# An "AI × astronomy" list built in one day that got 9 stars in 386 days: the content is real, the stars probably aren't · GitHub open-source project

> Source: https://stringers.io/en/github/gh_Xin-YuXiao__Awesome-AI-for-Astronomy.html (Stringers / 什么值得做, generated 2026-11-20)
> Stringers tracks which small products suddenly took off each day, investigates why one by one, and distils playbooks from them. Numbers come from 15 databases we collect daily, each with source and date.
> For agents: these investigations and the underlying databases (verified revenue, App Store / Google Play charts, ad libraries, GitHub, Product Hunt…) are queryable live over MCP — tools feed (daily breakouts) / plays (playbooks) / hot / products / data. Setup: https://github.com/komako-workshop/stringers (the user signs in with Google at stringers.io/en/start.html to get a personal URL). Site index: https://stringers.io/llms.txt

> The older lists in the same field all stop at 2023–2024. It used AI to fill in nearly two years of papers in one day, and peers are willing to send it PRs. But the only visible spread is one LinkedIn post with 45 likes, and the steady curve of dozens of stars a day, the very low fork ratio, and the bot-account forks all suggest most of the stars look manufactured.

[GitHub open-source project](https://stringers.io/en/github/index.md) · written 11-18

Xin-YuXiao/Awesome-AI-for-Astronomy · Xin-YuXiao

# An "AI × astronomy" list built in one day that got 9 stars in 386 days: the content is real, the stars probably aren't

**What it is** A curated list of papers and tools on "AI for astronomy and astrophysics", 436 papers, each with a one-line summary, plus a searchable web page, a page that screens new arXiv papers every day, and an interface for AI tools.

**Why it caught on** Older lists in the same field stopped at 2023–2024. It used AI to fill in nearly two years of papers in a day, and peers were willing to submit PRs. But the visible spread is only one LinkedIn post with 45 likes. The steady curve of dozens of stars a day, the very low fork ratio, and forks from bot accounts all suggest most of the stars look bought.

**C** **How you could follow it** Pick an "AI × some discipline" field where an old list has stopped updating and papers have exploded in the last two years, and build it the same way (YAML data, a filterable web page, verification annotations, daily arXiv screening, an AI tool interface), and spread it through peers in the field and authors of the papers it includes.

Grade C: the "AI builds the library in a day, a machine keeps it updated daily" approach is worth learning, but the stars are most likely inflated, so it proves nothing about how many real stars you can get. Academic circles in astronomy are small, so the natural ceiling is a few hundred stars over several years. Hardly anyone knows about it yet

Data (updated daily) · Repo created 2026-11-10 · 398 stars · 7 days +307

GitHub · stars

Data 2026-11-12 → 2026-11-20 (9 days, 9 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

A student at UCAS used an AI coding assistant to build a list of papers on "AI in astronomy" in one morning, and in 9 days it got 386 stars, more than the oldest list in the field had collected in 8 years. The list itself is genuinely well made and fills a gap left by lists that stopped updating years ago. But where the stars came from doesn't hold up to scrutiny. Below: what it did right, where the stars came from, and, if you want to build an open-source project that gets stars, what to copy and what not to trust.

### One morning, 233 commits

At 2026 Nov 11 7:08 Beijing time, the GitHub account Xin-YuXiao pushed the first commit to a new repository. 47 minutes later, the repo already had 233 paper entries. Commit messages read "Add 54 verified entries" and "Add 35 verified 2023-2026 entries", added in batches; up until recently, the commits still ended with "Co-Authored-By: Claude". It's clear this was assembled in batches by an AI coding assistant, with a human checking entries one by one. The next day, a web version and an auto-generated README pipeline went live.

That afternoon, a little after 3 o'clock, the author posted on LinkedIn: "Awesome-AI-for-Astronomy is now public. It collects 326 papers and 65 open-source tools, sorted into 18 topics, and about half were published between 2024 and 2026. Contributions welcome." It got 45 likes and 7 comments. The visible ones were "Have you done high-energy astrophysics?", "I'm an MBZ astronomer, let's talk sometime", and "Awesome".

