chore(P3): baseline 落库下沉 + MatchPredictPanel 拆分 + 多 worker/CSRF 文档

P3-2 baseline 落库从路由下沉到服务层(predict_baseline 内直接落库),
删除路由层 _persist_baseline,三种模式统一 result.prediction_id,对外 JSON 不变。

P3-1 MatchPredictPanel.PredictionPanel 拆为 OutcomePanel/AgentsPanel/ReasoningPanel
三个子组件,本文件保留 PredictModal/PredictProgress/Spinner,对外导出路径不变。

P3-3 docs 加 ⚠️ 多 worker 陷阱红字 + STRICT_SINGLE_WORKER 环境变量(启动期强制拒绝多 worker)。
P3-4 docs 新增「同站部署 vs 跨站 CSRF」节。

docs/06-deployment.md

@@ -46,6 +46,9 @@ curl http://localhost:8000/health
 - [ ] **6. 反代信任头** — `TRUST_PROXY_HEADERS=True`,且**仅可信反代可达 API**;反代需设置 `X-Forwarded-For`(`$proxy_add_x_forwarded_for`)与 `X-Real-IP`,否则限流/日志按反代 IP 计数
 - [ ] **7. 限流前置到网关** — 推荐 Nginx `limit_req`(配置见[安全与限流](#安全与限流));应用内限流与 KeyRing 为**单进程内存实现**,多 worker 各自独立计数会把实际配额放大 N 倍(启动时会打印一次性告警)
 - [ ] **8. uvicorn 单 worker** — compose/Dockerfile 默认单 worker,保持即可;需横向扩容时先在网关统一限流,再起多实例(每实例仍单 worker)
+
+> ⚠️ **多 worker 陷阱**:应用内限流(`_RateLimiter`)与 KeyRing 均为**进程内纯内存状态**,多 worker 部署(如 `uvicorn --workers 4`)时各进程**各自独立计数、互不共享**——实际限流配额会被放大 N 倍、KeyRing 限流状态也不同步。
+> 若确需多 worker,必须前置 Nginx/网关做**全局限流**(见[安全与限流](#安全与限流)),并设环境变量 `STRICT_SINGLE_WORKER=True`(见下)在启动期强制拒绝多 worker,避免静默配额漂移。
 - [ ] **9. 启动后健康检查** — `curl /health` 返回 200(存活);`curl /health/ready` 返回 200(就绪,校验数据库连通,不可达时 503)
 - [ ] **10. 数据库迁移** — compose/Dockerfile 启动命令已内置 `alembic upgrade head && uvicorn …`,升级镜像重启即自动迁移,无需手动执行
 
@@ -99,10 +102,29 @@ cd frontend && npm install && npm run dev
 | `LLM_AGGREGATOR_MODEL` | ❌ | — | 终裁模型(回落 `LLM_MODEL`) |
 | `BZZOIRO_KEY` | ✅ | — | bzzoiro 数据源 Key(唯一数据源) |
 | `CORS_ORIGINS` | ❌ | `http://localhost:5173,...` | 允许的跨域来源 |
+| `STRICT_SINGLE_WORKER` | ❌ | `False` | `True` 时若以多 worker 启动则拒绝(防限流配额漂移) |
 | `SECRET_KEY` | ❌ | — | 加密主密钥(生产环境必填) |
 | `ADMIN_PASSWORD` | ❌ | — | 管理后台密码(留空=不启用) |
 | `ADMIN_API_KEY` | ❌ | — | 机器/脚本调用的 API Key |
 
+## 同站部署 vs 跨站 CSRF
+
+ Profeto 管理鉴权使用 **HttpOnly Cookie 会话**(登录后服务端写入),`allow_credentials=True` 的 CORS 配置允许浏览器跨域携带 Cookie——这也引入了 CSRF 面。部署拓扑决定风险等级:
+
+**同站部署(推荐)**: 前端与 API 同域(反代把 `/` 与 `/api` 都转发到同一后端,或同源端口)。
+  - 浏览器视为 **same-origin**,CORS 不触发;`SameSite=Lax` 会话 Cookie 天然阻断跨站请求携带。
+  - 风险最低。`CORS_ORIGINS` 可设为空或同域来源,仅作兜底。
+
+**跨站部署**: 前端与 API 不同域(如前端 `app.example.com`、API `api.example.com`,或开发时 `localhost:3000` → `localhost:8000`)。
+  - 必须把 API 域名列入 `CORS_ORIGINS`,且 `allow_credentials=True` 才能携带 Cookie。
+  - 此时任何被允许域下的页面都能构造带 Cookie 的请求 → **CSRF 面**:
+    - 状态变更接口(采集/回测/改密等写操作)要求**管理员 Cookie + 同域**,攻击者无法从第三方站点读取 Cookie,但可构造跨域表单/请求——`SameSite=Lax` 会阻断跨站 POST 表单提交(顶级导航 GET 仍放行),这是当前主要防线。
+    - `GET /api/v1/admin/*` 只读接口受 `SameSite=Lax` 下顶级导航可能被利用,但攻击者无法读取响应(CORS 不匹配时浏览器拦截)。
+  - **加固建议**:
+    1. 反代层加 `Origin`/`Referer` 校验,仅放行 `CORS_ORIGINS` 列表中的来源(即便 FastAPI CORS 已通过,反代校验是多一层纵深)。
+    2. 写操作要求自定义请求头(如 `X-Requested-With: XMLHttpRequest`),第三方站点无法在无预检下添加自定义头,天然阻断简单跨站 POST。
+    3. 生产强制 HTTPS(`APP_ENV=production` 下 Cookie 自动 `Secure`),防中间人窃 Cookie。
+
 ## LLM 提供商配置示例
 
