skill-case-get/scripts/tests/classify.test.mjs

// Copyright (c) 未来飞马
//
// This Source Code Form is subject to the terms of the Mozilla Public
// License, v. 2.0. If a copy of the MPL was not distributed with this
// file, You can obtain one at https://mozilla.org/MPL/2.0/.
//
// Trademark Notice:
// The MPL-2.0 license grants copyright permissions for source code only.
// It does NOT grant any rights to use trademarks including "未来飞马",
// "Harness Loop", "RSI", and associated slogan "让AI进化提前发生，让AI落地快人一步".
// Any use of these trademarks requires separate written permission.
/**
 * 拆解 / 打标 / 合规 / 入库契约 测试：`node --test`
 */

import test from 'node:test';
import assert from 'node:assert/strict';
import path from 'node:path';
import fs from 'node:fs';
import os from 'node:os';

import { loadDictionary, buildIdempotencyKey, normalizeMaterialAsset, imageUrlsFrom, sortAssets, uniq } from '../lib.mjs';
import {
  assessCompliance, classifyCase, deriveCaseFields, extractOutcome, findPrivacy,
  inferFitStatus, normalizeSchool, splitCaseAndMaterials,
} from '../classify.mjs';
import { runIngest, makeTinyPng } from '../case-intake.mjs';
import { makeFixtures } from './fixtures.mjs';

const dict = loadDictionary();

// ---------------------------------------------------------------------------
// 1. 输入识别 / 拆解
// ---------------------------------------------------------------------------

test('splitCaseAndMaterials：案例说明 与 素材 严格分开', () => {
  const split = splitCaseAndMaterials({
    textBundle: { paragraphs: ['案例说明段落'], notes: ['备注'], titles: [], transcript: '视频转写内容' },
    visionResults: [
      { role: 'material', ocrText: '素材图里的字', description: '成绩单截图' },
      { role: 'description', ocrText: '说明配图的字', description: '流程图' },
    ],
  });
  assert.ok(split.descriptionCorpus.includes('案例说明段落'));
  assert.ok(split.descriptionCorpus.includes('视频转写内容'));
  assert.ok(split.descriptionCorpus.includes('流程图'), 'description 角色归案例说明');
  assert.ok(split.materialCorpus.includes('成绩单截图'), 'material 角色归素材');
  assert.ok(!split.materialCorpus.includes('案例说明段落'));
});

// ---------------------------------------------------------------------------
// 2. 打标
// ---------------------------------------------------------------------------

test('学校别名归一：IC → 帝国理工，UCL → 伦敦大学学院；不误伤 MAGIC', () => {
  const a = normalizeSchool('IC 的计算机课，考前冲刺', dict);
  assert.equal(a.schoolCanonical, '帝国理工');
  assert.deepEqual(a.schoolAliases, ['IC']);
  assert.equal(a.country, '英国');

  const b = normalizeSchool('UCL 的学生找我辅导', dict);
  assert.equal(b.schoolCanonical, '伦敦大学学院');

  assert.equal(normalizeSchool('MAGIC 系统很好用', dict).schoolCanonical, '');
});

test('打标：亮点 / 场景 / 异议 / 静态属性 按字典与策略命中', () => {
  const { tags } = classifyCase({
    dict,
    groups: [],
    visionResults: [],
    textBundle: {
      paragraphs: ['帝国理工 计算机科学 本科二年级，期末考前冲刺，提分 18 分，家长很满意', '客户说太贵了，不放心老师'],
      notes: [], titles: [], transcript: '',
    },
    authorizationStatus: 'authorized',
  });

  assert.deepEqual(tags.schoolCanonical, ['帝国理工']);
  assert.deepEqual(tags.country, ['英国']);
  assert.ok(tags.major.includes('计算机科学'));
  assert.ok(tags.highlightTypes.includes('提分'));
  assert.ok(tags.highlightTypes.includes('家长认可'));
  assert.ok(tags.scenarioTags.includes('考前冲刺'));
  assert.ok(tags.scenarioTags.includes('期末考试'));
  assert.ok(tags.objectionTags.includes('贵'));
  assert.ok(tags.objectionTags.includes('不放心老师'));
});

test('适用跟进状态：按 docs/prd/dashboard/10 策略命中（异议处理中）', () => {
  const fit = inferFitStatus({
    highlightTypes: ['提分', '家长认可'],
    scenarioTags: ['考前冲刺'],
    objectionTags: ['贵'],
    subject: [],
  });
  assert.ok(fit.fitStatus.includes('异议处理中'));
  assert.match(fit.reason, /跟进状态/);
});