The author, Xin Xinyu, lists astronomy and lunar exploration applications of large models on his profile. His GitHub account was registered in 2022, has 0 followers, and this is the only original repository he has starred. 9 days later (11-19), the repo had 386 stars.

### What it actually is

A curated list of "AI for astronomy, astrophysics, cosmology, heliophysics, and planetary science", but not the kind of README that just dumps links:

- **436 papers, 60+ software packages** , sorted into 20+ sections (foundation models, large language models, agents, benchmarks, galaxies, gravitational lensing, time domain, exoplanets, cosmology...), with each section in reverse chronological order.

- **A one-line summary for each paper** , marked with the journal or conference (bolded when verified via DOI), and whether code, weights, and data are public. The README states that every arXiv ID, DOI, and repository link was checked before inclusion, and an automated dead-link check runs as well.

- **A searchable, filterable web version** : filter by year, publication type, whether code is available, and data type. The whole list can be downloaded as JSON and CSV.

- **The "AstroAI Today" page** : GitHub Actions pulls new arXiv papers daily, DeepSeek screens for AI-related ones, and each gets a one-line TL;DR. There are also weekly and monthly digests, marked separately from the hand-curated portion.

- **A skill for AI tools** : lets Claude Code and similar tools query this catalog directly.

- A Zenodo DOI is attached, so people writing papers can cite it directly.

### The gap it fills is real

"AI × astronomy" lists already existed, but they had all stopped:

| List | Launched | Scale (November) | How it differs |
|---|---|---|---|
| georgestein/ml-in-cosmology | 2018-12 | 436 stars, stopped updating after 2023-04 | Covers only cosmology, pure README, no summaries or web page |
| dr-guangtou/taotie | 2018-12 | 472 stars, stopped updating after 2024-07 | A grab-bag of astrophysics resources, not focused on AI |
| jonathansick/awesome-astronomy | 2015-05 | 631 stars | Collects astronomy software, not AI papers; this list lists it as a supplement in its own README |
| mpi-astronomy/awesome-astro-datascience | 2022-04 | 67 stars, stopped updating after 2024-01 | Classic astronomical data science, doesn't chase large models |
| ai4s-research/awesome-ai-for-science | 2025-10 | 2,104 stars | All disciplines, astronomy is just one section |
| hyperai/awesome-ai4s | 2024-04 | 3,406 stars | A collection of Chinese-language paper walkthroughs, very light on astronomy |

(GitHub, November)

Since 2024, foundation models, large models, agents, and benchmark papers in astronomy have grown fast, and the old lists haven't kept up at all. This is currently the only one that is still updated, structured, updated automatically every day, and callable by AI tools. On GitHub, new repos on the same topic created after September have all gotten 0 to 1 stars, and nobody has followed up.

Peers have noticed too. Within a week, five or six outside people opened issues and PRs: Yu-Yang-Li, who builds the astronomy large model StarWhisper (332 stars), submitted 30+ PRs and also wrote that skill; several domestic astronomy ML researchers and one overseas developer opened issues and PRs, and a Turkish astronomy master's student forked it. For a list that's 9 days old, that's real response.

### Where the stars came from: only one visible post

Daily stars (GitHub, November):

| Date | star | Added that day |
|---|---|---|
| 11-12 | 39 | — |
| 11-13 | 91 | +52 |
| 11-14 | 154 | +63 |
| 11-15 | 182 | +28 |
| 11-16 | 216 | +34 |
| 11-17 | 276 | +60 |
| 11-18 | 325 | +49 |
| 11-19 | 379 | +54 |

The only traces of it spreading are the author's LinkedIn post, which got 45 likes, and a small circle of domestic peers. I found no discussion on X, Hacker News, Reddit, or Chinese web pages. The PullRepo 11-14 daily report ranks it #1 in Other AI Tools by growth, and trendshift lists it, but these sites pick up repos after the stars have already risen, so they are a result, not a cause. It also never made GitHub trending (repos need about 60 days since creation to qualify).

