candidate-pool 3.0.0 → 3.0.2

package/eod/lib/http.js

@@ -22,6 +22,10 @@
 
 const https = require('https');
 
+// keep-alive 连接复用：省掉每次请求的 TCP/TLS 握手（全流程 ~500-700 次请求）。
+// 经验证（Node 22）：空闲套接字不阻塞进程退出，脚本可正常结束。
+const AGENT = new https.Agent({ keepAlive: true, maxSockets: 8, maxFreeSockets: 8, keepAliveMsecs: 30000 });
+
 // 用完整浏览器 UA：残缺 UA（如裸 "Mozilla/5.0"）本身就是风控指纹特征。
 const UA = 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/126.0.0.0 Safari/537.36';
 
@@ -60,68 +64,67 @@ function looksLikeWaf(txt) {
  *   - isWaf    自定义「这条响应算不算被拦截」判定（默认 looksLikeWaf）
  * @returns {Promise<any>} 文本或已解析的 JSON
  */
-function httpGet(url, opts = {}) {
+async function httpGet(url, opts = {}) {
   const {
     enc = 'utf8', json = false, retries = 4, timeout = 15000,
     referer = 'https://gu.qq.com/', isWaf = null,
   } = opts;
 
-  return new Promise(async (resolve, reject) => {
-    let lastErr = null;
-    let wafHits = 0;
-
-    for (let i = 0; i < retries; i++) {
-      if (i > 0) {
-        // 被拦截 -> 长退避；普通网络抖动 -> 短退避
-        const wait = (lastErr && lastErr.waf)
-          ? WAF_BACKOFF[Math.min(wafHits - 1, WAF_BACKOFF.length - 1)]
-          : 500 * i;
-        await sleep(wait);
-      }
-      try {
-        const { status, txt } = await new Promise((res, rej) => {
-          const req = https.get(url, {
-            timeout,
-            headers: {
-              'User-Agent': UA,
-              Referer: referer,
-              Accept: '*/*',
-              'Accept-Language': 'zh-CN,zh;q=0.9',
-            },
-          }, (resp) => {
-            const chunks = [];
-            resp.on('data', (c) => chunks.push(c));
-            resp.on('end', () => res({ status: resp.statusCode, txt: decode(Buffer.concat(chunks), enc) }));
-          });
-          req.on('error', rej);
-          req.on('timeout', () => { req.destroy(); rej(new Error('timeout')); });
+  let lastErr = null;
+  let wafHits = 0;
+
+  for (let i = 0; i < retries; i++) {
+    if (i > 0) {
+      // 被拦截 -> 长退避；普通网络抖动 -> 短退避
+      const wait = (lastErr && lastErr.waf)
+        ? WAF_BACKOFF[Math.min(wafHits - 1, WAF_BACKOFF.length - 1)]
+        : 500 * i;
+      await sleep(wait);
+    }
+    try {
+      const { status, txt } = await new Promise((res, rej) => {
+        const req = https.get(url, {
+          timeout,
+          agent: AGENT,
+          headers: {
+            'User-Agent': UA,
+            Referer: referer,
+            Accept: '*/*',
+            'Accept-Language': 'zh-CN,zh;q=0.9',
+          },
+        }, (resp) => {
+          const chunks = [];
+          resp.on('data', (c) => chunks.push(c));
+          resp.on('end', () => res({ status: resp.statusCode, txt: decode(Buffer.concat(chunks), enc) }));
         });
+        req.on('error', rej);
+        req.on('timeout', () => { req.destroy(); rej(new Error('timeout')); });
+      });
 
