SBrainCO/docs/智脑实施方法论/07-步骤六-AI协同后端API构建与验证.md

07 · 步骤六：AI协同后端API构建与验证

目标：AI参与API开发，逐个验证——快速交付能用的代码，不追求完美架构。 [Delta主导，AI为主]

1. 指标分层设计

L0: 原始数据层 (bill_records, salary_detail_records, attendance_records)
       ↓
L1: 预聚合层 (物化视图 mv_*)
       ↓
L2: API层 (后端路由，SQL查询 + 业务逻辑)
       ↓
L3: 展示层 (前端页面，图表 + 表格)

2. 页面与API总览

页面	文件	调用的 API
总部驾驶舱	DashboardPage.tsx	/overview, /overview/daily, /stores/risk, /stores/priority, /tasks/loop-health, /stores/quadrant, /platform/economics, /situational-awareness/alerts, /store-expense/overview, /overview/yoy, /overview/mom, /overview/trend, /overview/profit-trend
老板驾驶舱	BossPage.tsx	/overview, /overview/daily, /store-expense/overview, /overview/profit-waterfall, /stores/risk, /stores/priority, /cost-analysis/store-overview, /store-expense/expense-structure, /overview/store-profit-ranking, /overview/profit-opportunity
门店工作台	StorePage.tsx	/stores/risk, /stores/priority, /tasks/stores/:code/daily-card, /tasks, /stores/:code/daily, /tasks/followup, /sku/attach, /sku/abc, /stores/:code, /situational-awareness/health-score, /stores/:code/meal-period, /stores/:code/category-mix, /stores/:code/cost, /stores/:code/member, /stores/:code/anomalies, /stores/:code/staffing
成本分析	CostPage.tsx	/cost/comparison, /cost/inventory, /cost/category-benchmark
成本分析(Tab)	cost-analysis/*.tsx	/cost-analysis/overview, /category-comparison, /margin-deviation, /variance-top, /menu-engineering, /profitability, /pricing, /store-overview, /store-ranking, /bom-, /packaging-, /material-, /data-quality, /unmatched-
费用分析(Tab)	store-expense/*.tsx	/store-expense/overview, /expense-structure, /store-ranking, /store-contribution, /break-even, /loss-diagnosis, /delivery-commission, /rent-risk, /fixed-variable, /efficiency, /store-evaluation
风险监控	RiskPage.tsx	/risk/anomaly, /risk/zero-received, /risk/cashier
态势感知	SituationalAwarenessPage.tsx	/situational-awareness/health-score, /alerts, /correlation, /forecast
会员复购	MemberPage.tsx	/member/comparison, /member/repeat
SKU分析	SKUPage.tsx	/sku/abc, /sku/category
菜单工程	MenuEngineeringPage.tsx	/analytics-enhanced/menu-engineering/actions
产品生命周期	ProductLifecyclePage.tsx	/product/products, /product/reviews/overview, /product/reviews
BOM穿透	BomPenetrationPage.tsx	/central-kitchen/bom-penetration
分销对账	DistributionReconciliationPage.tsx	/distribution/reconciliation
生产计划	ProductionPlanPage.tsx	/sales-driven/production-plan
任务闭环	TasksPage.tsx	/tasks
数据导入	DataImportPage.tsx	/product/import-logs, /product/quality-rules
本体浏览	OntologyPage.tsx	/tasks/ontology/*

详细的逐页面→API→SQL→数据源字段映射，见 08-数据溯源与API全量清单.md

3. 后端API规范

3.1 路由组织

server/src/routes/
├── data.ts              # 核心数据API（概览、门店、风险明细等）
├── situational-awareness.ts  # 态势感知（健康度、预警、关联分析）
├── smart-scheduling.ts  # 智能排班
├── cost-analysis.ts     # 成本分析
├── store-expense.ts     # 门店费用
├── analytics-enhanced.ts # 增强分析
├── admin.ts             # 管理
└── tts.ts               # 语音

2.2 通用模式

// 月份参数解析
const month = parseMonth(req)  // 从 query 参数获取，默认 '2026-04'

// 分页
const { page, pageSize, offset } = parsePagination(req)

// 数据权限
const scope = getDataScope(req)  // 返回 store_names 数组或 null（全部）

// 响应格式
sendSuccess(res, data, meta?)   // { success: true, data, meta }
sendError(res, message)         // { success: false, error: message }