 ### OpenAI

frontend/src/pages/matches/components/AgentsPanel.tsx

@@ -0,0 +1,171 @@
+/**
+ * AgentsPanel: 五路专家意见 —— 可折叠 + 状态摘要 + 权重条形图 + 单路详情。
+ *
+ * P3-1: 从 MatchPredictPanel.PredictionPanel 拆出,渲染逻辑原样搬迁。
+ */
+import { useState } from 'react'
+import type { AgentReport, Prediction } from '../../types'
+import { AGENT_LABELS, CN_NUM } from '../../types'
+
+const STATUS_BADGE: Record<string, { label: string; cls: string }> = {
+  ok: { label: '正常', cls: 'text-ink-500' },
+  no_data: { label: '无数据', cls: 'text-ink-400' },
+  error: { label: '调用失败', cls: 'text-press' },
+  parse_error: { label: '解析失败', cls: 'text-press' },
+}
+
+const SUFFICIENCY_LABEL: Record<string, string> = {
+  high: '充分',
+  medium: '一般',
+  low: '偏少',
+  none: '无',
+}
+
+/** home_edge(-1~1,正=利主队)的可视化:以中线为原点的双向细条 */
+function EdgeBar({ value }: { value: number }) {
+  const v = Math.max(-1, Math.min(1, value))
+  const half = Math.abs(v) * 50
+  return (
+    <div className="relative h-px w-full bg-ink-200" role="presentation">
+      <span className="absolute left-1/2 top-1/2 h-2 w-px -translate-x-1/2 -translate-y-1/2 bg-ink-400" />
+      <span
+        className={`absolute top-0 h-px transition-all duration-500 ${v >= 0 ? 'bg-press' : 'bg-ink-600'}`}
+        style={
+          v >= 0
+            ? { left: '50%', width: `${half}%` }
+            : { right: '50%', width: `${half}%` }
+        }
+      />
+    </div>
+  )
+}
+
+/** 单路专家意见:汉字编号 + 细线行 */
+function AgentCard({ report: r, no }: { report: AgentReport; no: string }) {
+  const badge = STATUS_BADGE[r.status] ?? { label: r.status, cls: 'text-ink-400' }
+  const inactive = r.status !== 'ok'
+
+  return (
+    <details className="group border-b border-ink-200">
+      <summary className="flex cursor-pointer list-none items-baseline gap-2.5 px-1 py-3">
+        <span className="font-serif text-sm text-ink-400">{no}</span>
+        <span className="text-sm font-medium text-ink-900">{AGENT_LABELS[r.agent] ?? r.agent}</span>
+        <span className={`text-2xs ${badge.cls}`}>{badge.label}</span>
+
+        <span className="ml-auto flex items-baseline gap-3 text-2xs tabular-nums text-ink-500">
+          {r.status === 'ok' && r.subjective_confidence !== null && (
+            <span>信心 {Math.round(r.subjective_confidence * 100)}%</span>
+          )}
+          {r.status === 'ok' && r.probable_score && (
+            <span className="font-serif font-bold text-ink-800">{r.probable_score}</span>
+          )}
+          <svg viewBox="0 0 20 20" className="h-3 w-3 self-center text-ink-300 transition-transform group-open:rotate-90" fill="currentColor" aria-hidden="true">
+            <path d="M7.3 5.3a1 1 0 011.4 0l4 4a1 1 0 010 1.4l-4 4a1 1 0 01-1.4-1.4L10.6 10 7.3 6.7a1 1 0 010-1.4z" />
+          </svg>
+        </span>
+      </summary>
+
+      <div className="space-y-3 px-1 pb-4 pl-7">
+        {inactive && (
+          <p className="text-xs leading-relaxed text-ink-500">
+            {r.status === 'no_data' && '该维度没有可用数据,已跳过 LLM 分析以节省额度(不影响其他专家)。'}
+            {r.status === 'error' && '该专家调用失败,本次结论未纳入其视角(fail-open 设计,不阻断整体预测)。'}
+            {r.status === 'parse_error' && '模型输出未通过格式校验,该报告已丢弃。'}
+          </p>
+        )}
+
+        {!inactive && r.home_edge !== null && (
+          <div>
+            <div className="mb-1.5 flex items-baseline justify-between text-2xs">
+              <span className="text-ink-500">主队优势</span>
+              <span className={`font-semibold tabular-nums ${r.home_edge > 0 ? 'text-press' : r.home_edge < 0 ? 'text-ink-700' : 'text-ink-500'}`}>
+                {r.home_edge > 0 ? '+' : ''}{r.home_edge.toFixed(2)}
+              </span>
+            </div>
+            <EdgeBar value={r.home_edge} />
+            <div className="mt-1 flex justify-between text-2xs text-ink-400">
+              <span>利客队</span>
+              <span>利主队</span>
+            </div>
+          </div>
+        )}
+
+        {r.analysis && (
+          <p className="font-serif text-sm leading-loose text-ink-700">{r.analysis}</p>
+        )}
+
+        {r.key_evidence.length > 0 && (
+          <ul className="space-y-1.5">
+            {r.key_evidence.map((e, i) => (
+              <li key={i} className="flex gap-2 text-xs leading-relaxed text-ink-600">
+                <span className="flex-shrink-0 text-ink-300" aria-hidden="true">—</span>
+                <span>{e}</span>
+              </li>
+            ))}
+          </ul>
+        )}
+
+        {r.exp_home_goals !== null && r.exp_away_goals !== null && (
+          <p className="text-xs text-ink-500">
+            进球期望 <span className="font-serif font-bold tabular-nums text-ink-900">{r.exp_home_goals.toFixed(1)} - {r.exp_away_goals.toFixed(1)}</span>
+          </p>
+        )}
+
+        {!inactive && (
+          <p className="border-t border-ink-100 pt-2.5 text-2xs text-ink-400">
+            数据充分度 {SUFFICIENCY_LABEL[r.data_sufficiency] ?? r.data_sufficiency}
+            <span className="mx-2 text-ink-200">|</span>
+            <span className="font-mono">{r.model}</span>
+            {r.latency_ms !== null && <span className="ml-2 tabular-nums">{r.latency_ms}ms</span>}
+          </p>
+        )}
+      </div>
+    </details>
+  )
+}
+
+export function AgentsPanel({ prediction }: { prediction: Prediction }) {
+  const [expertsOpen, setExpertsOpen] = useState(false)
+  const degraded = prediction.status === 'degraded' || prediction.status === 'failed'
+  const reports = prediction.agent_outputs ?? []
+  const okReports = reports.filter(r => r.status === 'ok')
+
+  if (reports.length === 0) return null
+
+  return (
+    <section>
+      <button
+        onClick={() => setExpertsOpen(o => !o)}
+        className="flex w-full items-center justify-between border-b border-ink-200 pb-2 text-left"
+      >
+        <span className="section-head mb-0">五路专家意见({okReports.length}/{reports.length} 路有效)</span>
+        <span className="text-2xs text-ink-400">{expertsOpen ? '收起' : '展开'}</span>
+      </button>
+
+      {!degraded && prediction.agent_weights && Object.keys(prediction.agent_weights).length > 0 && (
+        <div className="mt-3 space-y-1.5">
+          <span className="text-2xs text-ink-500">终裁权重分布</span>
+          {Object.entries(prediction.agent_weights)
+            .sort((a, b) => b[1] - a[1])
+            .map(([k, v]) => (
+              <div key={k} className="grid grid-cols-[96px_minmax(0,1fr)_40px] items-center gap-2">
+                <span className="truncate text-2xs text-ink-500">{AGENT_LABELS[k] ?? k}</span>
+                <div className="h-1.5 bg-paper-100">
+                  <div className="h-full bg-press" style={{ width: `${Math.round(v * 100)}%` }} />
+                </div>
+                <span className="text-right text-2xs tabular-nums text-ink-500">{Math.round(v * 100)}%</span>
+              </div>
+            ))}
+        </div>
+      )}
+
+      {expertsOpen && (
+        <div className="mt-2">
+          {reports.map((r, i) => (
+            <AgentCard key={r.agent} report={r} no={CN_NUM[i] ?? String(i + 1)} />
+          ))}
+        </div>
+      )}
+    </section>
+  )
+}

