一个判定不够：学习型世界模型及其测试框架的回归测试
原标题：One Verdict Is Not Enough: Regression Testing Learned World Models and Their Test Harnesses
AI 摘要
论文提出 TetraState 回归测试协议，针对学习型世界模型在模型版本、预处理和测试框架变化时的回归问题。该协议将每个测试在四种模型原生配置下运行，用结果类别（一致通过、配置相关通过、配置相关失败、一致失败）替代单一通过/失败判定，并将测试框架纳入被测系统版本管理。在 IntPhys 2 的 506 个测试及 Cosmos 3 Nano 与 V-JEPA 2 两个系统上验证发现，V-JEPA 2 测试框架修订仅使模态准确率变化 0.2 和 2.6 个百分点，却分别改变 11 和 214 个测试标签，说明准确率会掩盖框架缺陷。
正文节选
One Verdict Is Not Enough: Regression Testing Learned World Models and Their Test Harnesses Abstract Learned world models ship as software components whose model versions, preprocessing, and test harnesses change between releases. Work on model updates already counts per-sample regressions, or negative flips, that an accuracy check hides. Two problems remain: a test can pass under one configuration and fail under another, and a harness fault can produce erroneous scores and wrong verdicts withou