A 2017 English guide grew into a book whose build refuses any Node but 24

The six-step loop, and the three labels the book puts on every claim

The README opens with one loop and asks readers to run it repeatedly: find a problem, learn actively, work with AI, finish a real task, save the evidence, review and transfer. Nothing in this repository runs that loop for you. What the book does is sort every statement into one of three kinds. Research conclusions must name a source and say how far the evidence reaches. Personal experience keeps the story without being sold as a general rule. Hypotheses may enter the discussion, but only to be tested by the next round of action.

The author holds himself to the same rule. He is Han Xiankai, writing as Lipu, states that he is chairman of 中国词元云计算有限公司, says his commercial relationships are written in the open, and accepts claims about ability, product and income only when work, users, cost and time leave a trace behind. That disclosure is what docs/projects.md is for: the reading path describes it as disclosing relationships, purpose, update time and non-sponsorship.

Reading group 04 is an AI gateway and a Telegram channel, not a chapter

The front page sorts the book into four numbered groups. Groups 01 to 03 are reading paths, and every link is a Markdown file under docs/threads/ or docs/templates/. Group 04 is a different kind of entry. It holds biezou.com, publicly positioned as a unified AI API gateway and management panel, and the OpenHuge_ai Telegram channel, described as a way to find AI resources.

The README hedges this group on purpose. It says an optional third-party entry point is not an endorsement of security, quality or compliance by the book or by the author, and it tells you to check terms, price, privacy and availability before you use either one. Both are plain external anchors carrying target="_blank" and rel="noopener noreferrer", and nothing in the repository pins or mirrors them, so what they point at can change without a commit. One detail to keep in view: the repository's homepage field is https://biezou.com, the same gateway the README labels as optional.

Seven parts on disk, plus a glossary that ends in a next step

The spine is a table of parts, each with one core question and a file list. Part 0 holds a reading guide and a prologue. Part 1 opens the input and carries English from a CEFR self-test through vocabulary, grammar, listening, reading, speaking, writing and job-search English. Part 2 puts ability, career, relationships, failure and recovery back together. Part 3 covers learning with AI, attention, work and evidence. Part 4 returns to the body, family and daily rhythm, including how a parent protects a young learner's own agency instead of monitoring them. Part 5 is a ninety-day plan plus a chapter that treats the book as its own proof. Part 6 is a single afterword.

Two navigation files sit outside the parts. docs/reference/glossary.md is ordered as definition, then evidence, then next step, so no term dead-ends. docs/templates/toolkit.md indexes the worksheets by the problem in front of you. The older shape of the project is still in legacy/ at the root, and the difference in approach is the argument of the book: English used to be the whole map, and is now one branch feeding the same loop as everything else.

Building the site: Node 24 only, then six checks in a chain

There is no install command anywhere in the README. What it offers instead is the finished book: four files under docs/public/downloads/, Chinese and English, EPUB and PDF, plus a link back to this repository labelled as the source and the errata. The npm side exists to render the text and to produce those downloads. package.json is named life-level-up-guide, marked "private": true at version 1.0.0, and it pins one major version of Node:

"engines": {
  "node": ">=24 <25"
},
"docs:dev": "vitepress dev docs",
"docs:build": "vitepress build docs && npm run check:bundle",
"docs:preview": "vitepress preview docs"

The dev server is VitePress 1.6.4, alongside markdownlint-cli2 0.23.2, sharp 0.35.3 and @playwright/test 1.62.1. A first-time contributor needs three of the scripts:

npm run docs:dev
npm run check
npm run book:build

`check` is not a test. It is a chain, and the order matters because each step reads files an earlier step writes:

"check": "npm run check:navigation && npm run check:readme && npm run check:content && npm run check:format && npm run book:check && npm run book:pdf:check"

Inside it, check:navigation runs sync-navigation.mjs, sync-word-lists.mjs and sync-public-assets.mjs, each with --check, so a drifted sidebar fails the build instead of shipping. check:format hands every Markdown file to markdownlint-cli2. When generated files have drifted, sync:content rewrites navigation, word lists, public assets, the README and the EPUB in one pass. Plan for one gap: requirements-fonts.txt and requirements-pdf.txt sit at the root, and the README does not say what they ask for, so the PDF side has a dependency the npm scripts never mention.

No releases, no version to pin, and three license files

Three facts decide what you can rely on. There are no GitHub releases, and package.json is private at 1.0.0, so there is no artifact to pin and nothing to roll back to. The default branch is master, the last push was on 2026-09-20, and the README front matter carries updated: 2026-09-02 while calling the book a manuscript still being written.

For a reader that is fine. For anyone mirroring it, translating it or hanging a curriculum on a chapter URL, the cost is concrete: the text under a stable path can change with no version marker, and the only guard is the script chain you have to run yourself. Note that `check` never calls test:smoke, so a green chain does not tell you the built pages still open. playwright.config.mjs and that smoke script are a separate step nobody wires in for you.

On licensing, the body text is published under CC BY-NC 4.0. The root carries LICENSE.md, LICENSE-CONTENT.md, LICENSE-CODE.md and ATTRIBUTIONS.md, and GitHub's license field for the repository reads NOASSERTION, so the repository page itself will not tell you which file covers the part you want to reuse.

The evidence this book wants is evidence only you can produce

The method is to write a baseline before starting and retest after. docs/templates/learning-state.md records what you already know, what you do not, and what result would count as done. The grammar path and the family path both use a fourteen-day retest, the second one to replace monitoring and doing a young learner's work for them. Part 5 asks for ninety days, with docs/templates/90-day-cycle.md as the sheet to copy.

Those are Markdown files you fill in by hand. No script reads them, scores them or reminds you. The receipt the book offers, docs/templates/reader-field-note.md, is a form a reader sends back to the author. So here is the honest limit: progress in this project is self-reported, and a reader who needs evidence a third party can check will not find any in this repository. The book says close to as much itself when it states that it does not promise a tool will change your life, and that what stays scarce is knowing which question to press and owning the final decision.

The JavaScript here only renders and checks the book

VitePress builds the site, markdownlint-cli2 lints the Markdown, sharp handles images, Playwright smoke-tests pages, and the book: scripts drive build-epub.mjs and run-pdf-build.mjs. That is the whole of the code.

If you came for the AI material in part 3, the chapter on AI projects and resource-layer business walks requirements, prototypes, code, tests, model access, governance, enterprise delivery and commercial validation, and it is prose. There is no model, no dataset, no starter repository and no service to start. The English path has the same shape: a CEFR self-test and a fourteen-day retest on one high-impact structure are exercises for a person, not commands you can invoke. Treat the repository as a book with a small publishing toolchain attached, and if your decision is about a library, a model or an agent you intend to run in production, the first thing to check is which of LICENSE-CONTENT.md and LICENSE-CODE.md covers the part you want to take.

Editorial conclusion

Read it if you want a structured plan and are willing to keep the paper trail yourself. Skip it if you need a runnable tool or a result somebody else can verify. Before reusing any of it, read LICENSE-CONTENT.md and LICENSE-CODE.md, because the README puts the text under CC BY-NC 4.0, and if you fork the site, run npm run check on Node 24 first to see whether your copy still builds.

Frequently asked questions

Official sources

1. byoungd/up on GitHub
2. Issues
3. Project website
4. README