-        // WAF 拦截：状态码判定 + 正文嗅探，两者都查
-        const waf = (isWaf ? isWaf(txt, status) : false) || status === 501 || looksLikeWaf(txt);
-        if (waf) {
-          wafHits++;
-          const e = new Error(`被接口反爬拦截（HTTP ${status}）`);
-          e.waf = true;
-          throw e;
-        }
-        if (status && status >= 400) throw new Error(`HTTP ${status}`);
-
-        if (!json) return resolve(txt.trim());
-        const body = txt.trim();
-        try { return resolve(JSON.parse(body)); }
-        catch (e) {
-          // 兼容 JSONP 包裹：name({...});
-          const m = body.match(/^[a-zA-Z0-9_$.]+\(([\s\S]*)\);?$/);
-          if (m) { try { return resolve(JSON.parse(m[1])); } catch (e2) { /* 落到下面报错 */ } }
-          throw new Error('JSON 解析失败: ' + body.slice(0, 120));
-        }
-      } catch (e) {
-        lastErr = e;
+      // WAF 拦截：状态码判定 + 正文嗅探，两者都查
+      const waf = (isWaf ? isWaf(txt, status) : false) || status === 501 || looksLikeWaf(txt);
+      if (waf) {
+        wafHits++;
+        const e = new Error(`被接口反爬拦截（HTTP ${status}）`);
+        e.waf = true;
+        throw e;
+      }
+      if (status && status >= 400) throw new Error(`HTTP ${status}`);
+
+      if (!json) return txt.trim();
+      const body = txt.trim();
+      try { return JSON.parse(body); }
+      catch (e) {
+        // 兼容 JSONP 包裹：name({...});
+        const m = body.match(/^[a-zA-Z0-9_$.]+\(([\s\S]*)\);?$/);
+        if (m) { try { return JSON.parse(m[1]); } catch (e2) { /* 落到下面报错 */ } }
+        throw new Error('JSON 解析失败: ' + body.slice(0, 120));
       }
+    } catch (e) {
+      lastErr = e;
     }
-    reject(lastErr || new Error('httpGet 失败'));
-  });
+  }
+  throw lastErr || new Error('httpGet 失败');
 }
 
 /**

package/eod/lib/preopen_store.js

@@ -4,6 +4,9 @@
  * ---------------------------------------------------------------------------
  * 一天一份文件：preopen/data/preopen-YYYYMMDD.json
  *   - records: code -> 候选明细（含 P1/P2/P3 分项、基线、标签、总分）
+ *   - nearMiss: 贴线样本（总分落在 [minTotal-10, minTotal) 的 Top10，带完整分项）——
+ *     竞价特征无法事后从 K 线重建，被拒样本不留就无法回溯调「入选线」（2026-10-22 补）
+ *   - meta.rejByReason: 各淘汰原因计数（漏斗/通过率分析）
  *   - runs[]: 当日多次运行留痕（幂等：同文件整文件覆盖，records 按 code 唯一）
  *   - close：收盘回填后追加当日收盘/涨跌幅/是否涨停/命中标记（preopen_backfill.js）
  *
@@ -27,10 +30,13 @@ function loadRun(cfg, tradeDate) {
 
 /**
  * 落库（幂等：同文件整文件覆盖，runs[] 追加本次运行）。
- * @param {object} p { cfg, tradeDate, scope, candidates, meta, marketDate }
+ * @param {object} p { cfg, tradeDate, scope, candidates, nearMiss?, rejByReason?, meta, marketDate }
+ *   - nearMiss：贴线样本（总分 [minTotal-10, minTotal) 的 Top10，带完整分项），
+ *     供日后调「入选线」用；竞价特征无法事后重建，必须运行时留。语义 = 最近一次运行覆盖。
+ *   - rejByReason：各淘汰原因计数（漏斗分析用）。
  * @returns {{file, recordCount, runCount}}
  */
-function saveRun({ cfg, tradeDate, scope, candidates, meta, marketDate }) {
+function saveRun({ cfg, tradeDate, scope, candidates, nearMiss, rejByReason, meta, marketDate }) {
   const dir = path.join(paths.eodDir(), cfg.store.dataDir);
   fs.mkdirSync(dir, { recursive: true });
   const f = dataFile(cfg, tradeDate);
@@ -55,10 +61,13 @@ function saveRun({ cfg, tradeDate, scope, candidates, meta, marketDate }) {
     marketDate,
     meta: Object.assign({}, meta, { recordCount: Object.keys(records).length, runCount: runs.length }),
     runs,
+    // 贴线样本（最近一次运行覆盖）；rejByReason 挂在 meta 里
+    nearMiss: Array.isArray(nearMiss) ? nearMiss : (prev.nearMiss || []),
     // 保留历史回填结果（若有）
     close: prev.close || null,
     records,
   };
+  if (rejByReason && Object.keys(rejByReason).length) doc.meta.rejByReason = rejByReason;
   fs.writeFileSync(f, JSON.stringify(doc, null, 2), 'utf8');
   return { file: f, recordCount: doc.meta.recordCount, runCount: runs.length };
 }