2.3 SQL查询模式

安全参数化：

// 正确：参数化查询
const result = await query(`SELECT * FROM mv_risk_anomaly WHERE month = to_char($1::date, 'YYYY-MM')`, [month])

// 错误：字符串拼接
const result = await query(`SELECT * FROM mv_risk_anomaly WHERE month = '${month}'`)

动态条件构建：

const conditions: string[] = [`month = to_char($1::date, 'YYYY-MM')`]
const params: any[] = [month]
let paramIdx = 2
if (storeName) {
  conditions.push(`store_name = $${paramIdx}`)
  params.push(storeName)
  paramIdx++
}
const whereClause = conditions.join(' AND ')

聚合汇总+分页：

// 先查总数和汇总
const countResult = await query(`SELECT count(*) AS total, sum(consumption) AS sum_consumption FROM mv_risk_anomaly WHERE ${whereClause}`, params)
// 再查明细
const result = await query(`SELECT * FROM mv_risk_anomaly WHERE ${whereClause} ORDER BY ${sortCol} ${order} LIMIT $${paramIdx} OFFSET $${paramIdx + 1}`, [...params, pageSize, offset])

3. 常见SQL陷阱与修复

3.1 Schema前缀错误

症状：relation "analytics.mv_channel_daily" does not exist

排查：

SELECT schemaname, matviewname FROM pg_matviews WHERE matviewname LIKE '%channel%';

修复：确认实际schema，移除或修正前缀

// 错误：FROM analytics.mv_channel_daily
// 正确：FROM mv_channel_daily  (实际在public schema)

3.2 列名不存在

症状：column "risk_score" does not exist

排查：

SELECT column_name FROM information_schema.columns WHERE table_name = 'mv_store_risk_rating_monthly';

修复策略：

• 策略A：改用 SELECT *（当不需要特定列名时）
• 策略B：用 0 AS column_name 替代不存在的列（当前端期望该字段时）
• 策略C：修正为正确的列名

3.3 日期格式不匹配

症状：薪资相关查询返回0条

排查：

SELECT DISTINCT salary_period FROM salary_detail_records ORDER BY 1 DESC LIMIT 5;
-- 结果：2026年4月（非2026-04）

修复：

// 错误：s.salary_month = to_char($1::date, 'YYYY-MM')
// 正确：s.salary_period = to_char($1::date, 'YYYY"年"FMMM"月"')

3.4 物化视图stale

症状：API返回数据与直接查原始表不一致

排查：对比物化视图和原始表的count

SELECT count(*) FROM mv_risk_anomaly WHERE month = '2026-04';
SELECT count(*) FROM bill_records WHERE to_char(c176::timestamp, 'YYYY-MM') = '2026-04' AND ...;

修复：重建物化视图

DROP MATERIALIZED VIEW mv_risk_anomaly;
CREATE MATERIALIZED VIEW mv_risk_anomaly AS ...;
CREATE INDEX idx_mv_risk_anomaly_month ON mv_risk_anomaly(month);

4. 考勤打卡数据解析

4.1 department路径解析门店名

北京西部马华餐饮有限公司/西部马华品牌门店/胡庆鹏区/HQP2区/双安店/前厅/服务组

提取逻辑（从后往前找以"店"结尾的层级）：

function extractStore(dept: string): string {
  const parts = dept.split('/')
  for (let i = parts.length - 1; i >= 0; i--) {
    if (parts[i].endsWith('店')) return parts[i]
  }
  return ''
}

4.2 打卡时间解析

格式：08:43(考勤机：指纹)\n21:23(考勤机：指纹)

function parseClockTimes(raw: string): { start: number; end: number } | null {
  const times = raw.match(/(\d{1,2}):(\d{2})/g)
  if (!times || times.length < 2) return null
  const startHour = parseInt(times[0].match(/(\d{1,2}):/)[1])
  let endHour = parseInt(times[times.length - 1].match(/(\d{1,2}):/)[1])
  if (endHour < startHour) endHour = 23  // 跨天
  return { start: startHour, end: endHour }
}

4.3 日均在岗人数估算

// 每个员工每天打卡算1人在岗，按小时累计
// 月度汇总后除以30天得到日均同时在岗人数
staffSummary[store][hour] = Math.round(totalPersonHours[store][hour] / 30)

关键点：客流数据 mv_bill_hourly.bills 也是月度汇总，展示时需除以30天对齐。