frontend/src/pages/matches/components/MatchPredictPanel.tsx

@@ -6,37 +6,11 @@
  */
 import { useEffect, useState } from 'react'
 import TeamSideTag from '../../../components/TeamSideTag'
-import type { AgentReport, Match, Prediction } from '../types'
+import type { Match, Prediction } from '../types'
-import { AGENT_LABELS, CN_NUM, OUTCOME_LABEL } from '../types'
+import { AGENT_LABELS } from '../types'
-
+import { AgentsPanel } from './AgentsPanel'
-/** 置信度细线:0~1 数值的低调可视化 */
+import { OutcomePanel } from './OutcomePanel'
-function Meter({ value }: { value: number }) {
+import { ReasoningPanel } from './ReasoningPanel'
-  const pct = Math.max(0, Math.min(100, Math.round(value * 100)))
-  return (
-    <div className="h-px w-full bg-ink-200" role="presentation">
-      <div className="h-px bg-press transition-[width] duration-500" style={{ width: `${pct}%` }} />
-    </div>
-  )
-}
-
-/** home_edge(-1~1,正=利主队)的可视化:以中线为原点的双向细条 */
-function EdgeBar({ value }: { value: number }) {
-  const v = Math.max(-1, Math.min(1, value))
-  const half = Math.abs(v) * 50
-  return (
-    <div className="relative h-px w-full bg-ink-200" role="presentation">
-      <span className="absolute left-1/2 top-1/2 h-2 w-px -translate-x-1/2 -translate-y-1/2 bg-ink-400" />
-      <span
-        className={`absolute top-0 h-px transition-all duration-500 ${v >= 0 ? 'bg-press' : 'bg-ink-600'}`}
-        style={
-          v >= 0
-            ? { left: '50%', width: `${half}%` }
-            : { right: '50%', width: `${half}%` }
-        }
-      />
-    </div>
-  )
-}
 
 function Spinner({ className = '' }: { className?: string }) {
   return (
@@ -53,77 +27,73 @@ function Spinner({ className = '' }: { className?: string }) {
 }
 
 /** 胜平负一行文字:选中的红字加方块标记,未选中的退灰 */
-function OutcomeLine({
+.**
-  pick,
+ * P3-1:PredictionPanel 不再自绘,改为组合三个子组件:
-  confidence,
+ *   OutcomePanel(比分/胜平负/成本) / AgentsPanel(专家意见) / ReasoningPanel(终裁/降级)。
+ * 渲染输出与拆分前完全一致(仅降级警示 + 报头 + 元信息仍在此处)。
+ */
+function PredictionPanel({
+  prediction,
+  match,
+  embedded = false,
 }: {
-  pick: string | null
+  prediction: Prediction
-  confidence: number | null
+  match: Match
+  /** 弹窗嵌入模式:弹窗已提供报头,这里省略自带版头 */
+  embedded?: boolean
 }) {
-  const options = ['1', 'X', '2'] as const
+  const homeName = match.home_team_zh || match.home_team
+  const awayName = match.away_team_zh || match.away_team
+  const degraded = prediction.status === 'degraded' || prediction.status === 'failed'
+  const reports = prediction.agent_outputs ?? []
+  const okReports = reports.filter(r => r.status === 'ok')
+
   return (
-    <div>
+    <article className={embedded ? 'bg-paper-50' : 'border border-ink-900 bg-paper-50'}>
-      <div className="flex items-baseline justify-center gap-6 sm:gap-10">
+      {!embedded && (
-        {options.map(o => {
+      <div className="flex flex-wrap items-baseline justify-between gap-2 border-b border-ink-900 bg-paper-100 px-4 py-2.5 sm:px-5">
-          const on = pick === o
+        <h3 className="flex flex-wrap items-center gap-1.5 font-serif text-sm font-bold text-ink-900">
-          return (
+          预测版 ·
-            <div key={o} className="flex flex-col items-center gap-1">
+          <TeamSideTag side="home" />
-              <span className={`flex items-center gap-1.5 text-sm ${on ? 'font-semibold text-press' : 'text-ink-400'}`}>
+          {homeName}
-                {on && <span className="inline-block h-2 w-2 bg-press" aria-hidden="true" />}
+          <span>对</span>
-                {OUTCOME_LABEL[o]}
+          <TeamSideTag side="away" />
-              </span>
+          {awayName}
-              {on && confidence !== null && (
+        </h3>
         <span className="text-2xs tabular-nums text-ink-500">
-                  置信 {Math.round(confidence * 100)}%
+          {prediction.provider} / {prediction.model}
+          {prediction.latency_ms !== null && ` · ${(prediction.latency_ms / 1000).toFixed(1)}s`}
         </span>
-              )}
-            </div>
-          )
-        })}
-      </div>
-      {pick && confidence !== null && (
-        <div className="mx-auto mt-3 max-w-xs">
-          <Meter value={confidence} />
-          <p className="mt-1 text-center text-2xs text-ink-400">主观置信度,非统计概率</p>
       </div>
       )}
+
+      <div className="space-y-7 px-4 py-6 sm:px-5">
+        {degraded && (
+          <div className="border-l-2 border-press bg-press-wash/40 px-4 py-3">
+            <p className="font-serif text-sm font-bold text-press-dark">
+              {prediction.status === 'failed' ? '预测失败' : '预测降级(degraded)'}
+            </p>
+            <p className="mt-1.5 whitespace-pre-wrap text-xs leading-relaxed text-ink-600">
+              {prediction.reasoning || '所有专家均无有效数据或调用失败,无法生成可靠比分。'}
+            </p>
           </div>
+        )}
+
+        {!degraded && <OutcomePanel prediction={prediction} match={match} />}
+
+        <p className="text-center text-2xs text-ink-500">
+          `多专家模式 · ${okReports.length}/${reports.length} 路有效`
+          {prediction.prompt_version && ` · prompt ${prediction.prompt_version}`}
+        </p>
+
+        <AgentsPanel prediction={prediction} />
+
+        <ReasoningPanel prediction={prediction} />
+      </div>
+    </article>
   )
 }
 