Online, people claim it was passed around by several people in the LinkedIn astronomy scene. Checking those posts: one was published over 4 hours before the repo was created, and the other 3 days before. Both appear as auto-recommended 'related posts' at the bottom of LinkedIn pages. Neither is sharing this repo.

### The growth curve doesn't look like it came from a post

A repo lifted by a single post usually spikes the day of the post and then tapers off. This one does the opposite: the post went up on 11-11, and for the next 8 days the repo gained a steady 28 to 63 stars a day, with no single-day jump. Several mismatches stand out together:

- **Against the field's ceiling** : the best lists in this niche took 8 to 11 years to reach 436 to 631 stars. There are only a few thousand people doing ML in astronomy worldwide. A post with 45 likes bringing in 386 stars within 9 days doesn't add up.

- **Too few forks** : the fork-to-star ratio is 2.9% (11 / 386). Comparable lists sit at 11% to 14%.

- **Nobody watching** : there are 0 watchers, and the author doesn't watch it either.

- **11 of the 6 forks look like bot accounts** : they were mass-registered from late August to early October 2026, with usernames made of 'word + word + a few alphanumerics' and templated bios like 'Marketing Specialist | gardening' or 'Pharmacist — music, tennis', with 0 following and 0 followers. Each account forked only four to ten repos, nearly all small new domestic projects: a run of 'McDonald's MCP' repos (mcd-budget-meal, mcd-optimizer, mccarbon, ...), ling-stack, gezhi-sheets, and so on.

- **Repos serviced by the same accounts have had stars purged by GitHub** : mcd-budget-meal once lost 288 stars in a single day; mccarbon gained 937 in a day and then declined; gezhi-sheets, which shares a bot account with this list, has daily gains of 43, 63, 13, 41, 51, 85, 72, almost the same rhythm as here.

So the most reasonable explanation is that **most of the stars were bought** . That's an inference, and I'm fairly confident in it, but it isn't proven. GitHub doesn't publish stargazer lists for this repo, so I can't see who starred it. I also can't rule out that the bot accounts were parked on small repos by someone else rather than bought by the author. Domestic WeChat groups and lab forwarding, which don't show up in outside search, may explain part of it too.

The motive is easy to understand. Star counts look good on a CV or in application materials, and if you get to 60 in a few days you might even reach trending.

### What you can learn

Taken apart, the project has two things: one worth learning from, and one not to believe.

**What's worth learning is the method** : pick a niche where 'AI × a specific field' exists but the old lists stopped updating two or three years ago. Use an AI coding assistant to compile recent papers in a day or two, manually verifying each entry (journal, code, data). Store data in YAML and generate the README and a web page with scripts so people can filter and download. Add a daily page that uses a cheap model to screen arXiv and write a one-line summary, plus a skill for AI tools. Work that used to take one person years of maintenance can now be set up in a day, with the machinery keeping it fresh. This approach is what attracted real peer contributors.

**What not to believe is the star count** : it doesn't prove that 'doing this gets you 9 stars in 400 days.' Going by the experience with older lists in the same field, a few-thousand-person academic community naturally tops out at a few hundred stars and takes years to get there. Treating it as evidence that the approach 'has been validated' needs a heavy discount.

### If it were you

**Verdict** : C. You can learn a very efficient 'vertical paper list' method from it, but the stars are most likely bought, so it can't show how many real stars this path could bring in. The astronomy community is also small.

**Who it suits** : graduate students or postdocs already doing AI research in a particular discipline, who can judge paper quality and have some connections in that field's LinkedIn or academic circles. The goal is building a reputation for a list everyone in the field uses, not chasing stars.

**Who it's not for** : people outside the field who just want to swap in a different subject and chase stars. Papers can't be checked, peers won't open PRs, and you end up with another abandoned list. Anyone trying to copy "9 days, 400 stars" should know that curve can't be trusted.

