Comments20 pages, 1 figure, 9 tables. v2 adds Round 2: Russian-market coding agents (SourceCraft CLI, Koda CLI), Antigravity with Gemini 3.1 Pro / 3.5 Flash, and Codex CLI with GPT-5.6 on the same frozen task set, plus a tool-call contamination re-audit (network + disk layers). Data, full trajectories and harness: https://github.com/eugeneshilow/rubench
Form, Not Content? A Preregistered, Placebo-Controlled Evaluation of Learned Error-Conditioned Self-Repair Through Prompts and Weights in Frozen Small Code Models
形式而非内容?对冻结小代码模型中通过提示和权重进行的学习错误条件自修复的预注册、安慰剂对照评估
Mehmet Iscan
机构
*
PythaLab, Yıldız Technical University, Istanbul, Turkey(PythaLab,Yıldız技术大学,伊斯坦布尔,土耳其)
机构
*
Institute for Artificial Intelligence, Peking University(人工智能研究院,北京大学)
;
School of Intelligence Science and Technology, Peking University(智能科学与技术学院,北京大学)
;
College of Intelligence Science and Technology, National University of Defense Technology(智能科学与技术学院,国防科技大学)