-/** 预测成本展示:耗时 + token + 限流余量 */
-function PredictionCost({ prediction }: { prediction: Prediction }) {
-  const latency = prediction.latency_ms != null ? `${(prediction.latency_ms / 1000).toFixed(1)}s` : null
-  const tokens = prediction.prompt_tokens != null || prediction.completion_tokens != null
-    ? `${prediction.prompt_tokens ?? '?'}/${prediction.completion_tokens ?? '?'}`
-    : null
-
-  if (!latency && !tokens && prediction.rate_limit_remaining == null) return null
-
-  return (
-    <div className="border-t border-ink-200 pt-3 text-2xs text-ink-500">
-      <div className="flex flex-wrap items-center justify-center gap-x-4 gap-y-1">
-        {latency && (
-          <span className="inline-flex items-center gap-1">
-            <span aria-hidden="true" className="opacity-60">⏱</span>耗时 {latency}
-          </span>
-        )}
-        {tokens && (
-          <span className="inline-flex items-center gap-1">
-            <span aria-hidden="true" className="opacity-60">Tok</span>prompt/completion: {tokens}
-          </span>
-        )}
-        {prediction.rate_limit_remaining != null && prediction.rate_limit_remaining <= 3 && (
-          <span className="text-press" title="每分钟最多 10 次预测">
-            剩余配额: {prediction.rate_limit_remaining}/10(分钟)
-          </span>
-        )}
-      </div>
-    </div>
-  )
-}
-
-/** 预测过程阶段(按时长模拟;结果到达即跳到完成) */
 function PredictProgress() {
   const [elapsed, setElapsed] = useState(0)
   useEffect(() => {
@@ -199,256 +169,6 @@ function PredictProgress() {
     </div>
   )
 }
-
-const STATUS_BADGE: Record<string, { label: string; cls: string }> = {
-  ok: { label: '正常', cls: 'text-ink-500' },
-  no_data: { label: '无数据', cls: 'text-ink-400' },
-  error: { label: '调用失败', cls: 'text-press' },
-  parse_error: { label: '解析失败', cls: 'text-press' },
-}
-
-const SUFFICIENCY_LABEL: Record<string, string> = {
-  high: '充分',
-  medium: '一般',
-  low: '偏少',
-  none: '无',
-}
-
-/** 单路专家意见:汉字编号 + 细线行 */
-function AgentCard({ report: r, no }: { report: AgentReport; no: string }) {
-  const badge = STATUS_BADGE[r.status] ?? { label: r.status, cls: 'text-ink-400' }
-  const inactive = r.status !== 'ok'
-
-  return (
-    <details className="group border-b border-ink-200">
-      <summary className="flex cursor-pointer list-none items-baseline gap-2.5 px-1 py-3">
-        <span className="font-serif text-sm text-ink-400">{no}</span>
-        <span className="text-sm font-medium text-ink-900">{AGENT_LABELS[r.agent] ?? r.agent}</span>
-        <span className={`text-2xs ${badge.cls}`}>{badge.label}</span>
-
-        <span className="ml-auto flex items-baseline gap-3 text-2xs tabular-nums text-ink-500">
-          {r.status === 'ok' && r.subjective_confidence !== null && (
-            <span>信心 {Math.round(r.subjective_confidence * 100)}%</span>
-          )}
-          {r.status === 'ok' && r.probable_score && (
-            <span className="font-serif font-bold text-ink-800">{r.probable_score}</span>
-          )}
-          <svg viewBox="0 0 20 20" className="h-3 w-3 self-center text-ink-300 transition-transform group-open:rotate-90" fill="currentColor" aria-hidden="true">
-            <path d="M7.3 5.3a1 1 0 011.4 0l4 4a1 1 0 010 1.4l-4 4a1 1 0 01-1.4-1.4L10.6 10 7.3 6.7a1 1 0 010-1.4z" />
-          </svg>
-        </span>
-      </summary>
-
-      <div className="space-y-3 px-1 pb-4 pl-7">
-        {/* 无数据 / 失败时给出明确说明,避免用户以为是空白 bug */}
-        {inactive && (
-          <p className="text-xs leading-relaxed text-ink-500">
-            {r.status === 'no_data' && '该维度没有可用数据,已跳过 LLM 分析以节省额度(不影响其他专家)。'}
-            {r.status === 'error' && '该专家调用失败,本次结论未纳入其视角(fail-open 设计,不阻断整体预测)。'}
-            {r.status === 'parse_error' && '模型输出未通过格式校验,该报告已丢弃。'}
-          </p>
-        )}
-
-        {!inactive && r.home_edge !== null && (
-          <div>
-            <div className="mb-1.5 flex items-baseline justify-between text-2xs">
-              <span className="text-ink-500">主队优势</span>
-              <span className={`font-semibold tabular-nums ${r.home_edge > 0 ? 'text-press' : r.home_edge < 0 ? 'text-ink-700' : 'text-ink-500'}`}>
-                {r.home_edge > 0 ? '+' : ''}{r.home_edge.toFixed(2)}
-              </span>
-            </div>
-            <EdgeBar value={r.home_edge} />
-            <div className="mt-1 flex justify-between text-2xs text-ink-400">
-              <span>利客队</span>
-              <span>利主队</span>
-            </div>
-          </div>
-        )}
-
-        {r.analysis && (
-          <p className="font-serif text-sm leading-loose text-ink-700">{r.analysis}</p>
-        )}
-
-        {r.key_evidence.length > 0 && (
-          <ul className="space-y-1.5">
-            {r.key_evidence.map((e, i) => (
-              <li key={i} className="flex gap-2 text-xs leading-relaxed text-ink-600">
-                <span className="flex-shrink-0 text-ink-300" aria-hidden="true">—</span>
-                <span>{e}</span>
-              </li>
-            ))}
-          </ul>
-        )}
-
-        {r.exp_home_goals !== null && r.exp_away_goals !== null && (
-          <p className="text-xs text-ink-500">
-            进球期望 <span className="font-serif font-bold tabular-nums text-ink-900">{r.exp_home_goals.toFixed(1)} - {r.exp_away_goals.toFixed(1)}</span>
-          </p>
-        )}
-
-        {!inactive && (
-          <p className="border-t border-ink-100 pt-2.5 text-2xs text-ink-400">
-            数据充分度 {SUFFICIENCY_LABEL[r.data_sufficiency] ?? r.data_sufficiency}
-            <span className="mx-2 text-ink-200">|</span>
-            <span className="font-mono">{r.model}</span>
-            {r.latency_ms !== null && <span className="ml-2 tabular-nums">{r.latency_ms}ms</span>}
-          </p>
-        )}
-      </div>
-    </details>
-  )
-}
-
-function PredictionPanel({
-  prediction,
-  match,
-  embedded = false,
-}: {
-  prediction: Prediction
-  match: Match
-  /** 弹窗嵌入模式:弹窗已提供报头,这里省略自带版头 */
-  embedded?: boolean
-}) {
-  const homeName = match.home_team_zh || match.home_team
-  const [expertsOpen, setExpertsOpen] = useState(false)
-  const awayName = match.away_team_zh || match.away_team
-  const degraded = prediction.status === 'degraded' || prediction.status === 'failed'
-  const reports = prediction.agent_outputs ?? []
-  const okReports = reports.filter(r => r.status === 'ok')
-
-  return (
-    <article className={embedded ? 'bg-paper-50' : 'border border-ink-900 bg-paper-50'}>
-      {!embedded && (
-      <div className="flex flex-wrap items-baseline justify-between gap-2 border-b border-ink-900 bg-paper-100 px-4 py-2.5 sm:px-5">
-        <h3 className="flex flex-wrap items-center gap-1.5 font-serif text-sm font-bold text-ink-900">
-          预测版 ·
-          <TeamSideTag side="home" />
-          {homeName}
-          <span>对</span>
-          <TeamSideTag side="away" />
-          {awayName}
-        </h3>
-        <span className="text-2xs tabular-nums text-ink-500">
-          {prediction.provider} / {prediction.model}
-          {prediction.latency_ms !== null && ` · ${(prediction.latency_ms / 1000).toFixed(1)}s`}
-        </span>
-      </div>
-      )}
-
-      <div className="space-y-7 px-4 py-6 sm:px-5">
-        {/* ── degraded / failed 态:醒目警示 + 原因,不展示虚假比分 ── */}
-        {degraded && (
-          <div className="border-l-2 border-press bg-press-wash/40 px-4 py-3">
-            <p className="font-serif text-sm font-bold text-press-dark">
-              {prediction.status === 'failed' ? '预测失败' : '预测降级(degraded)'}
-            </p>
-            <p className="mt-1.5 whitespace-pre-wrap text-xs leading-relaxed text-ink-600">
-              {prediction.reasoning || '所有专家均无有效数据或调用失败,无法生成可靠比分。'}
-            </p>
-          </div>
-        )}
-
-        {/* ── 主结论(仅 success 展示) ── */}
-        {!degraded && (
-          <>
-            <div className="text-center">
-              <p className="font-serif text-5xl font-bold tabular-nums leading-none text-ink-900 sm:text-6xl">
-                {prediction.pred_home_goals ?? '-'}
-                <span className="mx-3 font-normal text-ink-300">:</span>
-                {prediction.pred_away_goals ?? '-'}
-              </p>
-              <p className="mt-3 text-2xs tracking-[0.5em] text-ink-400">预测比分</p>
-              {prediction.alt_pred_home_goals != null && prediction.alt_pred_away_goals != null && (
-                <p className="mt-2 text-2xs tabular-nums text-ink-400">
-                  备选{' '}
-                  <span className="font-serif text-sm font-bold tabular-nums text-ink-600">
-                    {prediction.alt_pred_home_goals}<span className="mx-0.5 font-normal text-ink-300">:</span>{prediction.alt_pred_away_goals}
-                  </span>
-                </p>
-              )}
-            </div>
-
-            <div className="border-y border-ink-200 py-4">
-              <OutcomeLine pick={prediction.pred_1x2} confidence={prediction.subjective_confidence} />
-            </div>
-          </>
-        )}
-
-        {/* ── 成本信息(耗时 + token + 限流余量) ── */}
-        {!degraded && (
-          <PredictionCost prediction={prediction} />
-        )}
-
-        {/* ── 元信息 ── */}
-        <p className="text-center text-2xs text-ink-500">
-          `多专家模式 · ${okReports.length}/${reports.length} 路有效`
-          {prediction.prompt_version && ` · prompt ${prediction.prompt_version}`}
-        </p>
-
-        {/* ── 终裁/降级说明意见 ── */}
-        {prediction.reasoning && degraded && (
-          <section>
-            <h4 className="section-head mb-2">降级原因</h4>
-            <blockquote className="border-l-2 border-press pl-4">
-              <p className="whitespace-pre-wrap font-serif text-sm leading-loose text-ink-700">{prediction.reasoning}</p>
-            </blockquote>
-          </section>
-        )}
-
-        {/* ── 专家意见(多模式):可折叠 + 状态摘要 + 权重条形图 ── */}
-        {reports.length > 0 && (
-          <section>
-            <button
-              onClick={() => setExpertsOpen(o => !o)}
-              className="flex w-full items-center justify-between border-b border-ink-200 pb-2 text-left"
-            >
-              <span className="section-head mb-0">五路专家意见({okReports.length}/{reports.length} 路有效)</span>
-              <span className="text-2xs text-ink-400">{expertsOpen ? '收起' : '展开'}</span>
-            </button>
-
-            {/* 权重条形图(仅 success 且有权重时显示) */}
-            {!degraded && prediction.agent_weights && Object.keys(prediction.agent_weights).length > 0 && (
-              <div className="mt-3 space-y-1.5">
-                <span className="text-2xs text-ink-500">终裁权重分布</span>
-                {Object.entries(prediction.agent_weights)
-                  .sort((a, b) => b[1] - a[1])
-                  .map(([k, v]) => (
-                    <div key={k} className="grid grid-cols-[96px_minmax(0,1fr)_40px] items-center gap-2">
-                      <span className="truncate text-2xs text-ink-500">{AGENT_LABELS[k] ?? k}</span>
-                      <div className="h-1.5 bg-paper-100">
-                        <div className="h-full bg-press" style={{ width: `${Math.round(v * 100)}%` }} />
-                      </div>
-                      <span className="text-right text-2xs tabular-nums text-ink-500">{Math.round(v * 100)}%</span>
-                    </div>
-                  ))}
-              </div>
-            )}
-
-            {expertsOpen && (
-              <div className="mt-2">
-                {reports.map((r, i) => (
-                  <AgentCard key={r.agent} report={r} no={CN_NUM[i] ?? String(i + 1)} />
-                ))}
-              </div>
-            )}
-          </section>
-        )}
-
-        {/* ── 终裁意见(success) ── */}
-        {prediction.reasoning && !degraded && (
-          <section>
-            <h4 className="section-head mb-3">终裁意见</h4>
-            <blockquote className="border-l-2 border-press pl-4">
-              <p className="whitespace-pre-wrap font-serif text-sm leading-loose text-ink-700">{prediction.reasoning}</p>
-            </blockquote>
-          </section>
-        )}
-      </div>
-    </article>
-  )
-}
-
-/** 预测弹窗:进行中显示过程可视化,完成后显示预测版,失败显示原因 */
 export function PredictModal({
   match,
   predicting,