package/eod/preopen_screener.js

@@ -249,6 +249,7 @@ async function run() {
   // 规则 + 打分
   log('[3/4] 应用 P1/P2/P3 打分…');
   const candidates = [];
+  const nearMissAll = [];
   const gaps = [];
   for (const q of quotes) {
     const b = baseline.get(q.code);
@@ -261,10 +262,8 @@ async function run() {
     const p2 = PR.calcP2(q, b, cfg);
     const p3 = PR.calcP3(b, cfg);
     const sc = PR.scorePreopen(p1, p2, p3, cfg);
-    if (sc.total < cfg.score.minTotal) continue;
-
     const sector = sectorOf(map, q.code) || UNKNOWN;
-    candidates.push({
+    const rec = {
       code: q.code, name: q.name, sector,
       board: /sh68[89]/.test(q.code) ? '科创板' : /^sz30/.test(q.code) ? '创业板' : /^bj/.test(q.code) ? '北交所' : '主板',
       open: q.open, prevClose: q.prevClose, price: q.price,
@@ -277,9 +276,23 @@ async function run() {
       total: +sc.total.toFixed(1),
       scoreItems: { p1: +p1.score.toFixed(2), p2: +p2.score.toFixed(2), p3: +p3.score.toFixed(2) },
       stamp: q.stamp,
-    });
+    };
+    if (sc.total < cfg.score.minTotal) {
+      // 贴线样本（2026-10-22 补）：总分落在 [minTotal-10, minTotal) 的进归档 nearMiss 段。
+      // 竞价量比等特征无法事后从 K 线重建，被拒样本不留就无法回溯调「入选线」本身。
+      if (sc.total >= cfg.score.minTotal - 10) {
+        nearMissAll.push(Object.assign(rec, { belowMin: +(cfg.score.minTotal - sc.total).toFixed(1) }));
+      }
+      continue;
+    }
+    candidates.push(rec);
   }
   candidates.sort((a, b) => b.total - a.total);
+  nearMissAll.sort((a, b) => b.total - a.total);
+  const nearMiss = nearMissAll.slice(0, 10);
+  // 各淘汰原因计数（gaps 明细不落盘，只留统计——6000+ 只/天的明细太肥且没用）
+  const rejByReason = {};
+  for (const g of gaps) rejByReason[g.reason] = (rejByReason[g.reason] || 0) + 1;
 
   // 分组
   const bySector = new Map();
@@ -297,8 +310,9 @@ async function run() {
 
   const result = {
     candidates, groups, top: candidates.slice(0, cfg.group.topN),
-    stats: { snapshotCount: quotes.length, baselineCount: baseline.size, candidateCount: candidates.length, groupCount: groups.length, gapCount: gaps.length },
+    stats: { snapshotCount: quotes.length, baselineCount: baseline.size, candidateCount: candidates.length, groupCount: groups.length, gapCount: gaps.length, rejByReason },
     gaps,
+    nearMiss,
   };
 
   log(`[4/4] 落库 + 报告 + 通知…`);
@@ -309,6 +323,7 @@ async function emitResult(p) {
   const { tradeDate, scope, result, marketDate } = p;
   const saved = store.saveRun({
     cfg, tradeDate, scope, candidates: result.candidates,
+    nearMiss: result.nearMiss, rejByReason: result.stats.rejByReason,
     meta: { snapshotCount: result.stats.snapshotCount, baselineCount: p.baselineCount, validQuotes: p.quotesCount, marketDate, elapsedMs: Date.now() - (p._t0 || Date.now()) },
     marketDate,
   });

package/eod/tail.config.js

@@ -30,7 +30,10 @@ const RANGES = [
 ];
 
 module.exports = {
-  version: 1,
+  // 口径版本：写进每日归档 JSON 顶层与报告页脚，参数/打分口径变更时 +1，
+  // 回溯按版本切段（v1 = 2026-09-18~10-21 六项打分含 sector；v2 = 2026-10-22 起五项 35/24/18/12/11 + 新门槛）。
+  // ⚠️ 新旧版本的总分/分项不可直接横向比较，复盘对比必须先按 version 分组。
+  version: 2,
 