test('适用跟进状态：命中七个枚举之一（全部在契约枚举内）', () => {
  const allowed = new Set(['新进线', '挖需中', '方案推荐中', '异议处理中', '待决策', '沉默待跟进', '已成交']);
  for (const scenario of [
    { highlightTypes: ['补考通过'], scenarioTags: [], objectionTags: [], subject: [] },
    { highlightTypes: ['导师匹配'], scenarioTags: [], objectionTags: [], subject: [] },
    { highlightTypes: [], scenarioTags: ['考前冲刺'], objectionTags: [], subject: [] },
    { highlightTypes: ['首课体验'], scenarioTags: [], objectionTags: [], subject: ['高等数学'] },
  ]) {
    for (const status of inferFitStatus(scenario).fitStatus) {
      assert.ok(allowed.has(status), `非法跟进状态：${status}`);
    }
  }
});

// ---------------------------------------------------------------------------
// 3. 隐私 / 合规
// ---------------------------------------------------------------------------

test('findPrivacy：片段一律掩码，绝不返回完整 PII', () => {
  const findings = findPrivacy('联系 13812345678 邮箱 a.b@example.com 微信：student_abc');
  const fields = findings.map((f) => f.field);
  assert.ok(fields.includes('手机号'));
  assert.ok(fields.includes('邮箱'));
  assert.ok(fields.includes('微信号'));
  const phone = findings.find((f) => f.field === '手机号');
  assert.ok(!phone.snippet.includes('13812345678'), '手机号必须已掩码');
  assert.match(phone.snippet, /^138\*{4}5678$/);
});

test('合规：客诉 / 负面事件 → riskFlags，且禁止入库', () => {
  const result = assessCompliance({
    corpus: '客户投诉退费纠纷，要求赔偿，还发了黑猫',
    authorizationStatus: 'authorized',
    hasMaterials: true,
  });
  assert.ok(result.riskFlags.includes('COMPLAINT'));
  assert.equal(result.authorizationOk, true);
  assert.ok(result.complianceBlockers.some((b) => b.includes('客诉')));
  assert.match(result.reviewHint, /驳回|复盘/);
});

test('合规：无授权一律不入库（硬规则）', () => {
  for (const status of ['pending', 'denied', '']) {
    const result = assessCompliance({
      corpus: '帝国理工 提分案例，很好',
      authorizationStatus: status,
      hasMaterials: true,
    });
    assert.equal(result.authorizationOk, false, `status=${status} 必须判为无授权`);
    assert.ok(result.riskFlags.includes('UNAUTHORIZED'));
    assert.ok(result.complianceBlockers.some((b) => b.includes('无授权')));
  }
});

test('合规：无素材 / 跑题 也拦下来', () => {
  assert.ok(assessCompliance({ corpus: '帝国理工 提分', authorizationStatus: 'authorized', hasMaterials: false })
    .complianceBlockers.some((b) => b.includes('可用素材')));
  assert.ok(assessCompliance({ corpus: '今天天气不错，出门散步', authorizationStatus: 'authorized', hasMaterials: false })
    .riskFlags.includes('OFF_TOPIC'));
});

test('合规：正常好案例 → 可入库待审，无 blocker', () => {
  const result = assessCompliance({
    corpus: '帝国理工 计算机 考前冲刺，期末提分 18 分，家长很满意',
    authorizationStatus: 'authorized',
    hasMaterials: true,
  });
  assert.equal(result.authorizationOk, true);
  assert.equal(result.complianceBlockers.length, 0);
  assert.match(result.reviewHint, /待审/);
});

// ---------------------------------------------------------------------------
// 4. 案例字段抽取
// ---------------------------------------------------------------------------

test('extractOutcome：提分幅度 / 周期 / 结果 / 录取', () => {
  assert.equal(extractOutcome('期末提分 18 分').scoreGain, '18分');
  assert.equal(extractOutcome('从 52 分到 78 分').scoreRange, '52 → 78');
  assert.equal(extractOutcome('辅导 6 周').period, '6 周');
  assert.equal(extractOutcome('拿到了帝国理工的录取').admittedTo, '帝国理工');
  assert.equal(extractOutcome('最终 pass 了').result, 'pass');
});

