feat: add health-aware P2C selector as default

Score is success-rate EWMA over (1 + inflight) with a 0.05 exploration
floor so quarantined sources can prove recovery; EWMA climb doubles as
slow-start. New optional OutcomeAwareSelector port feeds attempt
outcomes.

src/polygateway/client.py

@@ -24,7 +24,12 @@ from polygateway.middleware.structured import StructuredMW
 from polygateway.middleware.telemetry import TelemetryEmitter, TelemetryMW
 from polygateway.pricing import PricingTable
 from polygateway.providers import get_provider
-from polygateway.sources import LeastInflightSelector, RoundRobinSelector, SourceCooldownMemo
+from polygateway.sources import (
+    HealthAwareSelector,
+    LeastInflightSelector,
+    RoundRobinSelector,
+    SourceCooldownMemo,
+)
 from polygateway.transports.openai_compat import OpenAICompatTransport
 from polygateway.types import ChatRequest, LLMResponse
 
@@ -193,6 +198,7 @@ class GatewayClient:
         cache: CacheBackend | None = None,
         telemetry: TelemetryRecorder | None = None,
         registry: Mapping[str, ProviderProfile] | None = None,
+        rng: Any = random.random,
     ) -> GatewayClient:
         """按配置装配;显式传入的后端实例即共享(None 项按配置自建私有实例)。"""
         sources = list(settings.sources)
@@ -201,7 +207,7 @@ class GatewayClient:
         return cls(
             scope=settings.scope,
             sources=sources,
-            selector=_build_selector(settings.selector),
+            selector=_build_selector(settings.selector, rng=rng),
             limiter=limiter or _build_limiter(settings, sources),
             breaker=breaker or _build_breaker(settings),
             transport=OpenAICompatTransport(registry=registry),
@@ -272,8 +278,12 @@ def _build_breaker(settings: GatewaySettings) -> ProviderGate:
     return InMemoryGate(config=settings.breaker)
 
 
-def _build_selector(name: str) -> SourceSelector:
-    return RoundRobinSelector() if name == "round_robin" else LeastInflightSelector()
+def _build_selector(name: str, *, rng: Any = random.random) -> SourceSelector:
+    if name == "round_robin":
+        return RoundRobinSelector()
+    if name == "least_inflight":
+        return LeastInflightSelector()
+    return HealthAwareSelector(rng=rng)
 
 
 def _build_cache(settings: GatewaySettings) -> CacheBackend | None:

src/polygateway/ports.py

@@ -181,6 +181,18 @@ class SourceSelector(Protocol):
     ) -> list[SourceConfig]: ...
 
 
+@runtime_checkable
+class OutcomeAwareSelector(Protocol):
+    """可选选源器扩展(M2.5 设计 §3.2): 消费尝试结果以维护健康视图。
+
+    RetryMW 构造时 isinstance 判定一次;非本 Protocol 的选源器不受影响。
+    喂数口径: 真实成功 ok=True;Transient/SourceDead/429 ok=False;
+    ResultInvalid 与"网关健康拒坏请求"不喂(坏结果 ≠ 坏服务)。
+    """
+
+    def record_outcome(self, source_name: str, ok: bool) -> None: ...
+
+
 @runtime_checkable
 class StructuredOutputStrategy(Protocol):
     """结构化输出策略(D7/D14): 请求侧叠加 + 响应侧解析。"""

src/polygateway/sources.py

@@ -7,6 +7,7 @@ RetryMW 的进程本地状态——熔断开路的源在本地记冷却截止,
 
 from __future__ import annotations
 
+import random
 import time
 from typing import TYPE_CHECKING
 
@@ -15,6 +16,9 @@ if TYPE_CHECKING:
 
     from polygateway.types import SourceConfig, SourceStats
 
+_EWMA_ALPHA = 0.2  # 健康 EWMA 步长: 约 10 次成功从谷底爬回 0.9(天然 slow-start)
+_SCORE_FLOOR = 0.05  # 探索地板: 塌陷源保有微量被选概率,恢复靠真实成功自证
+
 
 class RoundRobinSelector:
     """轮转起点后移(CHS selector.py:20 同款);单 client 内游标推进。"""
@@ -41,6 +45,43 @@ class LeastInflightSelector:
         return sorted(sources, key=lambda s: stats[s.name].inflight if s.name in stats else 0)
 
 
+class HealthAwareSelector:
+    """健康感知选源(M2.5 设计 §3.2;蓝本 Envoy least-request + gRPC WRR)。
+
+    score = max(ewma_success, 地板) / (1 + inflight)。头名经 P2C(随机取
+    两源比分,高者先)引入探索;其余按分数降序。健康态为进程本地(业界
+    共识: Envoy/Finagle/gRPC 全本地),属 client 实例,不违反纯 asyncio 中立。
+    已知取舍(设计 §3.2): 仅头名随机化,多 worker 溢出会集中到同一次优源。
+    """
+
+    def __init__(self, *, rng: Callable[[], float] = random.random) -> None:
+        self._rng = rng
+        self._ewma: dict[str, float] = {}
+
+    def record_outcome(self, source_name: str, ok: bool) -> None:
+        """尝试结果喂数(OutcomeAwareSelector 端口);初始 1.0 乐观起步。"""
+        prev = self._ewma.get(source_name, 1.0)
+        self._ewma[source_name] = prev + _EWMA_ALPHA * ((1.0 if ok else 0.0) - prev)
+
+    def _score(self, name: str, stats: dict[str, SourceStats]) -> float:
+        ewma = max(self._ewma.get(name, 1.0), _SCORE_FLOOR)
+        inflight = stats[name].inflight if name in stats else 0
+        return ewma / (1.0 + inflight)
+
+    def order(
+        self, sources: list[SourceConfig], stats: dict[str, SourceStats]
+    ) -> list[SourceConfig]:
+        if len(sources) < 2:
+            return list(sources)
+        ranked = sorted(sources, key=lambda s: self._score(s.name, stats), reverse=True)
+        # P2C: 随机取两源比分,胜者提为头名(平分取采样序首位)
+        i = int(self._rng() * len(sources)) % len(sources)
+        j = int(self._rng() * len(sources)) % len(sources)
+        a, b = sources[i], sources[j]
+        head = a if self._score(a.name, stats) >= self._score(b.name, stats) else b
+        return [head] + [s for s in ranked if s.name != head.name]
+
+
 class SourceCooldownMemo:
     """进程本地的源冷却备忘(CHS governance.py:107 同款)。