frontend/src/pages/matches/components/OutcomePanel.tsx

@@ -0,0 +1,126 @@
+/**
+ * OutcomePanel: 预测主结论 —— 比分 / 胜平负 / 置信度 / 成本。
+ *
+ * P3-1: 从 MatchPredictPanel.PredictionPanel 拆出,渲染逻辑原样搬迁。
+ */
+import TeamSideTag from '../../../../components/TeamSideTag'
+import type { Match, Prediction } from '../../types'
+import { OUTCOME_LABEL } from '../../types'
+
+/** 置信度细线:0~1 数值的低调可视化 */
+function Meter({ value }: { value: number }) {
+  const pct = Math.max(0, Math.min(100, Math.round(value * 100)))
+  return (
+    <div className="h-px w-full bg-ink-200" role="presentation">
+      <div className="h-px bg-press transition-[width] duration-500" style={{ width: `${pct}%` }} />
+    </div>
+  )
+}
+
+/** 胜平负一行文字:选中的红字加方块标记,未选中的退灰 */
+function OutcomeLine({
+  pick,
+  confidence,
+}: {
+  pick: string | null
+  confidence: number | null
+}) {
+  const options = ['1', 'X', '2'] as const
+  return (
+    <div>
+      <div className="flex items-baseline justify-center gap-6 sm:gap-10">
+        {options.map(o => {
+          const on = pick === o
+          return (
+            <div key={o} className="flex flex-col items-center gap-1">
+              <span className={`flex items-center gap-1.5 text-sm ${on ? 'font-semibold text-press' : 'text-ink-400'}`}>
+                {on && <span className="inline-block h-2 w-2 bg-press" aria-hidden="true" />}
+                {OUTCOME_LABEL[o]}
+              </span>
+              {on && confidence !== null && (
+                <span className="text-2xs tabular-nums text-ink-500">
+                  置信 {Math.round(confidence * 100)}%
+                </span>
+              )}
+            </div>
+          )
+        })}
+      </div>
+      {pick && confidence !== null && (
+        <div className="mx-auto mt-3 max-w-xs">
+          <Meter value={confidence} />
+          <p className="mt-1 text-center text-2xs text-ink-400">主观置信度,非统计概率</p>
+        </div>
+      )}
+    </div>
+  )
+}
+
+/** 预测成本展示:耗时 + token + 限流余量 */
+function PredictionCost({ prediction }: { prediction: Prediction }) {
+  const latency = prediction.latency_ms != null ? `${(prediction.latency_ms / 1000).toFixed(1)}s` : null
+  const tokens = prediction.prompt_tokens != null || prediction.completion_tokens != null
+    ? `${prediction.prompt_tokens ?? '?'}/${prediction.completion_tokens ?? '?'}`
+    : null
+
+  if (!latency && !tokens && prediction.rate_limit_remaining == null) return null
+
+  return (
+    <div className="border-t border-ink-200 pt-3 text-2xs text-ink-500">
+      <div className="flex flex-wrap items-center justify-center gap-x-4 gap-y-1">
+        {latency && (
+          <span className="inline-flex items-center gap-1">
+            <span aria-hidden="true" className="opacity-60">⏱</span>耗时 {latency}
+          </span>
+        )}
+        {tokens && (
+          <span className="inline-flex items-center gap-1">
+            <span aria-hidden="true" className="opacity-60">Tok</span>prompt/completion: {tokens}
+          </span>
+        )}
+        {prediction.rate_limit_remaining != null && prediction.rate_limit_remaining <= 3 && (
+          <span className="text-press" title="每分钟最多 10 次预测">
+            剩余配额: {prediction.rate_limit_remaining}/10(分钟)
+          </span>
+        )}
+      </div>
+    </div>
+  )
+}
+
+export function OutcomePanel({ prediction, match }: { prediction: Prediction; match: Match }) {
+  const homeName = match.home_team_zh || match.home_team
+  const awayName = match.away_team_zh || match.away_team
+  const degraded = prediction.status === 'degraded' || prediction.status === 'failed'
+
+  return (
+    <>
+      {!degraded && (
+        <div className="text-center">
+          <p className="font-serif text-5xl font-bold tabular-nums leading-none text-ink-900 sm:text-6xl">
+            {prediction.pred_home_goals ?? '-'}
+            <span className="mx-3 font-normal text-ink-300">:</span>
+            {prediction.pred_away_goals ?? '-'}
+          </p>
+          <p className="mt-3 text-2xs tracking-[0.5em] text-ink-400">预测比分</p>
+          {prediction.alt_pred_home_goals != null && prediction.alt_pred_away_goals != null && (
+            <p className="mt-2 text-2xs tabular-nums text-ink-400">
+              备选{' '}
+              <span className="font-serif text-sm font-bold tabular-nums text-ink-600">
+                {prediction.alt_pred_home_goals}<span className="mx-0.5 font-normal text-ink-300">:</span>{prediction.alt_pred_away_goals}
+              </span>
+            </p>
+          )}
+        </div>
+      )}
+
+      {!degraded && (
+        <div className="border-y border-ink-200 py-4">
+          <OutcomeLine pick={prediction.pred_1x2} confidence={prediction.subjective_confidence} />
+        </div>
+      )}
+
+      {!degraded && <PredictionCost prediction={prediction} />}
+    </>
+  )
+}