   // =========================================================================
   // 1. 交易时段（FR-01 / Q1：不依赖外部日历，靠行情时间戳 + 本地时钟判断）
@@ -212,6 +215,9 @@ module.exports = {
     // 观察模式（14:30~14:50）另存 eod-<交易日>-obs.json，不污染正式口径；
     // 一天一个文件，多次观察靠 runs[] 留痕（--stamp 才会在文件名里带时点）。
     keepRunHistory: 20,      // 每个文件最多保留多少条运行历史
+    // raw/（快照+分时原始存档）**永久保留**（2026-10-22 确认）：
+    // 它是「改打分权重后离线重算历史」的唯一凭据（.tmp/smoke_newcfg.js 即靠它冒烟），
+    // 单份几 MB，磁盘成本可忽略；删除 = 该日历史永远只能按旧口径分析。
   },
 
   // =========================================================================

package/gen_candidates.js

@@ -42,6 +42,9 @@
 const fs = require('fs');
 const path = require('path');
 const https = require('https');
+// keep-alive 连接复用（与 eod/lib/http.js 同口径）：观察池逐只抓取 ~360×3 类请求，
+// 复用 TCP/TLS 省掉大半握手耗时。空闲套接字经验证不阻塞进程退出（Node 22）。
+const AGENT = new https.Agent({ keepAlive: true, maxSockets: 8, maxFreeSockets: 8, keepAliveMsecs: 30000 });
 const { now: nowStr, dayjs } = require('./time');
 // 工作目录解析：产物/配置的落点走 paths，代码资产（本文件等）仍锚在包内。
 // 默认 HOME = 包根（与历史 __dirname 行为一致）；可用 CANDIDATE_POOL_HOME / --cwd 重定位。
@@ -50,7 +53,7 @@ const paths = require('./paths');
 const scaffold = require('./scaffold');
 // 规则引擎纯函数（R01/R07/R05/行业中位数/成交量单位/时区锚定）统一走 rules.js，
 // 单测见 rules.test.js，避免与 gen_candidates 内联实现漂移导致双份真相。
-const { lastClosedTradingDay, nextTradingDay, calcTurnover, computeSectors, calcR01, isLaggard, verifyR05 } = require('./rules');
+const { lastClosedTradingDay, nextTradingDay, calcTurnover, computeSectors, calcR01, isLaggard, verifyR05, G } = require('./rules');
 
 // ---------- 数据源：腾讯公开行情接口（无需任何第三方 CLI / 内置技能）----------
 const QT_QUOTE = 'https://qt.gtimg.cn/q=';
@@ -121,44 +124,42 @@ function looksLikeWaf(txt) {
 const UA = 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/126.0.0.0 Safari/537.36';
 const WAF_BACKOFF = [2000, 5000, 12000, 25000]; // 命中拦截时的退避梯度（ms）
 
