"Tutoring centers could charge parents over ¥1,000 for this": a Chinese textbook knowledge graph built by a crypto-focused X account, 19 days, 6700 stars

On the afternoon of October 18, 2026, an X account in Chinese called "风清扬" created a GitHub repo. 24 minutes later it pinned a tweet that opened with "Let's put it this way: if the repo below ever reaches tutoring centers, they could charge parents over ¥1,000 a year." That tweet ended up with 227k views. By November 6, the repo, cn-math-taxonomy, had 6755 stars and 1758 forks. Below: what it is, where the stars came from, why comparable projects with more data only have a couple hundred stars, and how to build a project that can earn stars the same way.

What it is

A prerequisite map of Chinese primary and secondary school knowledge: each knowledge point is tagged with grade, volume, and chapter from People's Education Press / unified-edition textbooks. Edges between points mean "you must learn A before B," and each point has questions attached.

What actually gets shared is the web layer on top (math.pansgo.com): the kid starts from a target knowledge point, and every wrong answer traces back along the prerequisite chain until it hits the exact concept they never learned, then gives a step-by-step fix route. Get "completing the square" wrong and it traces back to "perfect square formula" from Grade 8. The diagnosis can be exported as a vertical share image with a QR code. Free, no sign-up; the data is under CC BY-NC 4.0, free for personal and educational use, commercial use needs permission from the author.

The first version covered only junior-high math: 333 knowledge points, 616 prerequisite edges, 1407 questions. After that, something new came out almost every day: primary school, high school, English, physics, Chinese, science, moral education and law, history, geography, biology, chemistry. By November there were 6 133 commits and 9 subjects. A Zhihu article introducing it said "five days." Judging by the pace, much of the content was likely AI-generated (inference), and the README itself admits that the high school and English questions "have not been blind-reviewed" and that many catalogs are "pending verification."

How the stars came

GitHub star count (11-06):

Three things came together.

1. It rode someone else's wave of attention. US company Marble released os-taxonomy, a knowledge graph of English and American primary school curricula, in July. In mid-October it was suddenly reshared all at once in the Chinese AI community on X, gaining 1200 stars over three days from 10-17 to 10-20. @jinglian 10-18 had just posted about it, and 10-19 followed with this "domestic version". @aigclink 10-20 said: "The project I shared a couple of days ago, a Chinese curriculum version has already been built." The Chinese community had just been introduced to "knowledge graphs", and people were asking "what about the Chinese textbook version?" This repo was right there. Its README also states that the data format references os-taxonomy.

2. The author already had traffic. "风清扬" (@hassanyehia) registered 2011 years ago, has 16.8k followers, and describes themselves as "an independent observer of the crypto market | AI Agent player," with a "#蓝鸟会成员" (Bluebird Club member) tag (a mutual-follow and promotion circle on Chinese X; how many of the stars came from mutual promotion is unknown). The tweet from the day the repo was created had a cold-start of 227k views. When the blogger reposted it the next day, the repo already had momentum worth citing.

3. The copy was written for parents, not programmers. "Tutoring centers could charge over ¥1,000," "diagnose exactly which knowledge point your child is stuck on": a developer dataset was framed as a tool for parents to save money. The Zhihu article was titled "I bet every tutoring center hopes Chinese parents never see this GitHub project," and its opening used a phrase every parent has heard: "weak foundations." Most people who reposted it also took that angle.

After that it never went quiet. The trick was posting in pieces: one subject per post (geography, biology, chemistry, Chinese literature...), each linking to the GitHub repo and the website. 11-04 The chemistry post alone got 16k views. The speed itself became the content. Over the last 7 days it gained 1578 stars, versus 3000 in the 7 days before. It's cooling, but still picking up a hundred or two a day.

The fork rate is unusually high: 1758 / 6755, about 26%, where repos with 5000+ stars of the same age usually sit around 11%. In an early November tweet the author said "Today I noticed something bad: someone has started charging for this." Many forks are probably parents or institutions pulling it down to deploy or resell (inference). The issue tracker also shows who is using it: someone built a Qingdao (Shandong) edition of 1–9 grade math, packaged it as an APK, and says "it's already being used on my kid"; a string of people asked for Shanghai, Jiangsu, and Lu editions and Foreign Language Teaching Press versions; and some carefully filed corrections backed by textbook page evidence.

Why projects with more data didn't take off

This topic wasn't first to market:

The clearest evidence is mmlong818: launched four months earlier with 20 times the data, the author even went under @jinglian's tweet to self-promote ("why is everyone sharing that incomplete version?"), and it still has only 265 stars.