**How to build it** : pick a field with the same pattern, where the old list has stopped updating and papers have exploded over the last two years. First search GitHub to confirm nobody is maintaining one (chemistry, materials, earth science, medical imaging, high-energy physics are worth a look; avoid all-discipline AI 4S lists that already have big ones). Copy the old list's structure: store the data in YAML, auto-generate the README and a filterable web page, tag every entry with code and data, attach DOIs for citation. Then add a daily arXiv screening page and a skill for AI tools. The old lists have neither, so those two are the easiest differentiator to point to.

**How to get users** : in academic circles, stars come from peers. Post on the field's LinkedIn and X, leading with how many papers it covers and how many are from the last two years. Send the list to authors of the papers it includes, since they are the most likely to share it. Ask people in your own lab and collaborators to submit PRs so the repo has activity from day one. Submit to workshops or mailing lists at relevant conferences. Don't buy stars: GitHub has already purged repos serviced by the same bot accounts.

**How early you need to be** : there's no rush, since academic lists compete on who occupies an empty niche first and who keeps updating it. But 'daily automatic updates' is something anyone can copy, and once another person has claimed a field it's hard to squeeze in, so if you're interested, move early.

**What would show I got this wrong** : (1) if this list is cited in astronomy circles over the next few months, gets picked up by prominent accounts or journals, and its external contributors grow from five or six to dozens, that would mean real demand for vertical lists is bigger than estimated here, and copying the approach to other disciplines is worth pursuing more actively; (2) conversely, if its stars are purged in bulk on some day (like mcd-budget-meal losing several hundred in one day), the star-buying judgment is confirmed; (3) if someone uses the same method to build a list for another discipline and gets several hundred stars naturally, without buying, in two or three months, then the ceiling for this path is higher than astronomy's, and that field is worth pursuing.

### Data and sources

GitHub: repo page, README, commit history, issues and PRs, fork account profiles, daily star records (11-12 to 11-19), and for the same list, star counts and last commit dates (November). LinkedIn: the author's 2026-11-11 post (45 likes, 7 comments) and the posting dates of two "related posts" (inferred from post IDs). Aggregators: PullRepo 2026-11-14 daily digest, trendshift. Searches on X, Hacker News, Reddit, and Chinese-language web found no discussion. Wherever it says "inferred", it is an inference.

[GitHub repo ↗](https://github.com/Xin-YuXiao/Awesome-AI-for-Astronomy)

**Research notes: competitor table, acquisition evidence, full research report**

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| georgestein/ml-in-cosmology | Open source | 2018-12 | 436 stars / 50 forks, no updates after 2023-04 | Cosmology only; plain README with no summaries, web page, or auto-updates |
| dr-guangtou/taotie | Open source | 2018-12 | 472 stars / 57 forks, no updates after 2024-07 | Grab-bag of astrophysics resources, not focused on AI |
| jonathansick/awesome-astronomy | Open source | 2015-05 | 631 stars / 88 forks | Collects astronomy software and resources, not AI papers |
| mpi-astronomy/awesome-astro-datascience | Open source | 2022-04 | 67 stars, stopped updating after 2024-01 | Covers classical astronomical data science, doesn't track large models or agents |
| astroML/astroML | Open source | 2012-10 | 1,209 stars | Codebase plus textbooks, not a paper index |
| ai4s-research/awesome-ai-for-science | Open source | 2025-10 | 2,104 stars | Full-subject AI4S list, astronomy is just one section |
| hyperai/awesome-ai4s | Open source | 2024-04 | 3,406 stars | Chinese AI4S paper digest collection, hidden very shallow |
| awesome-JAXtronomy / awesome-helio / awesome-astrophysical-simulation-codes | Open source | 2021–2024 | 31–127 stars | Each covers one subfield |
| benty-fields / arXiv astro-ph daily listings / NASA ADS | Web | Already around | Everyday tools for the astronomy community | Recommends papers daily, but doesn't filter for AI-related work or write TL;DRs |