frontend/src/pages/matches/components/ReasoningPanel.tsx

@@ -0,0 +1,34 @@
+/**
+ * ReasoningPanel: 终裁意见 / 降级原因 —— 预测的文本解释。
+ *
+ * P3-1: 从 MatchPredictPanel.PredictionPanel 拆出,渲染逻辑原样搬迁。
+ */
+import type { Prediction } from '../../types'
+
+export function ReasoningPanel({ prediction }: { prediction: Prediction }) {
+  const degraded = prediction.status === 'degraded' || prediction.status === 'failed'
+
+  if (!prediction.reasoning) return null
+
+  // 降级态:reasoning 展示为「降级原因」
+  if (degraded) {
+    return (
+      <section>
+        <h4 className="section-head mb-2">降级原因</h4>
+        <blockquote className="border-l-2 border-press pl-4">
+          <p className="whitespace-pre-wrap font-serif text-sm leading-loose text-ink-700">{prediction.reasoning}</p>
+        </blockquote>
+      </section>
+    )
+  }
+
+  // 成功态:reasoning 展示为「终裁意见」
+  return (
+    <section>
+      <h4 className="section-head mb-3">终裁意见</h4>
+      <blockquote className="border-l-2 border-press pl-4">
+        <p className="whitespace-pre-wrap font-serif text-sm leading-loose text-ink-700">{prediction.reasoning}</p>
+      </blockquote>
+    </section>
+  )
+}

src/api/app.py

@@ -2,6 +2,7 @@
 from __future__ import annotations
 
 import logging
+import os
 from collections.abc import AsyncIterator
 from contextlib import asynccontextmanager
 
@@ -40,6 +41,24 @@ async def lifespan(app: FastAPI) -> AsyncIterator[None]:
             "多 worker 部署请将限流前置到 Nginx/网关,或以单 worker 运行"
         )
 
+    # P3-3:STRICT_SINGLE_WORKER 启动期强制校验,拒绝多 worker 静默配额漂移。
+    # uvicorn 通过 --workers 传入;此处以环境变量 UVICORN_WORKERS 或启动参数判定。
+    # 为避免耦合 uvicorn 内部,仅校验一个显式传入的标记:当 STRICT_SINGLE_WORKER=True 时,
+    # 要求环境变量 UVICORN_WORKERS 不为空且 <=1,否则拒绝启动。
+    if settings.STRICT_SINGLE_WORKER:
+        workers = os.environ.get("UVICORN_WORKERS", "1")
+        try:
+            n_workers = int(workers)
+        except ValueError:
+            n_workers = 1
+        if n_workers > 1:
+            raise RuntimeError(
+                f"STRICT_SINGLE_WORKER=True 但以 {n_workers} worker 启动会被拒绝 "
+                f"(应用内限流/KeyRing 多 worker 下各自独立计数,配额放大 {n_workers} 倍)。"
+                f"请前置 Nginx/网关全局限流后再启用多 worker,或保持单 worker。"
+            )
+        logger.info("STRICT_SINGLE_WORKER=True:已确认单 worker 启动,限流配额不会漂移")
+
     # 注册默认定时任务(如果数据库中没有)
     from src.db.base import AsyncSessionLocal
     from sqlalchemy import select

src/api/routes/predict.py

@@ -64,10 +64,8 @@ async def predict(req: PredictRequest, request: Request):
         raise HTTPException(500, "预测失败,请查看服务器日志")
 
     # D2: 三种模式统一返回 PredictResult —— 字段映射单一化,无 dict 分支。
-    # 仅 baseline 的 prediction_id 需要在此落库补齐(服务层不落库)。
+    # P3-2:baseline 已在服务层(predict_baseline)落库并回填真实 prediction_id,
-    if req.mode == "baseline":
+    # 路由层不再需要特殊的 _persist_baseline,与 single/multi 路径统一。
-        prediction_id = await _persist_baseline(req.match_id, result)
-    else:
     prediction_id = result.prediction_id
 
     # 3. 结果映射(无 DB 访问)
@@ -101,36 +99,6 @@ async def predict(req: PredictRequest, request: Request):
     )
 
 
-async def _persist_baseline(match_id: int, baseline: PredictResult) -> int:
-    """将基线预测结果写入 prediction 表,复用 upsert 语义。"""
-    from src.db.unit_of_work import get_uow
-    from src.llm.predict import _upsert_prediction
-
-    async with get_uow() as session:
-        pred = await _upsert_prediction(
-            session,
-            match_id=match_id,
-            provider_name="baseline",
-            model="baseline",
-            mode="baseline",
-            run_type="live",  # baseline 是 live 预测的变体,符合 ck_run_type_enum
-            values={
-                "prompt_version": baseline.prompt_version,
-                "prompt_tokens": baseline.prompt_tokens or 0,
-                "completion_tokens": baseline.completion_tokens or 0,
-                "latency_ms": baseline.latency_ms or 0,
-                "pred_home_goals": baseline.pred_home_goals,
-                "pred_away_goals": baseline.pred_away_goals,
-                "pred_1x2": baseline.pred_1x2,
-                "subjective_confidence": baseline.subjective_confidence,
-                "reasoning": baseline.reasoning,
-                "raw_response": baseline.raw,
-                "status": "success",
-            },
-        )
-        return pred.id
-
-
 @router.get("/predictions", response_model=list[PredictionOut], dependencies=[Depends(require_admin)])
 async def list_predictions(
     match_id: int | None = None,

