A
数据处理与采集 Data processing and collection
把网页、CSV、XLSX 与数据库数据转化为来源清晰、可恢复、可复现的结果。 Turn web, CSV, XLSX, and database inputs into traceable, resumable, and reproducible results.
开源工程实践 / 2026 Open-source engineering / 2026
把遗留脚本变成
可验证、可迁移
的开源工具。
Turning legacy scripts into
verifiable, portable
open-source
tools.
我关注兼容迁移、数据工具和可访问前端。每个公开结论都应有代码、测试或发布记录作为证据。 I work on compatible migrations, data tools, and accessible front ends. Every public claim should be backed by code, tests, or a release.
-
01
兼容是契约 Compatibility is a contract
保留有效输入、输出和调用路径。 Preserve valid inputs, outputs, and invocation paths.
-
02
验证优先于承诺 Evidence before promises
Windows、Linux、浏览器和真实发布共同验收。 Validate on Windows, Linux, browsers, and real releases.
-
03
边界也是功能 Boundaries are features
明确不采集什么、不推断什么、不自动做什么。 State what is not collected, inferred, or automated.
01 / 03
可由公开仓库证明 Backed by public repositories
能力，不是关键词。 Capabilities, not keywords.
B
兼容性与迁移 Compatibility and migration
先冻结有效行为，再替换语言、依赖与内部结构，让旧用户能够无缝迁移。 Freeze valid behavior before replacing languages, dependencies, and internals so existing users can migrate safely.
C
可访问前端与验证 Accessible front ends and verification
用语义化 HTML、渐进增强和跨平台测试构建无需框架即可长期维护的页面。 Build durable, framework-free pages with semantic HTML, progressive enhancement, and cross-platform tests.
02 / 03
精选公开项目 Selected public work
五个项目，五种边界。 Five projects, five explicit boundaries.
状态来自 2026 年 7 月完成的仓库验证。深入命令、限制和迁移说明以各项目 README 为准。 Statuses reflect repository verification completed in July 2026. Each project README remains the source of truth for commands, limits, and migrations.
-
Python / Data collection
ttjjCrawler
面向中国公募基金公开数据的保守型、多来源、可恢复采集器。来源始终分开保存，不自动猜测哪个来源“正确”。 A conservative, multi-source, resumable collector for public Chinese mutual-fund data. Sources remain separate; the tool never guesses which one is “correct.”
-
Python / Privacy
ReadWechatMessage
完全离线的一对一聊天记录统计工具，兼容旧 MATLAB 输入列位与核心输出名称，同时不连接微信、不解密数据库、不上传内容。 A fully offline one-to-one chat statistics tool that preserves legacy MATLAB column positions and output names without connecting to WeChat, decrypting databases, or uploading content.
-
Python / OCR + Simulation
yuanshen
可复现、可批处理、兼容旧输入的五星圣遗物副词条评分实验，支持结构化 JSON 与可选 OCR，并明确限定为非官方研究工具。 A reproducible, batchable artifact sub-stat scoring experiment with legacy JSON compatibility and optional OCR, explicitly scoped as an unofficial research tool.
-
HTML / CSS / Accessibility
html_sample
无需框架或运行时构建的静态学习示例集，覆盖语义化 HTML、响应式 CSS、无障碍、表单交互和自动化浏览器测试。 A framework-free static learning gallery covering semantic HTML, responsive CSS, accessibility, form behavior, and automated browser testing.
-
Python / Time series / Legacy
ProcessSqlData-TS
对 Python 2 时序研究原型进行风险隔离、行为记录和迁移规划。仓库明确禁止把遗留脚本直接当作生产任务运行。 A safety and migration framework around a Python 2 time-series research prototype. The repository explicitly warns against running the legacy script as a production job.
03 / 03
工作方式 How I work
先定义边界，再写实现。 Define the boundary before writing the implementation.
冻结有效行为 Freeze valid behavior
把输入、输出、错误和调用路径写成兼容契约。 Turn inputs, outputs, errors, and invocation paths into a compatibility contract.
拒绝静默猜测 Refuse silent guesses
来源冲突、无效输入和结构变化应该可见地失败。 Source conflicts, invalid inputs, and structural changes should fail visibly.
让证据随代码发布 Ship evidence with the code
测试、文档、Release 和部署结果共同定义“完成”。 Tests, documentation, releases, and deployment results define “done” together.
本页的隐私承诺 This page's privacy promise
不使用 Cookie，不接入统计，不加载第三方脚本，也不收集任何访客输入。 No cookies, analytics, third-party scripts, or visitor input collection.