Where users come from

- **Verified** **Author's LinkedIn post** : posted at 2026-11-11 07:19 UTC, '326 papers and 65 tools', 45 likes / 7 comments

- **Confirmed** **Chinese astronomy ML community (UCAS / StarWhisper)** : Yu-Yang-Li submitted 30 several PRs adding skills; ruiyicheng, Maxwell-Jia, xuquanfeng, and HIGGS317 filed issues / PRs; uzayyaydiin forked

- **Verified** **GitHub star-growth aggregators (passive listing)** : PullRepo 11-14 daily report ranks it #1 in Other AI Tools (Growth Score 97.62); trendshift and risingtechrank list it

- Inferred **Suspected star-buying** : 6/11's forks come from 2026-08–10 mass-registered bot accounts with fake identities, which also fork a run of small domestic repos (the mcd-* McDonald's MCP series, ling-stack, gezhi-sheets, etc.), one of which (mcd-budget-meal) lost 288 stars in a day; this repo's smooth curve of 28–63 stars a day, its 2.9% fork ratio (comparable 11–14%), and its 0 watchers; the stargazer list endpoint isn't visible, so it can't be checked directly

- Inferred **LinkedIn third-party sharing (claim from the data card)** : the Charles McGowen post, decoded, was published at 11-10 18:50 UTC, before the repo was created; the Zahra Sharbaf post is dated 11-07; both appear as related posts on the page and are not sharing this repo

## Xin-YuXiao/Awesome-AI-for-Astronomy: why it 'took off'

### Bottom line first

- **What it is** : a curated list of papers, models, datasets, benchmarks and software for 'AI in astronomy / astrophysics.' It also has a searchable, filterable web version, an 'AstroAI Today' page that uses DeepSeek to screen new arXiv papers daily and write a one-line summary, a research skill that AI tools can call, and a Zenodo DOI. The author is Xin-Yu Xiao, a student at UCAS (University of the Chinese Academy of Sciences); his homepage lists research on large models for astronomy and lunar exploration.

- **The content is solid** : 436 papers and 77 software/resource links, each with a one-line summary, journal/conference tags, code, weights and data labels, plus automated link checking. The existing lists in the same field have stopped updating (ml-in-cosmology's last commit was in 2023-04, awesome-astro-datascience in 2024-01), so it fills a real gap.

- **But the star speed most likely didn't grow naturally** : the only visible organic channels are the author's own LinkedIn post (45 likes, 7 comments) and a circle of domestic astronomy ML peers. That can't support a curve of '9 stars in 386 days, a steady 28–63 a day, with no single-day spike.' The best lists in the field took 7–11 years to reach 436–631 stars. The fork ratio is only 2.9%, versus 11–14% for comparable lists. 6 of the 11 forks are mass-registered bot accounts with fake identities, and these accounts also fork a run of small domestic projects, one of which GitHub once purged of 288 stars in a single day.

- **Conclusion** : the product's differentiation is real (structured data, automatic updates, callable by agents), which is why it has genuine contributors in a small circle. But the star count that 'took off' mostly looks propped up by bought stars, not carried by organic spread.

### 1. Basic data

| items | value | Source |
|---|---|---|
| Repo created | 2026-11-10 23:13 UTC | GitHub API |
| star / fork | 386 / 11(11-19) | GitHub API, GraphQL |
| watchers | 0 | GitHub GraphQL |
| Daily stars | 11-12 39 → 11-13 91 → 11-14 154 → 11-15 182 → 11-16 216 → 11-17 276 → 11-18 325 → 11-19 379 | Data card / GitHub repo_day |
| Daily net gain | 52, 63, 28, 34, 60, 49, 54 | github repo_day |
| trending | Never listed (trending only includes repos about 60 days old, which proves nothing either way) | GitHub repo |
| Author | Xin-YuXiao, account registered 2022 years ago, 0 followers, and this is the only one of his original repos with any stars | GitHub API |
| Commits | 161 in total: the author 86 times, Yu-Yang-Li 42 times, the actions bot 26 times, others 4 times | GitHub commits |

Content: 436 papers, grouped into 20 sections, newest first within each section; each has a one-line summary, journal/conference (bolded if verified via DOI), and code/weights/data links. The web version can filter by year, publication type, whether code is available, and data modality, and can download JSON / CSV.

