Harness-IF：用执行证据评估编码Agent规则有效性
If you write rules in an AGENTS.md, this one is worth your time.
If you write rules in an AGENTS.md, this one is worth your time.
如果你在AGENTS.md中编写规则，那么这条规则值得你花时间阅读。
When a coding agent follows your rule, it may have been going to do that anyway.
当编码代理遵循你的规则时，它可能本来就会这样做。
Harness-IF separates the two by scoring 256 rules one at a time from execution evidence, then re-running every task with the rule withheld across nine probe builds to find which rules actually oppose the model's defaults.
Harness-IF通过从执行证据中逐一评分256条规则，然后在九个探针构建中重新运行每个任务并省略该规则，以找出哪些规则实际上与模型的默认行为相悖，从而将两者区分开来。
Across 12 frontier models, raw accuracy runs 72.1 to 85.9%, and Against-Prior Accuracy runs 66.1 to 78.6%. Every model gets worse once coincidence is stripped out, by 3.6 to 7.4 points.
在12个前沿模型中，原始准确率在72.1%到85.9%之间，而对抗先验准确率在66.1%到78.6%之间。一旦去除巧合因素，每个模型的性能都会下降3.6到7.4个百分点。
One finding worth flagging. Precedence does not follow prompt depth. System prompts, project files, and user instructions all outrank tool and skill descriptions.
一个值得注意的发现是：优先级并不遵循提示深度。系统提示、项目文件和用户指令的优先级都高于工具和技能描述。
Paper: https://arxiv.org/abs/2608.11727
论文：https://arxiv.org/abs/2608.11727
Track more trending AI papers in our academy: https://academy.dair.ai/
在我们的学院中追踪更多热门AI论文：https://academy.dair.ai/
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力