def __init__(self):
+        self._lock = threading.Lock()
+        self._total = 0
+        self._fast_path = 0
+        self._breach = 0
+        self._by_status: Dict[str, int] = {}
+        self._rounds: List[int] = []
+        self._api_input_tokens = 0
+        self._api_output_tokens = 0
+        self._api_cost_usd = 0.0
+        self._recent: List[Dict[str, Any]] = []
+
+    def record(self, result) -> None:
+        """记录一次 PipelineResult。"""
+        with self._lock:
+            self._total += 1
+            if getattr(result, "fast_path", False):
+                self._fast_path += 1
+            status = getattr(result, "status", "?")
+            self._by_status[status] = self._by_status.get(status, 0) + 1
+            self._rounds.append(getattr(result, "rounds_used", 0))
+            if "breach" in " ".join(getattr(result, "route", [])):
+                self._breach += 1
+            self._api_input_tokens += getattr(result, "api_input_tokens", 0)
+            self._api_output_tokens += getattr(result, "api_output_tokens", 0)
+            self._api_cost_usd += getattr(result, "cost_est", 0.0)
+            self._recent.append({
+                "request_id": getattr(result, "request_id", ""),
+                "status": status,
+                "fast_path": getattr(result, "fast_path", False),
+                "api_input_tokens": getattr(result, "api_input_tokens", 0),
+                "api_output_tokens": getattr(result, "api_output_tokens", 0),
+                "rounds_used": getattr(result, "rounds_used", 0),
+            })
+            if len(self._recent) > 200:
+                self._recent = self._recent[-200:]
+
+    def summary(self) -> Dict[str, Any]:
+        with self._lock:
+            total = self._total
+            rounds = self._rounds
+            return {
+                "total_requests": total,
+                "fast_path_rate": round(self._fast_path / total, 4) if total else 0.0,
+                "status_distribution": dict(self._by_status),
+                "breach_count": self._breach,
+                "rounds_used": {
+                    "avg": round(sum(rounds) / len(rounds), 2) if rounds else 0.0,
+                    "max": max(rounds) if rounds else 0,
+                    "distribution": _histogram(rounds),
+                },
+                "api_tokens": {
+                    "input": self._api_input_tokens,
+                    "output": self._api_output_tokens,
+                    "total": self._api_input_tokens + self._api_output_tokens,
+                    "cost_est_usd": round(self._api_cost_usd, 6),
+                },
+            }
+
+
+def _histogram(values: List[int], max_bucket: int = 10) -> Dict[str, int]:
+    out: Dict[str, int] = {}
+    for v in values:
+        key = str(v) if v <= max_bucket else f">{max_bucket}"
+        out[key] = out.get(key, 0) + 1
+    return out

router_system/worker.py

@@ -154,9 +154,14 @@ class WorkerLoop:
 
 def build_worker(cfg: Dict[str, Any], kb: Any = None,
                  generate: Optional[Callable[[str], Awaitable[str]]] = None) -> WorkerLoop:
-    """cfg 为 config.worker 段。generate 缺省时用 llama-server 端点客户端（惰性）。"""
+    """cfg 为 config.worker 段。generate 缺省时按 backend 选择：
+    mock（零运行时演示）| llama_server（真实本地模型，惰性连接）。"""
+    backend = cfg.get("backend", "llama_server")
     if generate is None:
-        generate = _make_llama_generate(cfg)
+        if backend == "mock":
+            generate = _mock_generate()
+        else:
+            generate = _make_llama_generate(cfg)
     verifier = Verifier(code_timeout_s=float(cfg.get("code_timeout_s", 10)))
     return WorkerLoop(
         generate=generate,
@@ -167,6 +172,16 @@ def build_worker(cfg: Dict[str, Any], kb: Any = None,
     )
 
 
+def _mock_generate() -> Callable[[str], Awaitable[str]]:
+    """零运行时 mock 生成器：返回一段确定性文本（演示/测试，不连真实模型）。"""
+
+    async def _gen(prompt: str) -> str:
+        return ("（mock worker）以下是对当前步骤的实现说明："
+                "步骤已完成，内容足够长且非占位，可供接地验证通过。")
+
+    return _gen
+
+
 def _make_llama_generate(cfg: Dict[str, Any]) -> Callable[[str], Awaitable[str]]:
     """返回调用本地 llama-server（OpenAI 兼容 /v1/chat/completions）的生成器。"""
     base_url = cfg.get("base_url", f"http://127.0.0.1:{cfg.get('port', 8901)}/v1")