Repository navigation
docs(agent-harness): Agent4Sci（Automating ML for Science）精读与通俗拆解入库（181，I-241） - #1219
Merged
Merged
Conversation
…1，I-241）——防作弊三阶段协议 + 增益成分表 + 三技能干预，附原型与 4 张论文原图; - 新增 181 精读（arXiv:2610.28763v1，2026-10-20，Anthropic Fellows Program，arXiv perpetual non-exclusive）：20 个已发表科学 ML 基准的长时程爬坡测试台；信源经官方 citation 元数据双通道核验（webReader 首抓为幽灵内容已弃用，I-241）。 - 核心拆解：四类评测作弊与标准化+三阶段协议（封存测试集/独立评测算分/轨迹审计，120 对三家族审计全一致）；Δ% 方向归一与增益成分表（工程型 σ=1.5 vs 洞见型 σ=19.6，HPO 消融只回收小部分）；三缺陷三技能近翻倍（10.4→20.1，PFDelta 64.4→91.0 物理洞见）与三天长跑反例（洞见门控、过拟 −25%）；五规律三争议五批判边界。 - 配套纯标准库原型 agent4sci_lab.py（15 断言 selftest + E1–E5 破坏实验实测退化）；4 张论文原图入 source-figures（fig2 等比降采样过 1280KiB 门，RGB 无透明）。 - 登记面：research readme、knowledge-map、source-figures README、CHANGELOG、issue I-241。 - 验收：源稿对账 34 项全绿（含 4 返工复判）+ 增量 Checker 8 行；盲评两判官出闸；外行四测四项出闸（LLM 代理作答）；字数门 5000 整（中文+英文单词）。 🤖 Generated with [Claude Code](https://github.com), [CodeX](https://openai.com), [Gemini](https://github.com/apps/gemini-code-assist) Co-Authored-By: Aurelius Huang<threefish.ai@gmail.com>
…/../assets/→../../assets/、2 处 lab 链接 ./agent-harness/assets/→./assets/，7 处对齐兄弟文档 pattern（I-241 增量）; 🤖 Generated with [Claude Code](https://github.com/claude), [CodeX](https://openai.com), [Gemini](https://github.com/apps/gemini-code-assist) Co-Authored-By: Aurelius Huang<threefish.ai@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
摘要
Automating ML for Science: Can Frontier Agents Climb Scientific Hills in the Wild?（arXiv:2610.28763v1，2026-10-20，Ming Zhong / Stacy Li / Nicholas Carlini / Matthew Jagielski，Anthropic Fellows Program）首读入库：agent-harness 181。论文把「前沿编码 agent 能不能做科学 ML 研究」变成可防作弊的测量问题——20 个已发表科学基准、8×H200/12h 预算内测全自动爬坡增益及其构成。
核心拆解
交付物
docs/research/agent-harness/181-agent4sci-scientific-hills.mddocs/research/agent-harness/assets/agent4sci_lab.pydocs/assets/source-figures/agent-harness/登记面：research readme、knowledge-map、source-figures README、CHANGELOG、issue I-241。
验收
🤖 Generated with Claude Code