test.jsonl
2.62 KB
{"id":"test-01","source":"real","provenance":"data/companies/e2e-demo/runs/run_9168bd24f46c/nodes/rn_6bcda7e1fd6a/contract.json（attempt-1修复后的最终版本，attempt-1.json的addressed逐条对应本文件字段）","excerpt":"verifiable_assertions A-1..A-14独立可复核断言 + requires_browser_measurement MB-1..MB-6要求实测数值对 + style_budget量化风格上限 + content_fidelity_floor占位退化兜底 + error_mapping九条覆盖E-1~E-9 + 六项data_shapes均含where/shape","scores":{"D1":5,"D2":5,"D3":5,"D4":5,"D5":5},"weighted_sum":100,"final_score":100,"cap_triggered":null,"level":"优秀","label_note":"train-01的修复后版本，同一案例的另一半：验证benchmark对'同一份契约改好之后'能正确从45分（不合格）跳到100分（优秀），而不是对同一案例系统性偏向某个分数。"}
{"id":"test-02","source":"constructed","provenance":"构造样本，非真实案例，混合中等质量","excerpt":"接口签名给了主要键但1处缺省值未写明（D1=3）；验收判据里1条'响应速度良好'是结论句、其余判据可数（D2=3）；data_shapes字段清单完整但1个字段命名与PRD术语脱节未标注映射（D3=4）；1条PRD边界场景（并发提交）在error_mapping找不到对应（D4=3）；无越界内容（D5=5）","scores":{"D1":3,"D2":3,"D3":4,"D4":3,"D5":5},"weighted_sum":70,"final_score":65,"cap_triggered":"D2=3 → 上限65（加权和70>65，取65）","level":"合格","label_note":"边界样本：D2=3恰好命中封顶规则的中间档，验证'合格但有薄弱维度'区间（65-79）在触发封顶后被压到区间下限65，而非直接落在70——测试封顶规则对D2=3这一档是否会被正确应用而非只在D2<=2时生效。"}
{"id":"test-03","source":"constructed","provenance":"构造样本，非真实案例，D2严重违规的另一种表现形式（与train-01/04/05均不同的具体缺陷）","excerpt":"整份契约里，唯一的验收相关判据是正文一句话'页面视觉效果达到设计要求即可，由验收人员目测判断'——既非可数断言也非实测数值，也没有声明谁来判断'达到'的边界；接口签名与数据形态本身写得完整","scores":{"D1":4,"D2":1,"D3":4,"D4":3,"D5":5},"weighted_sum":62,"final_score":45,"cap_triggered":"D2=1 → 上限45","level":"不合格","label_note":"D2=1的另一种触发形式：不是'自证式脚本'或'结论句代替实测'这类曾在train里出现的具体模式，而是'纯主观目测无任何可复核判据'，用于验证D2评分标准能泛化到未见过的具体缺陷表现，而不是只认识train里列举过的四种范例写法。"}