Practical notes on research engineering, scientific software, reproducibility, and AI-assisted technical work.
This repository grew out of a small collection of MATLAB utilities for psychology and behavioral-research workflows. The original material is preserved under legacy-matlab-tools/. New material focuses on reusable engineering lessons learned from real research and software work, rewritten so that it stands on its own without exposing private project details.
Open the demo index. It has three status sections:
- 使用中 / In use — current learning, revision, and everyday workflows.
- 尝试中 / Trying — methods under exploration, including AI执行效率检查.
- 暂停中 / On hold — paused prototypes, currently GrapePaper, with existing links preserved.
Status describes the author's current use, not a quality certification. Public examples remain generic or independently fictional.
continuous-learning/ai-use.html— 使用中下的“持续积累”:AI 使用经验的单一入口,集中记录模型连通边界、Dots 云端分工、对话交接、自查与常见问题。continuous-learning/writing-feedback.html— 接续现有写作入口:公开文章收集、导师反馈提纯与写作检查;只保留通用方法。notion-learning-quests.html— 中文经验网页:Notion 学习闯关的现成方案、虚构三关示意与待验证的搭建检查清单;下载后可直接在浏览器打开。agentic-engineering/— reviewing and governing AI-assisted implementation work.research-engineering/— reproducibility, model-fitting, evidence, and scientific-software practice.legacy-matlab-tools/— the original MATLAB scripts and demos kept for continuity.PUBLICATION_POLICY.md— privacy and provenance gate for all public notes.
These notes favor methods that can be checked over claims that merely sound convincing. In particular:
- bind reviews to an exact revision;
- review evidence, not confidence;
- separate implementation success from scientific validity;
- distinguish
PASS,FAIL, andUNVERIFIED; - keep scope changes explicit;
- preserve provenance where it matters;
- publish only project-agnostic lessons.
The repository is intentionally incremental. Notes are added when a recurring lesson is useful enough to keep, not to satisfy a posting schedule.
See LICENSE. Legacy material retains its existing history and attribution context.