src/core/config.py

@@ -11,6 +11,10 @@ class Settings(BaseSettings):
     # --- app ---
     APP_ENV: str = "development"
     LOG_LEVEL: str = "INFO"
+    # P3-3:多 worker 时应用内限流与 KeyRing 各自独立计数(配额放大 N 倍)。
+    # 设为 True 时若以多 worker 启动 uvicorn 则拒绝启动,避免静默配额漂移。
+    # 仅在你已前置 Nginx/网关做全局限流、确认不需要此守护时留空/False。
+    STRICT_SINGLE_WORKER: bool = False
     # 生产环境强制要求管理鉴权配置,即使 APP_ENV=production 也生效。
     # True 时若 auth_configured() 为 False 则拒绝(503),development 保持 fail-open。
     REQUIRE_ADMIN_AUTH: bool = False

src/llm/baseline.py

@@ -12,7 +12,7 @@ from sqlalchemy import case, func, select
 
 from src.db.base import AsyncSession, AsyncSessionLocal
 from src.db.models import Match
-from src.llm.predict import PredictResult
+from src.llm.predict import PredictResult, _upsert_prediction
 
 logger = logging.getLogger(__name__)
 
@@ -58,19 +58,23 @@ async def predict_baseline(
 
     返回 PredictResult(D2 统一结果类型):
       provider=model="baseline", 不调用 LLM,latency_ms≈0。
-    prediction_id 为占位 0 —— baseline 不在服务层落库,
+
-    由路由层 _persist_baseline 落库后取得真实 id。
+    P3-2:baseline 落库下沉到服务层 —— 直接在服务层完成落库并回填真实
+    prediction_id,路由层不再需要特殊的 _persist_baseline,与 single/multi
+    路径统一(result.prediction_id 即可用)。对外 JSON 不变。
     """
+    from src.db.unit_of_work import get_uow
+
     async with AsyncSessionLocal() as db:
         match = await db.get(Match, match_id)
         if match is None:
             raise ValueError(f"match {match_id} not found")
 
         before = None
-        if backtest and match.match_dt:
+        if backtest and match.match_date:
             from datetime import timedelta
 
-            before = match.match_dt - timedelta(days=1)
+            before = match.match_date - timedelta(days=1)
         elif cutoff_at is not None:
             before = cutoff_at
 
@@ -93,8 +97,38 @@ async def predict_baseline(
     else:
         pred_1x2 = "X"
 
+    values = {
+        "prompt_version": "baseline_v1",
+        "prompt_tokens": 0,
+        "completion_tokens": 0,
+        "latency_ms": 0,
+        "pred_home_goals": float(pred_home),
+        "pred_away_goals": float(pred_away),
+        "pred_1x2": pred_1x2,
+        "subjective_confidence": 0.5,
+        "reasoning": (
+            f"基线估计(非投注建议): 主队主场场均进球 {home_avg:.2f} → 预测 {pred_home}; "
+            f"客队客场场均进球 {away_avg:.2f} → 预测 {pred_away}。"
+        ),
+        "raw_response": {"home_avg": round(home_avg, 2), "away_avg": round(away_avg, 2)},
+        "status": "success",
+    }
+
+    # P3-2:服务层落库,回填真实 prediction_id(与 single/multi 统一)。
+    async with get_uow() as session:
+        pred = await _upsert_prediction(
+            session,
+            match_id=match_id,
+            provider_name="baseline",
+            model="baseline",
+            mode="baseline",
+            run_type="live",
+            values=values,
+        )
+        prediction_id = pred.id
+
     return PredictResult(
-        prediction_id=0,  # 占位:真实 id 由路由层 _persist_baseline 落库后返回
+        prediction_id=prediction_id,
         provider="baseline",
         model="baseline",
         prompt_version="baseline_v1",
@@ -105,14 +139,11 @@ async def predict_baseline(
         alt_pred_away_goals=None,
         pred_1x2=pred_1x2,
         subjective_confidence=0.5,
-        reasoning=(
+        reasoning=values["reasoning"],
-            f"基线估计(非投注建议): 主队主场场均进球 {home_avg:.2f} → 预测 {pred_home}; "
-            f"客队客场场均进球 {away_avg:.2f} → 预测 {pred_away}。"
-        ),
         context="",  # baseline 不构建 LLM 上下文
         status="success",
         latency_ms=0,
         prompt_tokens=0,
         completion_tokens=0,
-        raw={"home_avg": round(home_avg, 2), "away_avg": round(away_avg, 2)},
+        raw=values["raw_response"],
     )

tests/test_baseline.py

@@ -5,6 +5,7 @@
 """
 from __future__ import annotations
 
+from types import SimpleNamespace
 from unittest.mock import patch
 
 import pytest
@@ -12,6 +13,24 @@ import pytest
 from src.llm.baseline import _avg_goals, predict_baseline
 
 
+class _FakeUoW:
+    """P3-2:baseline 在服务层落库,测试需 mock get_uow。"""
+
+    async def __aenter__(self):
+        return SimpleNamespace(
+            execute=lambda *a, **k: SimpleNamespace(scalar_one_or_none=lambda: None),
+            add=lambda *a, **k: None,
+            flush=lambda *a, **k: None,
+        )
+
+    async def __aexit__(self, *a):
+        return None
+
+
+async def _fake_upsert(session, **kw):
+    return SimpleNamespace(id=1)
+
+
 @pytest.mark.asyncio
 async def test_avg_goals_no_data_returns_zero():
     """无历史数据时场均进球为 0(不抛异常)。"""
@@ -68,7 +87,9 @@ async def test_predict_baseline_no_llm():
         match_status = "scheduled"
 
     with patch("src.llm.baseline._avg_goals", fake_avg), \
-         patch("src.llm.baseline.AsyncSessionLocal") as SLC:
+         patch("src.llm.baseline.AsyncSessionLocal") as SLC, \
+         patch("src.db.unit_of_work.get_uow", _FakeUoW), \
+         patch("src.llm.baseline._upsert_prediction", _fake_upsert):
         class FakeSession:
             async def get(self, cls, mid):
                 return FakeMatch()
@@ -93,6 +114,8 @@ async def test_predict_baseline_no_llm():
     assert result.pred_1x2 == "X"
     assert result.subjective_confidence == 0.5
     assert "非投注建议" in result.reasoning
+    # P3-2:服务层落库,回填真实 prediction_id
+    assert result.prediction_id == 1
     # 确认未调用任何 LLM 相关模块
     assert "home_10" in captured and "away_20" in captured
 
@@ -112,7 +135,9 @@ async def test_predict_baseline_clamps_to_range():
         match_status = "scheduled"
 
     with patch("src.llm.baseline._avg_goals", fake_avg), \
-         patch("src.llm.baseline.AsyncSessionLocal") as SLC:
+         patch("src.llm.baseline.AsyncSessionLocal") as SLC, \
+         patch("src.db.unit_of_work.get_uow", _FakeUoW), \
+         patch("src.llm.baseline._upsert_prediction", _fake_upsert):
         class FakeSession:
             async def get(self, cls, mid):
                 return FakeMatch()
@@ -128,3 +153,5 @@ async def test_predict_baseline_clamps_to_range():
     assert result.pred_home_goals == 10.0  # clamped
     assert result.pred_away_goals == 0.0   # clamped
     assert result.pred_1x2 == "1"         # 10:0 主胜
+    # P3-2:服务层落库,回填真实 prediction_id
+    assert result.prediction_id == 1

tests/test_d2_predict_unification.py

@@ -62,12 +62,35 @@ async def test_predict_baseline_returns_predict_result():
         async def __aexit__(self, *a):
             return None
 