-function httpGet(url, { json = true, retries = 4, timeout = 15000 } = {}) {
-  return new Promise(async (resolve, reject) => {
-    let lastErr;
-    let wafHits = 0;
-    for (let i = 0; i < retries; i++) {
-      if (i > 0) {
-        // 拦截用长退避，普通网络错误用短退避
-        await sleep(lastErr && lastErr.waf
-          ? WAF_BACKOFF[Math.min(wafHits - 1, WAF_BACKOFF.length - 1)]
-          : 500 * i);
-      }
-      try {
-        const { status, txt } = await new Promise((res, rej) => {
-          const req = https.get(url, { timeout, headers: { 'User-Agent': UA, 'Referer': 'https://gu.qq.com/', 'Accept': '*/*' } }, (resp) => {
-            let d = ''; resp.on('data', c => d += c); resp.on('end', () => res({ status: resp.statusCode, txt: d.trim() }));
-          });
-          req.on('error', rej);
-          req.on('timeout', () => { req.destroy(); rej(new Error('timeout')); });
-        });
-        // WAF 拦截返回 HTTP 501 + 跳转页；状态码判定比嗅探 body 更可靠，两者都查。
-        if (status === 501 || looksLikeWaf(txt)) {
-          wafHits++;
-          const e = new Error(`被接口反爬拦截（HTTP ${status}）`);
-          e.waf = true;
-          throw e;
-        }
-        if (status && status >= 400) throw new Error(`HTTP ${status}`);
-        if (!json) return resolve(txt);
-        try { return resolve(JSON.parse(txt)); }
-        catch (e) {
-          const m = txt.match(/^[a-zA-Z0-9_]+\(([\s\S]*)\);?\s*$/);
-          if (m) { try { return resolve(JSON.parse(m[1])); } catch (e2) {} }
-          throw new Error('JSON 解析失败: ' + txt.slice(0, 120));
-        }
-      } catch (e) { lastErr = e; }
+async function httpGet(url, { json = true, retries = 4, timeout = 15000 } = {}) {
+  let lastErr;
+  let wafHits = 0;
+  for (let i = 0; i < retries; i++) {
+    if (i > 0) {
+      // 拦截用长退避，普通网络错误用短退避
+      await sleep(lastErr && lastErr.waf
+        ? WAF_BACKOFF[Math.min(wafHits - 1, WAF_BACKOFF.length - 1)]
+        : 500 * i);
     }
-    reject(lastErr || new Error('httpGet 失败'));
-  });
+    try {
+      const { status, txt } = await new Promise((res, rej) => {
+        const req = https.get(url, { timeout, agent: AGENT, headers: { 'User-Agent': UA, 'Referer': 'https://gu.qq.com/', 'Accept': '*/*' } }, (resp) => {
+          let d = ''; resp.on('data', c => d += c); resp.on('end', () => res({ status: resp.statusCode, txt: d.trim() }));
+        });
+        req.on('error', rej);
+        req.on('timeout', () => { req.destroy(); rej(new Error('timeout')); });
+      });
+      // WAF 拦截返回 HTTP 501 + 跳转页；状态码判定比嗅探 body 更可靠，两者都查。
+      if (status === 501 || looksLikeWaf(txt)) {
+        wafHits++;
+        const e = new Error(`被接口反爬拦截（HTTP ${status}）`);
+        e.waf = true;
+        throw e;
+      }
+      if (status && status >= 400) throw new Error(`HTTP ${status}`);
+      if (!json) return txt;
+      try { return JSON.parse(txt); }
+      catch (e) {
+        const m = txt.match(/^[a-zA-Z0-9_]+\(([\s\S]*)\);?\s*$/);
+        if (m) { try { return JSON.parse(m[1]); } catch (e2) {} }
+        throw new Error('JSON 解析失败: ' + txt.slice(0, 120));
+      }
+    } catch (e) { lastErr = e; }
+  }
+  throw lastErr || new Error('httpGet 失败');
 }
 
 // 候选源：HOME 下 candidates.json 配置的可编辑观察池（替代 westock ranking CompScore）
@@ -423,6 +424,10 @@ async function main() {
   paths.ensureDir(path.dirname(snapshotPath));   // --dump 指向新目录时也要能写出去
   const snap = {
     anchor, target, limit, topN,
+    generatedAt: new Date().toISOString(),
+    // 快照当时的规则常量（rules.js G）：日后调参（如 CHG_MIN 3→4）时，
+    // 历史门槛可按当时的常量重算，也可按新常量对 stocks[].chg 等原始值离线重切。
+    rulesG: { ...G },
     breadth, breadthIsAnchor, overview,
     stocks: stocks.map(s => { const c = Object.assign({}, s); delete c.kline; delete c.quote; delete c.minute; return c; }),
     sectors: validSectors.map(s => { const c = Object.assign({}, s); delete c.members; return c; }),
@@ -483,7 +488,7 @@ function buildHTML(m) {
 
