tihao-ai/claude-code-tihao-sourcing/scripts/overnight-manual-review-smoke.js

#!/usr/bin/env node
const fs = require('fs');
const os = require('os');
const path = require('path');
const { spawnSync } = require('child_process');
const { buildMetrics, readCsv } = require('./review-metrics');

const root = path.resolve(__dirname, '..');

function main() {
  const output = fs.mkdtempSync(path.join(os.tmpdir(), 'tihao-overnight-review-'));
  const result = spawnSync(process.execPath, [
    path.join(root, 'scripts', 'overnight-quality-campaign.js')
  ], {
    cwd: root,
    encoding: 'utf8',
    env: {
      ...process.env,
      TIHAO_OVERNIGHT_LIVE: 'false',
      TIHAO_OVERNIGHT_OUTPUT: output,
      TIHAO_OVERNIGHT_FIXTURES: 'dha-mom-baby',
      TIHAO_OVERNIGHT_VARIANTS: 'baseline-live,result-first',
      TIHAO_OVERNIGHT_MAX_RUNS: '2',
      TIHAO_OVERNIGHT_FAIL_ON_GATES: 'false'
    }
  });

  if (result.status !== 0) {
    process.stderr.write(result.stderr || result.stdout || '');
    throw new Error('overnight quality smoke run failed');
  }

  const reviewCsv = path.join(output, 'manual-review-sample.csv');
  const aggregateReport = path.join(output, 'aggregate-report.md');
  assert(fs.existsSync(reviewCsv), 'manual-review-sample.csv should be written');
  assert(fs.existsSync(aggregateReport), 'aggregate-report.md should be written');

  const rows = readCsv(reviewCsv);
  for (const column of ['人工复核标签', '客户选择', '归因类型', '反馈原因']) {
    assert(rows.header.includes(column), `manual review sample should include ${column}`);
  }

  const metrics = buildMetrics({ inputPath: reviewCsv, rows });
  assert(metrics.headerOk, 'manual review sample should keep software table prefix header');
  assert(metrics.duplicateKeyCount === 0, 'manual review sample should have zero duplicate keys');
  assert(metrics.rankContinuous, 'manual review sample ranks should be continuous per brief');

  const report = fs.readFileSync(aggregateReport, 'utf8');
  assert(report.includes('归因类型'), 'aggregate report should instruct failure attribution');
  assert(report.includes('客户选择'), 'aggregate report should instruct customer decision labeling');
  assert(report.includes('参考账号不像'), 'aggregate report should list reference mismatch label');

  console.log(JSON.stringify({
    ok: true,
    output,
    reviewCsv,
    rows: metrics.total,
    duplicateKeyCount: metrics.duplicateKeyCount,
    rankContinuous: metrics.rankContinuous
  }, null, 2));
}

function assert(condition, message) {
  if (!condition) throw new Error(message);
}

main();