So the product difference is real: among its peers, it's the only one that is open source, aligned to Chinese textbook sections, and includes a diagnostic web page a parent can open right away, which can be explained in one sentence. But that only makes it "good"; what made it "get shared" was the author's own 220k views and landing the timing right on the os-taxonomy wave.

Others are already following

• Someone is selling its data for a fee (the author says so, violating CC BY-NC).
• A batch of reposts exists on CSDN, GitCode, Instagram, and YouTube, but the volume is small.
• Issue threads show someone built a Qingdao edition and someone wants a Shanghai edition, but these are all add-ons inside its framework. No project that started fresh has gained stars (since September, new repos searchable by "prerequisite / pre-diagnosis / knowledge graph" haven't gained traction).

The Chinese K12 slot is already held by it, and the author keeps adding a subject every day.

If it were you

B. The recipe is learnable, but the two things that matter most are the ones it has and you probably don't. One is your own audience (an account with tens of thousands of followers, where posts routinely pull hundreds of thousands of views). The other is timing: catching the one or two days when an overseas hit goes viral in Chinese circles. Without those two, the same-topic mmlong818 is your outcome. The Chinese textbook niche is already full.

Who it suits: people who already have several thousand to tens of thousands of readers on X, Jike, RedNote, or WeChat Official Accounts; people who can write front-end code and quickly fill content with AI; people who know a particular country's curriculum or a type of exam well enough to judge whether AI output is right.

Who it doesn't suit: anyone who just wants to build another "Chinese textbook knowledge graph" (no matter how good, resource accounts won't repost the same thing twice); anyone with no distribution channel who expects a good README to get discovered; anyone without the patience to handle corrections and textbook edition issues, because parents and teachers will flag errors one by one.

How to do it:

• Other places: Marble only covered English and American primary schools. Japan (Courses of Study), Korea, India's CBSE, Vietnam, and Indonesia are still open. Build "prerequisite graph + free diagnostic web page" in the local language and share it with local parent communities and education bloggers.
• Other exams: skip grade levels and build prerequisite graphs around a single exam: graduate-school math entrance, civil service aptitude tests, IELTS, AP, algorithm questions for programming interviews. Adults anxious about their own exams spend their own time and money, so it spreads more directly than targeting children.
• Do what it didn't do well: textbook editions (Shanghai, Jiangsu, Beijing Normal, five-four system), the new editions from 2026 September, and question quality (blind review). The issue threads already have a line of people asking for these. But that path only suits contributing PRs to it or building a small dedicated "edition mapping" project. It's hard to earn lots of stars on its own.
• Whichever path you choose, keep the product shape the same: open data, a web page anyone can open right away, and a result image people can share in their friend circles. Write the copy for the end user's anxiety, not for programmers.

How to earn stars: watch the overseas open-source scene closely. When a dataset or tool catches fire in English circles and Chinese circles are only starting to repost it, produce a localized version within 48 hours so the people reposting can write "X exists abroad, and someone in China has already built Y." Post from your own account on launch day, write the copy around what users save, and end with "if it helps, give it a star." Afterward, don't post everything at once; break it into subject-by-subject or feature-by-feature posts, each with a link.

When to move: this wave of knowledge-graph hype has passed, and there's already a Chinese version. So you're chasing the next wave, which means there's no fixed date, just one rule: move within two or three days after the overseas hit starts catching on in Chinese circles.

What would signal you read it wrong: (1) Over the next month or two, a comparable project for Japan, Korea, or some exam gets picked up by big accounts and reaches a thousand stars. That would mean the "switch country / switch exam" route works, so act fast. (2) Its own growth drops within a month to tens of stars a day, and the issues are more corrections than suggestions. That would mean people wanted the "free diagnosis" story, not the data, and later entrants will find it harder to earn stars from data alone. (3) You have no audience, you build something more solid than it, and you send it to a few education and AI bloggers, yet nobody reposts it within a month. Then you're on the mmlong818 path: the topic itself isn't enough, and distribution is the bottleneck.

Data and sources

• Stars and forks: GitHub, 10-19 to 11-06 daily.
• Author's tweets: X @hassanyehiaa 10-18 pinned post (227k views), early-November geography / history tweets, 11-04 chemistry tweets.
• Reposts: X @jinglian (10-19, 10-23), @aigclink (10-20); Zhihu column; GitCode blog (10-21,2564 reads).
• Star counts of comparable projects: GitHub, 11-06.
• Issue quotes: repo issues #4, #3, #33 and others.
• Actual website usage is unknown; stars indicate interest, not how many parents are using it.