+    # P3-2:baseline 在服务层落库(get_uow + _upsert_prediction),需 mock 掉。
+    class FakeUoW:
+        async def __aenter__(self):
+            return _make_session()
+
+        async def __aexit__(self, *a):
+            return None
+
+    captured = {}
+
+    async def fake_upsert(session, **kw):
+        captured.update(kw)
+        return SimpleNamespace(id=77)
+
+    # baseline.py 内部 from-import get_uow / _upsert_prediction,需 patch 真实来源模块。
     with patch("src.llm.baseline._avg_goals", fake_avg), \
-         patch("src.llm.baseline.AsyncSessionLocal") as SLC:
+         patch("src.llm.baseline.AsyncSessionLocal") as SLC, \
+         patch("src.db.unit_of_work.get_uow", FakeUoW), \
+         patch("src.llm.baseline._upsert_prediction", fake_upsert):
         SLC.return_value = FakeCM()
 
         result = await predict_baseline(1)
 
+    # P3-2:验证服务层落库被调用且属性映射正确
+    assert captured["match_id"] == 1
+    assert captured["provider_name"] == "baseline"
+    assert captured["run_type"] == "live"
+    assert captured["values"]["pred_home_goals"] == 2.0
+
     assert isinstance(result, PredictResult)
     assert result.mode == "baseline"
     assert result.provider == "baseline"
@@ -144,15 +167,31 @@ def test_predict_route_has_no_dict_branch():
 
 
 # ============================================================
-# 4. _persist_baseline 属性映射(baseline 落库语义不变)
+# 4. P3-2:baseline 服务层落库属性映射(落库已从路由移到 baseline.py)
 # ============================================================
 
 
+class _FakeResult:
+    """支持 .scalar_one_or_none() 的最小假结果集。"""
+
+    def __init__(self, items):
+        self._items = list(items)
+
+    def scalars(self):
+        return self
+
+    def all(self):
+        return self._items
+
+    def scalar_one_or_none(self):
+        return self._items[0] if self._items else None
+
+
 class _FakeUoW:
-    """替代 get_uow 的最小上下文管理器。"""
+    """替代 get_uow 的最小上下文管理器(session.execute 是 async 的)。"""
 
     def __init__(self):
-        self.session = SimpleNamespace()
+        self.session = _make_session()
 
     async def __aenter__(self):
         return self.session
@@ -160,44 +199,67 @@ class _FakeUoW:
     async def __aexit__(self, *a):
         return None
 
+    def __call__(self):
+        return self
+
+
+def _make_session(existing=None):
+    """构造带 async execute / add / flush 的假 session。"""
+    sess = SimpleNamespace()
+
+    async def execute(*a, **k):
+        return _FakeResult(existing or [])
+
+    sess.execute = execute
+    sess.add = lambda *a, **k: None
+
+    async def flush(*a, **k):
+        return None
+
+    sess.flush = flush
+    return sess
+
 
 @pytest.mark.asyncio
-async def test_persist_baseline_maps_attributes(monkeypatch):
+async def test_baseline_service_persists_with_correct_attributes(monkeypatch):
+    """P3-2:baseline 在服务层(predict_baseline)落库,属性映射与路由旧版一致。"""
     captured = {}
 
     async def fake_upsert(session, **kwargs):
         captured.update(kwargs)
         return SimpleNamespace(id=77)
 
+    class FakeMatch:
+        id = 1
+        home_team_id = 10
+        away_team_id = 20
+        league_id = 1
+        match_status = "scheduled"
+
+    class FakeSession:
+        async def get(self, cls, mid):
+            return FakeMatch()
+
+    class FakeSLC:
+        async def __aenter__(self):
+            return FakeSession()
+
+        async def __aexit__(self, *a):
+            return None
+
+    async def fake_avg(db, *, team_id, side, league_id, before):
+        return 2.0 if side == "home" else 1.0
+
+    monkeypatch.setattr("src.llm.baseline._avg_goals", fake_avg)
+    monkeypatch.setattr("src.llm.baseline.AsyncSessionLocal", FakeSLC)
     monkeypatch.setattr("src.db.unit_of_work.get_uow", lambda: _FakeUoW())
-    monkeypatch.setattr("src.llm.predict._upsert_prediction", fake_upsert)
+    # baseline.py 模块级 import _upsert_prediction(第 15 行),需 patch baseline 模块属性
+    monkeypatch.setattr("src.llm.baseline._upsert_prediction", fake_upsert)
 
-    from src.api.routes.predict import _persist_baseline
+    result = await predict_baseline(1)
 
-    baseline = PredictResult(
+    # 落库被调用且属性映射正确
-        prediction_id=0,  # baseline 不在服务层落库,由 _persist_baseline 落库后取得真实 id
+    assert captured, f"predict_baseline 应调用 _upsert_prediction 落库,但 captured 为空(result.prediction_id={result.prediction_id!r})"
-        provider="baseline",
-        model="baseline",
-        prompt_version="baseline_v1",
-        mode="baseline",
-        pred_home_goals=2.0,
-        pred_away_goals=1.0,
-        alt_pred_home_goals=None,
-        alt_pred_away_goals=None,
-        pred_1x2="1",
-        subjective_confidence=0.5,
-        reasoning="r",
-        context="",
-        status="success",
-        latency_ms=0,
-        prompt_tokens=0,
-        completion_tokens=0,
-        raw={"home_avg": 2.1, "away_avg": 1.4},
-    )
-
-    pid = await _persist_baseline(1, baseline)
-
-    assert pid == 77
     assert captured["match_id"] == 1
     assert captured["provider_name"] == "baseline"
     assert captured["model"] == "baseline"
@@ -212,5 +274,28 @@ async def test_persist_baseline_maps_attributes(monkeypatch):
     assert v["prompt_tokens"] == 0
     assert v["completion_tokens"] == 0
     assert v["latency_ms"] == 0
-    assert v["raw_response"] == {"home_avg": 2.1, "away_avg": 1.4}
+    assert v["raw_response"] == {"home_avg": 2.0, "away_avg": 1.0}
     assert v["status"] == "success"
+
+    # 回填真实 prediction_id(服务层落库后取得)
+    assert result.prediction_id == 77
+    assert result.pred_1x2 == "1"
+    assert captured["match_id"] == 1
+    assert captured["provider_name"] == "baseline"
+    assert captured["model"] == "baseline"
+    assert captured["mode"] == "baseline"
+    assert captured["run_type"] == "live"
+    v = captured["values"]
+    assert v["prompt_version"] == "baseline_v1"
+    assert v["pred_home_goals"] == 2.0
+    assert v["pred_away_goals"] == 1.0
+    assert v["pred_1x2"] == "1"
+    assert v["subjective_confidence"] == 0.5
+    assert v["prompt_tokens"] == 0
+    assert v["completion_tokens"] == 0
+    assert v["latency_ms"] == 0
+    assert v["raw_response"] == {"home_avg": 2.0, "away_avg": 1.0}
+    assert v["status"] == "success"
+
+    # 回填真实 prediction_id(服务层落库后取得)
+    assert result.prediction_id == 77