把不同理由放上桌。 / Put more than one reason on the table.
开始讨论 / Play · 方法与权重 / Methods · 互动工作室 / Studio
AI 筛选简历、安排班次、整理福利咨询、标记作业、推荐信息时，你会先争取什么？这款轻量讨论游戏面向学生与初入职场的人：九个原创虚构情境，每个提供三种可以解释、也各有代价的政策回应。选择后查看收益、取舍和其他理由，最后得到本轮治理关注图谱与一个原创讨论角色。
When AI screens a CV, allocates shifts, organizes benefit enquiries, flags coursework, or recommends information, what would you ask for first? This discussion game is for students and early-career users: nine original fictional dilemmas, with three reasoned policy responses each. Compare benefits, costs, and alternatives, then view an attention map and an original discussion role for this round.
- 九个原创情境，涉及招聘、平台排班、公共服务、作业检测、推荐、职场监控、申诉、可及性与研究引文。/ Nine original cases on hiring, shifts, public services, coursework flags, recommendations, monitoring, appeals, accessibility, and research citations.
- 查看三种回应的收益、代价与替代理由；原文明确标注缺失证据。/ Compare three responses with benefits, tradeoffs, alternative reasons, and missing evidence.
- 回到任意案例改选或清除；进度与可选笔记自动保存在当前浏览器。/ Revisit, change, or clear choices; progress and optional reflections save locally.
- 查看机会可及性、透明、隐私、人工复核、责任纠错五类关注的份额。/ View attention shares for access, transparency, privacy, human review, and accountability.
- 得到与本轮选择关联的原创讨论角色；并列或接近的维度形成“多声部协商者”。/ Receive an original round-specific role; tied or close priorities use a Many-voice Convenor.
- 复制不含私人选择的讨论题，导出含选择与笔记的 Markdown 工作纸。/ Copy a debate prompt without personal answers, or export a Markdown worksheet that includes choices and notes.
- 下载 1080 × 1620 PNG 分享卡，只包含汇总维度和角色，不含具体选择或私人笔记。/ Download a 1080 × 1620 PNG with aggregate dimensions and a role; no raw answers or private notes.
- 中文默认，切换英语；64 KB 上限的 JSON 备份在完整校验后才替换当前记录。/ Chinese by default, with English switching; JSON backups up to 64 KB replace records only after validation.
示例分享卡 / Demo share card
以下图片由预设测试选择生成，仅用于展示下载格式。 / This image uses predetermined test choices solely to demonstrate the download format.
每个选项向五个维度分配 相同的 10 个关注点数，没有奖励分或惩罚分。将已选回应的点数相加后换算成合计 100% 的关注份额；改选会替换原分配，不会重复累加。五个维度与权重属于明确公开的编辑设计，不是经过验证的心理测量。
Every option allocates the same 10 attention points across five dimensions, with no reward or penalty points. Selected responses are summed and converted into shares totaling 100%; edits replace the previous allocation. Dimensions and weights are explicit editorial designs, not validated psychometrics.
这些结果描述本轮游戏选择，不评判人格、道德、政治立场或法律能力。角色名称是便于继续讨论的原创表达，没有正确角色或标准判决。题目中的组织、事实、政策回应均为虚构，不作现行法律、服务效果或具体争议的判断。
Results describe choices in this round. They assess no personality, morality, political stance, or legal competence. Role names are original discussion devices, with no correct role or official verdict. Organizations, facts, and policy responses are fictional; the app makes no current-law, service-effectiveness, or real-dispute judgment.
五个讨论维度借鉴 OECD AI Principles 的部分治理主题。案例、回应、权重和角色由本项目原创设计，不能被表述为 OECD 的完整原则目录、量表或认可产品。具体映射与来源核验说明见 方法文档。
The five discussion dimensions adapt selected themes in the OECD AI Principles. Cases, responses, weights, and roles are original project designs; they are not OECD's complete principle list, a validated scale, or an endorsed product. See the method note.
没有账户、后端、分析统计、外部字体或模型 API。选择和笔记保存在当前浏览器的 localStorage，不上传；跨设备继续需要导出和恢复 JSON。浏览器清理、无痕模式结束、存储拒绝或容量不足都可能影响保留。写入失败会持续提示当前变化仅在页面内，请关闭前导出备份。
No account, backend, analytics, external fonts, or model API. Choices and notes stay in this browser's localStorage; use JSON export/restore to move devices. Clearing data, ending private browsing, denied storage, or quota limits affect persistence. Failed writes show a persistent warning that changes remain in memory only; export before closing.
导入会先校验格式、案例版本、语言、进度、已知案例与选项、笔记长度和容量，再询问是否替换。无效导入不修改原记录。工作纸和 JSON 包含具体选择与私人笔记，请在分享前检查；PNG 与复制讨论题不包含这些内容。点击外部阅读链接仍会正常访问其网站，托管网站会收到普通页面请求。
Imports validate format, case version, language, progress, known case/option IDs, notes, and size before asking to replace records. Invalid imports leave existing data intact. Worksheets and JSON contain raw choices and reflections; review before sharing. PNG cards and copied prompts exclude them. External reading links visit their websites normally; hosting receives ordinary page requests.
纯静态 HTML/CSS/ES modules，无构建或应用依赖安装。使用本地 HTTP 服务，不直接双击 HTML：
Static HTML/CSS/ES modules, with no build or application dependency installation. Serve over HTTP:
python -m http.server 8000
node --check app.mjs
node --check core.mjs
node --check data.mjs
node --test tests/core.test.mjs
Node.js 20+；npm test 和 npm run check 提供等价命令。19 个核心测试覆盖所有 19,683 种完整选择组合、等额分配、百分比舍入、并列、改选、导入原子性、隐私导出和 Markdown 转义。GitHub Actions 重复运行检查。
Node.js 20+; npm test and npm run check are equivalent shortcuts. Nineteen core tests cover all 19,683 complete answer patterns, equal allocation, percentage rounding, ties, editing, atomic invalid imports, private-data exclusion, and Markdown escaping. GitHub Actions repeats the checks.
Optional browser verification uses a separately installed Playwright for development only:
npm install --no-save playwright
npx playwright install chromium
# Keep the HTTP service running; default test URL is http://127.0.0.1:8030/09-algorithm-jury/
node tests/browser-check.mjs
可设置 JURY_URL 指向本地服务、BROWSER_PATH 指向已安装 Chrome/Edge。开发检查验证 390 px 布局、键盘操作、刷新续玩、改选、无效/有效恢复、存储失败、讨论工作纸和 PNG 下载。页面运行不依赖 Playwright。
Set JURY_URL for a different local URL and BROWSER_PATH for an installed Chrome/Edge. Development checks cover 390 px layout, keyboard choices, resume, editing, invalid/valid restore, storage failure, worksheet and PNG downloads. Visitors require none of these development packages.