   // 核心结论
   const best = [...universe].sort((a, b) => (b.r01.core || 0) - (a.r01.core || 0))[0];
-  const bestStr = best && best.r01.core ? `${esc(best.name)}（${best.code.replace(/^[sh|sz]/, '')}），四项中达成 ${best.r01.core} 项${best.r01.capOk ? '，市值达标' : ''}` : '无';
+  const bestStr = best && best.r01.core ? `${esc(best.name)}（${best.code.replace(/^(sh|sz|bj)/, '')}），四项中达成 ${best.r01.core} 项${best.r01.capOk ? '，市值达标' : ''}` : '无';
   const conclusion = [
     `<b>${b('R01 ' + (r01Triggers ? `硬触发 ${r01Triggers} 只` : '零硬触发'), 'R01 ' + (r01Triggers ? `${r01Triggers} hard trigger(s)` : 'zero hard triggers'))}。</b>` +
       (r01Triggers ? '' : `${universe.length} 只标的无一满足全部门槛。最接近者为 ${bestStr}。`),
@@ -518,7 +523,7 @@ function buildHTML(m) {
     const tier = tierOf.get(s.code) || 'excluded';
     const cls = r.ok ? ' class="hit"' : '';
     matrixRows += `<tr data-tier="${tier}"${cls}>` +
-      `<td>${s.code.replace(/^[sh|sz]/, '')}</td>` +
+      `<td>${s.code.replace(/^(sh|sz|bj)/, '')}</td>` +
       `<td>${b(s.name, s.name)}</td>` +
       `<td><span class="tag tier-${tier}">${TIER_LABEL[tier]}</span></td>` +
       `<td>${num(r.close)}</td>` +
@@ -541,7 +546,7 @@ function buildHTML(m) {
   let exRows = '';
   for (const s of excluded) {
     exRows += `<tr>` +
-      `<td>${s.code.replace(/^[sh|sz]/, '')}</td>` +
+      `<td>${s.code.replace(/^(sh|sz|bj)/, '')}</td>` +
       `<td>${b(s.name, s.name)}</td>` +
       `<td class="wrap">${(s.excludeReasons || ['—']).map(r => b(r, r)).join('；')}</td>` +
       `</tr>`;
@@ -557,7 +562,7 @@ function buildHTML(m) {
     const r07 = s.r07 || {};
     return `<div class="card ${label === 'priority' ? 'priority' : ''}">` +
       `<div class="grid-2">` +
-      `<div><div class="kv"><span class="label">${b('名称', 'Name')}：</span><span class="value">${b(s.name, s.name)}（${s.code.replace(/^[sh|sz]/, '')}）</span></div>` +
+      `<div><div class="kv"><span class="label">${b('名称', 'Name')}：</span><span class="value">${b(s.name, s.name)}（${s.code.replace(/^(sh|sz|bj)/, '')}）</span></div>` +
       `<div class="kv"><span class="label">${b('R01 达成', 'R01 score')}：</span><span class="value">${r.core}/4${r.capOk ? '，市值达标' : '，市值越线'}</span></div>` +
       `<div class="kv"><span class="label">${b('行业', 'Sector')}：</span><span class="value">${r07.sectorName || '—'}（${pct(r07.sectorPct)}，排名 ${r07.sectorRank || '—'}/${r07.total || '—'}）</span></div></div>` +
       `<div>${fund}` +

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "candidate-pool",
-  "version": "3.0.0",
+  "version": "3.0.2",
   "description": "A股短线选股工具包：次日候选池初筛（R01 量能验证突破 + R07 板块内补涨）与尾盘选股（14:50 口径），数据取自腾讯公开行情接口，输出单文件双语 HTML 报告",
   "bin": {
     "candidate-pool": "bin/candidate-pool.js",
@@ -85,16 +85,16 @@
     "prepare": "husky"
   },
   "devDependencies": {
-    "@commitlint/cli": "^21.2.3",
-    "@commitlint/config-conventional": "^21.2.3",
+    "@commitlint/cli": "^21.2.4",
+    "@commitlint/config-conventional": "^21.2.4",
     "changelogen": "^0.6.2",
     "husky": "^9.1.7"
   },
   "engines": {
-    "node": ">=18"
+    "node": ">=22.5"
   },
   "dependencies": {
     "dayjs": "^1.11.23",
-    "dotenv": "^18.0.1"
+    "dotenv": "^18.0.8"
   }
 }