- **AstroAI Today** : a GitHub Actions job pulls arXiv every day, filters it with DeepSeek V4.1 Flash, writes a TL;DR for each paper, and also publishes weekly and monthly digests.

- **`.agents/skills/astroai-research/SKILL.md`** : lets Claude Code and similar tools query this catalog directly.

- The repo has `.claude/skills/`, and the early commits are bulk imports like "Add 54 verified entries" and "Add 35 verified 2023-2026 entries". First commit on the day the database was created was 23:08, and 233 entries landed 47 minutes later. The next day it had a web version and a YAML generation pipeline. Clearly built with an AI coding agent plus manual verification.

### 2. Competitors and similar projects

| Name | Where | Launched | Scale | How it differs |
|---|---|---|---|---|
| georgestein/ml-in-cosmology | Open source | 2018-12 | 436 stars / 50 forks, no updates after 2023-04 | Cosmology only; plain README with no summaries, web page, or auto-updates |
| dr-guangtou/taotie | Open source | 2018-12 | 472 stars / 57 forks, no updates after 2024-07 | Grab-bag of astrophysics resources, not focused on AI |
| jonathansick/awesome-astronomy | Open source | 2015-05 | 631 stars / 88 forks | It collects astronomy software and resources, not AI papers, and the README itself lists this one as a supplement. |
| mpi-astronomy/awesome-astro-datascience | Open source | 2022-04 | 67 stars / 12 forks, no updates after 2024-01 | It covers classic astronomical data-science resources and doesn't chase large models or agents. |
| astroML/astroML | Open source | 2012-10 | 1,209 stars | One code repo plus a textbook, not a paper index. |
| JAXtronomy/awesome-JAXtronomy, i4Ds/awesome-helio, pmocz/awesome-astrophysical-simulation-codes | Open source | 2021–2024 | 31–127 stars | Each covers one subfield (JAX, heliophysics, simulation codes). |
| ai4s-research/awesome-ai-for-science | Open source | 2025-10 | 2,104 stars | A cross-discipline AI4S list where astronomy is just one section. |
| hyperai/awesome-ai4s | Open source | 2024-04 | 3,406 stars | A Chinese collection of AI4S paper reviews, broad coverage, shallow on astronomy. |
| Yu-Yang-Li/StarWhisper | Open source | 2023-07 | 332 stars | The astronomy LLM "StarWhisper" itself, not an index; its author is also the second-largest contributor to this list. |
| benty-fields, the arXiv astro-ph daily listing, NASA ADS | Web | Already around | Everyday tools for the astronomy community | They push papers daily, but don't filter for "AI-related" and don't write TL;DRs. |

Summary: The "AI × astronomy" niche already has lists, but they're all 2018–2022 era lists that have stopped updating, and they're just plain READMEs. This one is the only one still maintained for 2024–2026 years, structured, auto-updated, and callable by agents. No copycats found in the same period: among the newer repos on the same topic on GitHub searched by 2026-09, the rest have only 0–1 stars.

### 3. Growth: where users come from

#### The organic channels I can confirm (all very small)

1. **The author's LinkedIn launch post** (2026-11-11 07:19 UTC): it says "326 papers, 65 open-source tools, 18 topics", with 45 likes and 7 comments (LinkedIn page num-reactions). [www.linkedin.com/posts/xin-yu-xiao-344146313_github-xin-yuxi…](https://www.linkedin.com/posts/xin-yu-xiao-344146313_github-xin-yuxiaoawesome-ai-for-astronomy-activity-7526181692429230080-gMo7)

2. **The Chinese astronomy-ML peer circle** : there is real outside participation.
    - Yu-Yang-Li (UCAS, author of StarWhisper) opened 30 PRs plus a research skill. The PRs came in every 10–15 seconds, which is also agent-batched.
    - ruiyicheng opened PR #3, Maxwell-Jia opened issues #2 and #4, and xuquanfeng opened issue #14.
    - HIGGS317 (Divyansh Tripathi) opened PR #15.
    - uzayyaydiin (a Turkish astronomy master's student) forked it.
    - These are the real signs of "it's being talked about in a small circle."