test('deriveCaseFields：标题 / 摘要 / 使用建议 / 结果证据 都产出', () => {
  const fields = deriveCaseFields(
    'IC 计算机科学考前冲刺案例\n客户在读本科二年级，期末前两周找到我们，担心挂科。\n结果：期末提分 18 分，家长很满意。',
    [{ count: 3, preview: '' }],
    { productLine: ['大学课程'], schoolCanonical: ['帝国理工'], subject: ['高等数学'], stage: ['本科二年级'], scenarioTags: ['考前冲刺'], objectionTags: [] },
    { fitStatus: ['挖需中'] },
  );
  assert.ok(fields.title.length > 0);
  assert.ok(fields.summary.length > 0);
  assert.match(fields.usageSuggestion, /痛点|挂|时间/);
  assert.match(fields.targetCustomer, /帝国理工/);
  assert.match(fields.resultEvidence, /提分/);
});

// ---------------------------------------------------------------------------
// 5. materialAssets 契约（逐字对齐 SHARED-CONTRACT.md）
// ---------------------------------------------------------------------------

test('normalizeMaterialAsset：字段齐全、order 1 起、非法值报问题', () => {
  const { asset, problems } = normalizeMaterialAsset({ kind: 'image', url: 'https://x/a.jpg', order: 3 }, 0);
  assert.deepEqual(Object.keys(asset).sort(), [
    'description', 'groupId', 'kind', 'label', 'localPath', 'ocrText', 'order', 'role', 'url', 'usageSuggestion',
  ]);
  assert.equal(asset.order, 3);
  assert.equal(asset.role, 'material');
  assert.equal(problems.length, 0);

  const bad = normalizeMaterialAsset({ kind: 'pdf' }, 0);
  assert.ok(bad.problems.some((p) => p.includes('kind')));
  assert.ok(bad.problems.some((p) => p.includes('url')));
});

test('imageUrlsFrom：= kind=image 的 url，按 order 排序（兼容字段）', () => {
  const urls = imageUrlsFrom([
    { order: 2, kind: 'image', url: 'https://x/b.jpg' },
    { order: 1, kind: 'image', url: 'https://x/a.jpg' },
    { order: 3, kind: 'video', url: 'https://x/c.mp4' },
  ]);
  assert.deepEqual(urls, ['https://x/a.jpg', 'https://x/b.jpg']);
});

test('buildIdempotencyKey：同内容同键，内容变则变（tenantId + idempotencyKey 幂等）', () => {
  const base = { tenantId: 'lumi-demo', sourceType: 'image_group', sourceRef: 'wx-2026-10', title: 'IC 提分', materialAssets: [{ order: 1, kind: 'image', url: 'u1' }] };
  const k1 = buildIdempotencyKey(base);
  const k2 = buildIdempotencyKey({ ...base });
  const k3 = buildIdempotencyKey({ ...base, title: '别的标题' });
  assert.equal(k1, k2);
  assert.notEqual(k1, k3);
  assert.match(k1, /^caseget-[0-9a-f]{16}$/);
  assert.equal(buildIdempotencyKey({ ...base, explicit: 'manual-key' }), 'manual-key');
});

test('uniq：多参数/嵌套都支持，并丢空值', () => {
  assert.deepEqual(uniq(['a'], ['b', 'a'], [], ['c']), ['a', 'b', 'c']);
  assert.deepEqual(uniq([[1, 2], [2, 3]]), [1, 2, 3]);
  assert.deepEqual(uniq([null, undefined, '', 'x']), ['x']);
});

test('sortAssets：按 order 升序且稳定', () => {
  const sorted = sortAssets([{ order: 2, label: 'b' }, { order: 1, label: 'a' }, { order: 2, label: 'c' }]);
  assert.deepEqual(sorted.map((a) => a.label), ['a', 'b', 'c']);
});

// ---------------------------------------------------------------------------
// 6. 端到端：夹具 → ingest → 案例包
// ---------------------------------------------------------------------------

test('端到端 ingest：多图(9)+docx+pptx → 案例包，字段对齐共享契约，不自动提交', async (t) => {
  const workRoot = fs.mkdtempSync(path.join(os.tmpdir(), 'caseget-e2e-'));
  t.after(() => fs.rmSync(workRoot, { recursive: true, force: true }));

  const fixtures = makeFixtures(path.join(workRoot, 'fixtures'));
  const outDir = path.join(workRoot, 'out');

  const report = await runIngest({
    inputs: [...fixtures.images, fixtures.docx, fixtures.pptx],
    outDir,
    sourceRef: 'e2e-fixture',
    authorizationStatus: 'authorized',
    assetBaseUrl: 'https://cdn.example.invalid/cases/e2e',
    // 单测不联网：关掉真上传，专门验证 --asset-base-url 派生的旧行为
    upload: false,
    skipVision: true, // 不调模型：OCR 留空，其余链路全跑
    noLearn: true,
  });

  const pkg = report.casePackage;

