个人网站：网页改好了，下载的简历为什么还是另一版？

简历页接进个人网站后，同一份经历有了几个出口：中文网页、英文网页、下载的 PDF，以及浏览器打印。改好眼前这页，其他出口也可能继续提供旧内容。

这次我负责项目取舍、事实核对和文字删改，让 AI 编程工具调整页面、生成文件并补充检查。代码实现由 AI 完成。工作集中在 2026 年 10 月 30 至 31 日，没有访问转化或招聘结果数据。

先决定哪些经历能写进去

内容整理时，我把真实会务工作、个人原型和软件验证分开。会务实践的场次数不能写成系统服务了多少客户；合成数据上的流程通过，也不能写成社区已经使用。尚未核实的证书名称和项目日期继续留空。

我还要求把个人网站和 Lumi Lab 放进项目经历：网站本身也有设计、交互、双语内容、发布与维护工作。声音手帖作为其中的实验举例，不再重复占据一个主项目位置。随后又删去了顶部个人概述。这些修改需要同时反映到中英文网页与 PDF。

下载按钮指向一个独立文件

网页读取结构化的简历内容，PDF 则是另一个实际文件。两者可以引用同一组事实和项目顺序，但改网页不会重新生成那份文件。

因此，这次更新还包含重新生成单页 PDF、渲染检查文字与链接，再替换下载文件。验收时直接下载按钮指向的字节，与仓库里的 PDF 比较 SHA-256。这样才能发现“页面是新的，附件还是旧的”这种发布遗漏。

浏览器打印有自己的用途：它打印当前网页语言，隐藏导航和按钮，使用白底深色文字。英文页面下载的仍是中文投递版，界面也需要明确说明，避免读者下载后才发现语言不同。

深色主题下，还要检查切换的瞬间

另一个修复围绕主题切换后的文字对比度。只分别截取稳定的浅色、深色页面，可能漏掉切换时的短暂不可读状态。相关回归检查因此关注切换动作后的即时样式，打印样式则固定为适合纸面的颜色。

这里的协作分工很具体：我决定需要保留的内容、删去的段落与三个出口的一致性要求；AI 修改数据、样式和测试。最终核对仍需要回到渲染后的页面和真正下载的 PDF。

当前依据是版本记录、简历文档和浏览器回归记录。本轮写作未重新执行简历专项测试，也没有证明这些修改提高了求职成功率。可以查看网页简历与当前下载文件；后续更新仍要逐一核对这些出口。

Portfolio: keeping the web resume and its download in step

Adding a resume to my portfolio created several outputs: Chinese and English HTML, a downloadable PDF, and a browser print view. Updating the visible page could leave another output behind.

I chose the projects, checked the claims, and revised the copy. AI coding tools implemented the page, generated the file, and added checks. This account covers October 30–31, 2026; it has no recruiting-outcome or conversion data.

Decide what the evidence supports

I kept conference assignments, personal prototypes, and software validation distinct. An approximate event count did not become a count of customers using an AI system. Synthetic workflow tests did not become community adoption. Unverified certificates and dates stayed out.

I also added the portfolio and Lumi Lab as project work, with Sound Notebook as one experiment inside it. A later revision removed the introductory profile paragraph. These decisions had to reach both languages and the PDF.

Check the file behind the button

The HTML reads structured resume data. The PDF is a separate artifact. Sharing facts and project order does not make a page edit regenerate that file.

The update therefore included generating and rendering the one-page PDF, inspecting text and links, and replacing the download. Browser checks downloaded the actual response and compared its SHA-256 with the repository file. This checks whether the attachment caught up with the page.

Printing serves a different purpose: it prints the current page language, hides navigation and controls, and uses dark text on white. The English page still downloads a Chinese application PDF, so its button description needs to say that.

Inspect the transition between themes

A separate fix concerned text contrast immediately after a theme change. Stable light and dark screenshots can miss the transition itself. The regression checks therefore inspect styles after the toggle, while print styles keep their own fixed colors.

I set the content and consistency requirements; AI changed data, styles, and tests. We still needed the rendered page and downloaded artifact to judge the result.

This article reviews version history, documentation, and prior browser records. I did not rerun the resume suite while writing it, and there is no evidence of improved hiring outcomes. The web resume and download remain separate outputs to check during future edits.

不是每件事都需要一个标签。

探索线索