3. **Aggregator reposts** : all were passive pickups after the repo gained traction, so they are a result, not a cause.
    - PullRepo ranked 11-14 #1 for growth in "Other AI Tools" in its daily report (Growth Score 97.62, 157 stars at the time). [pullrepo.com/report/todays-other-ai-tools-fastest-growing-pr…](https://pullrepo.com/report/todays-other-ai-tools-fastest-growing-projects-november-14-2026)
    - There are also trendshift.io/repositories/353039 and risingtechrank.

#### The data card's claim of "spread in the LinkedIn astronomy circle" doesn't hold up.

- The post the card names, by Charles McGowen, decodes from its LinkedIn activity ID to 2026-11-10 18:50 UTC, more than 4 hours before the repo was created (23:13 UTC).

- The Zahra Sharbaf post is from 2026-11-07 23:08 UTC.

- Both are "related posts" at the bottom of LinkedIn pages, not posts sharing this repo.

- X (x-radar, 0 results over the past 30 days), HN (0 hits in the hackernews library), Reddit, and Chinese web searches (Brave / Jina / Exa) turned up no discussion.

#### Signs of star-farming (strong, but the stargazer list can't be seen directly)

1. **Curve shape** : over 8 days it gained 28–63 stars a day, with no sudden spike. Repos lifted by a post usually spike on the day of the post and then decay, but the author's only post was on 11-11, and afterward the daily gain stayed steady around 50.

2. **Against the field's ceiling** : the most successful comparable list in this niche, ml-in-cosmology, took 8 years to reach 436 stars, and awesome-astronomy took 11 years to reach 631. There are only a few thousand people doing astronomy ML worldwide. A list gaining 386 stars in 9 days off 45 likes is implausible.

3. **Fork ratio** : 2.9% (11/386). Comparable repos sit at 11–14%: ml-in-cosmology 11.5%, taotie 12%, awesome-astronomy 14%, awesome-ai-for-science 13%. And of these 11 forks, only 5 are from real people.

4. **Watchers is 0** : not even the author watches it.

5. **6 bot-like accounts forked it** : cobwebgoes4b, crispsatlas-3q, repairs78peaches, dibs-livings-7u, docents-node6r, cyanmistier-5g.
    - They were all registered in batches between 2026-08-31 and 10-06. Usernames are "word+word+alphanumeric", display names are made-up Western names, and bios follow templates like "Marketing Specialist | gardening" or "Pharmacist — music, tennis", with 0 followers / 0 following.
    - Each account forked only 4–10 repos, and nearly all were new, small Chinese projects: mcd-budget-meal, mcd-optimizer, mcd-calorie-butler, mcd-kcal-picker, mcd-points-actuary, mcd-saver-checkup, mcd-wrapped, mccarbon (a cluster of "McDonald's MCP" projects), plus ling-stack (forked by 3 of the accounts), gezhi-sheets, winutil-cn, ai-mingren-video-generation, and others.
    - Accounts meant to look real usually fork famous repos. This batch forks only obscure small ones, which looks more like the customer list of a star-buying service.

6. **Other repos served by the same accounts were later purged of stars by GitHub** :
    - Sam-Hui-dot/mcd-budget-meal lost 288 stars in a single day, netting down.
    - yeqv111/mccarbon gained 937 in one day, then declined instead of rising.
    - beiwocaishijia/gezhi-sheets (created 11-09, forked by the same bot account as this repo) gained 43, 63, 13, 41, 51, 85, 72 per day, almost the same rhythm as this repo.
    - The above comes from the github library's repo_day / repo_profile data.