  // 输入识别：多图 + 文档 + PPT → mixed
  assert.equal(pkg.sourceType, 'mixed');

  // 素材：9 张九宫格 + 1 张异批 + docx 内嵌 1 + pptx 内嵌 1
  assert.equal(pkg.materialAssets.length, 12);
  assert.deepEqual(pkg.materialAssets.map((a) => a.order), Array.from({ length: 12 }, (_, i) => i + 1));

  // 四个分组：九宫格 / 异批单图 / docx 内嵌图 / pptx 内嵌图
  // （两份文档的内嵌图不会互相并组——sourceKey 不同）
  assert.equal(pkg.groupInfoList.length, 4);
  assert.equal(pkg.groupInfoList.filter((g) => g.count === 1).length, 3);
  assert.ok(pkg.materialAssets.some((a) => a.description.includes('文档') || a.label.includes('内嵌')), '应包含文档内嵌图素材');

  // 九宫格分组
  const gridGroup = pkg.groupInfoList.find((g) => g.count === 9);
  assert.ok(gridGroup, '应识别出 9 张的九宫格组');
  assert.equal(gridGroup.layout, '3x3');
  assert.equal(gridGroup.orderingRule, 'left-to-right,top-to-bottom');
  assert.equal(pkg.groupInfo.layout, '3x3');

  // 契约必填字段
  for (const key of ['sourceType', 'idempotencyKey', 'title', 'summary', 'materialAssets', 'groupInfo', 'tags', 'highlightTypes', 'scenarioTags', 'objectionTags', 'riskFlags', 'authorizationStatus']) {
    assert.ok(key in pkg, `案例包缺少契约字段 ${key}`);
  }

  // 落库预期：进待审清单，不进公共素材库
  assert.deepEqual(pkg.expected, { reviewStatus: 'pending', readyForUse: false });

  // 文档抽取的文本进入打标语料
  assert.ok(pkg.schoolCanonical === '帝国理工', `学校归一失败：${pkg.schoolCanonical}`);
  assert.ok(pkg.highlightTypes.includes('提分'));
  assert.ok(pkg.imageUrls.length >= 10);
  assert.ok(pkg.imageUrls.every((u) => u.startsWith('https://cdn.example.invalid/cases/e2e/')), 'url 应按 --asset-base-url 派生');
  assert.equal(pkg.hasImage, true);
  assert.equal(pkg.materialAssets.filter((a) => !a.url).length, 0, '提供了 base url 后不应再有缺 url 的素材');

  // 产物落盘
  for (const file of Object.values(report.artifacts)) {
    assert.ok(fs.existsSync(file), `产物缺失：${file}`);
  }
  const pkgOnDisk = JSON.parse(fs.readFileSync(report.artifacts.casePackage, 'utf8'));
  assert.equal(pkgOnDisk.idempotencyKey, pkg.idempotencyKey);

  // 报告提到九宫格顺序与待审
  const md = fs.readFileSync(report.artifacts.report, 'utf8');
  assert.match(md, /materialAssets/);
  assert.match(md, /3x3/);
});

test('端到端 ingest：未授权时不产出可提交案例包（canSubmit=false）', async (t) => {
  const workRoot = fs.mkdtempSync(path.join(os.tmpdir(), 'caseget-unauth-'));
  t.after(() => fs.rmSync(workRoot, { recursive: true, force: true }));

  const fixtures = makeFixtures(path.join(workRoot, 'fixtures'));
  const report = await runIngest({
    inputs: [fixtures.images[0]],
    outDir: path.join(workRoot, 'out'),
    authorizationStatus: 'pending',
    upload: false, // 单测不联网
    skipVision: true,
    noLearn: true,
  });

  assert.equal(report.canSubmit, false);
  assert.ok(report.casePackage.riskFlags.includes('UNAUTHORIZED'));
  assert.ok(report.casePackage.complianceBlockers.length > 0);
});