NiceEval 是什么

// evals/weather-tool.eval.ts
import { defineEval, defineJudge } from "niceeval";
import { includes } from "niceeval/expect";

const groundedAnswer = defineJudge({
  name: "grounded-weather-answer",
  rubric: "根据 question 评价 answer 是否给出具体天气信息，而不是拒绝或含糊作答。",
});

export default defineEval({
  description: "问天气时调用工具，并基于工具结果作答",
  judge: groundedAnswer,

  async test(t) {
    const turn = await t.send("北京今天天气怎么样？");
    turn.succeeded().label("Turn 成功");

    t.calledTool("get_weather").label("调用天气工具");
    t.check(turn.message, includes("天气")).label("说明天气");

    const second = await t.send("那上海明天呢？");
    t.check(second.message, includes("上海")).label("第二轮接得上上下文");

    t.judge({ question: turn.input, answer: turn.message }, groundedAnswer).gate(0.7)
      .label("回答有依据");
  },
});

// evals/social-journey.eval.ts —— 被测对象是一个 AI 驱动的社交应用
export default x.defineEval({
  description: "发帖、回复后，AI 角色作出回应，社交世界保持自洽",
  async test(t) {
    const initial = await t.visitDiscoveryPage();
    const post = await t.post({ intent: "邀请大家今晚一起拍摄城市夜景", withImage: true });
    t.check(post, authoredPost(initial.viewerId)).label("帖子属于当前玩家");

    const reply = await t.reply({ postId: post.id, intent: "补充集合地点在河边步道入口" });
    const responses = await t.waitForReplies(reply.id);
    t.check(responses.length, greaterThan(0)).label("AI 角色作出回应");
    t.check(worldMaterial(await t.refreshFeed()), coherentSocialWorld()).label("刷新后社交关系仍完整");
  },
});