7. **Limitations** : GitHub's stargazers endpoint returns 404 for this repo, and GraphQL returns an empty list too, so the individual people who starred it can't be seen. "Star-farming" is therefore a strong inference, not proof. It's also possible these bot accounts fork small popular repos to age themselves; but the ones they fork are small Chinese repos with 100–2000 stars, which doesn't fit well with riding on popularity.

### IV. Why it took off (in two layers)

**As a product, it really took off in a small circle** , for these reasons:

1. **It filled a gap left by abandoned lists** : the comparable lists in the field all stopped around 2022–2024, while papers on astronomy foundation models, agents, and benchmarks exploded in 2024–2026. About half its entries are from 2024–2026 (the author's own LinkedIn words).

2. **An AI agent built a "verified" high-density list in a day** , and DeepSeek keeps it updated automatically every day. The barrier shifted from "one person maintaining it for years" to "one person sets it up in a day, the machine keeps it running."

3. **Details useful to researchers** : journal verification, code / weights / dataset labels, one-line summaries, JSON / CSV, DOIs for citation, plus an agent skill. This made peers willing to open PRs, with 5–6 external contributors within a week.

**As a star count, its "breakout" doesn't come mainly from the things above** : the only visible spread is one LinkedIn post with 45 likes. The curve is smooth, the fork ratio is extremely low, some forks come from bot accounts, and those accounts overlap with known star-farming networks. Taken together, the most likely explanation is that most of the stars were bought or traded. The motive is common among students: résumé and academic clout, plus a shot at trending once the stars pass 60 days.

### 5. Differentiation

- **Compared with the old lists** : it's a "structured database," not a pile of README links. It's filterable, has summaries and verification, can be downloaded and cited, and updates automatically every day.

- **Compared with cross-discipline AI4S lists** : it covers only astronomy, and splits it finely (galaxies, gravitational lensing, time domain, exoplanets, heliophysics, radio, planets, and 20 more sections).

- **Compared with arXiv daily-push tools** : it first uses an LLM to filter for "is it AI-related," then writes the TL;DR, and labels human picks separately from machine recommendations.

- **Is it why it grew** : it explains why it got real contributors and peer recognition. It doesn't explain the 9 stars in 386 days; the star count mainly didn't come from differentiation.

### 6. What to take away

- The combination "AI coding agent builds the repo in a day + human verification + LLM auto-updates daily + skill / JSON interface for agents" can be carried over to any abandoned "AI for X" vertical list (chemistry, materials, earth science, medical imaging...). What's actually valuable is "an abandoned vertical + structured + automatically updated."

- Launch playbook: post once on LinkedIn in the academic circle, have people in the same lab batch-open PRs, and make the repo look active.

- The downside: the same star-farming accounts get cleaned up by GitHub (mcd-budget-meal dropped 288 stars in a day), so star counts are unreliable. Discount it heavily as evidence that demand was validated.

### VII. Uncertain

- Without the stargazer list, the share of farmed stars can't be quantified.

- I don't know whether WeChat groups, Moments, or the group meetings and chats of the Chinese astronomy community passed it along. That's the only channel that might explain part of the organic growth and still leave no trace online.

- Web demo and AstroAI Today page: no real traffic data.

- Fork accounts can also be that same batch of accounts 'farming' themselves, not necessarily bought by the author.

### Sources

- GitHub API: repo, author, commits, issues / PRs, forks, fork account profiles; GraphQL (stargazerCount 386, watchers 0, stargazers list empty)

- github library (agent-im DuckDB): repo_profile / repo_day, including this repo, comparable lists, and repos forked by the bot accounts

- hackernews library: 0 entries

- x-radar: 0 posts in the last 30 days

- LinkedIn author post (read via Jina, 45 reactions / 7 comments) and posts by Charles McGowen and Zahra Sharbaf (activity IDs decoded to get the posting time)

- PullRepo 2026-11-14 daily report, trendshift.io/repositories/353039

- Web version entries.json (meta: 436 papers, 77 links, generated 2026-11-18)

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.
