Buying GitHub stars is the sort of thing that invites the wrong kind of laughter. It sounds like a startup founder paying for imaginary applause. Embarrassing, yes. Slightly comic, yes. Also not the point.

The interesting part is that GitHub stars are not just decorative pixels. They are a public reputation signal on a platform where people discover code, evaluate projects, shortlist dependencies, and make rough judgments about what looks active, credible, and worth a closer look. Once that signal is cheap to launder, the problem stops being vanity and starts becoming trust.

That is what makes the market for fake stars worth writing about. Not because I think a few bought stars magically turn a mediocre repository into the next Linux. The evidence for that stronger claim is weak, and there is now direct experimental work suggesting the effect on downloads may be much smaller than people assume. The real issue is simpler and more structural: code hosting has acquired the counterfeit-attention economy of every other platform, except that the thing being counterfeited here is credibility around executable artifacts.


Why Stars Are Worth Buying At All

Developers like to pretend they do not use heuristics. They do.

The 2018 paper What’s in a GitHub Star? by Hudson Borges and Marco Tulio Valente found that stars are not merely bookmarks or idle gestures [[1]]. They function as a visible signal of interest, approval, and project standing. Of the 791 developers surveyed, about half said they star repositories to show appreciation and about half to bookmark them, and more than a third because they use the project; of the 777 who answered the follow-up question, three out of four (567) said they consider star counts before using or contributing to a repository. Earlier work by Borges, Andre Hora, and Valente treated stars directly as a popularity measure [[2]]; later work on repository centrality argued that star-network structure itself carries meaningful information about project prevalence and lifespan [[3]].

None of this means stars are a perfect metric. It means they are a real one. Imperfect metrics are exactly the kind people manipulate, because they are visible enough to matter and simple enough to game.

There is also a social layer here that is hard to quantify but easy to observe. A repository with five stars looks like somebody’s weekend experiment. A repository with five thousand looks like infrastructure. That impression may be lazy, unfair, and often wrong. It is still how human beings scan crowded information environments. A star count is not a proof of quality. It is a shortcut. Shortcuts are where fraud lives.


The Business Is Surprisingly Banal

The market for fake GitHub engagement is less “underground sophistication” than ordinary engagement fraud wearing a developer T-shirt.

The best recent academic anchor here is Hao He and colleagues’ Six Million (Suspected) Fake Stars on GitHub, presented at ICSE 2026 [[4]]. The title sounds dramatic, but the really useful part is how undramatic the mechanism is. The authors’ tool, StarScout, searched GitHub’s public event archive for two signatures: accounts whose entire history is a single star (plus at most one other event in the same repository that day), and groups of accounts that star the same repositories in lockstep. Between July 2019 and December 2024 it flagged six million suspected fake stars. The authors then kept only repositories that look like buyers rather than bystanders: a month with more than 50 suspect stars making up more than half of that month’s stars, and more than 10% suspect stars overall. That leaves 18,617 repositories with what they call fake-star campaigns, holding 3.81 million of the suspect stars from 301,000 accounts that do little else. The activity rose sharply in 2024. This is not a rare exotic trick. It is a market.

It is small in volume and large where it is aimed. Fake stars were at most 1% of all stars in any month, but in July 2024, the peak, 16.66% of the repositories that gained at least 50 stars that month had run a campaign. These are detector outputs, not confirmed purchases. The first preprint, with data to October 2024, counted 4.5 million suspected fake stars, 3.1 million of them in campaigns; those are the figures BleepingComputer reported [[5]].

Reporting and vendor pages make the market look even more ordinary. WIRED managed to buy 50 stars for a dormant repository for USD 6 paid in ether [[6]]. Dagster, while building its own fake-star detection work, bought stars from public services including Baddhi Shop and GitHub24 [[7]]: the first quoted 1,000 stars for as little as USD 64, the second EUR 0.85 a star. The table of seven storefronts in He and colleagues’ paper, compiled in July 2024, lists USD 0.10 to 1.62 per star (the text says up to 2.00), and the paper notes that stars usually sell in batches of 50 or 100 [[4]]. In April 2026, public storefronts advertised GitHub stars with the same language used across the wider SMM economy: visibility, credibility, promotion, “real engagement”, instant delivery, a 30-day guarantee [[8]] [[9]]. Both pages have since been rewritten.

The payment story is also more boring than the confident crypto-only versions of this thesis. Cryptocurrency does show up repeatedly. WIRED’s test purchase used ether. Some vendors clearly lean on crypto rails. But the visible market is mixed, not exclusively crypto. Followdeh’s checkout now takes cryptocurrencies alongside Alipay, WeChat Pay, and PhonePe, with PayPal handled by hand at a 20% surcharge [[8]], and Baddhi Shop takes PhonePe [[9]]. PlayerUp, a marketplace for second-hand accounts that lists star offers too, pays its sellers by PayPal, bank wire, UPI, or card [[10]]. That matters because it tells us what kind of market this is. Not an occult darknet specialty. Just a grubby, normalised, public-facing attention business that happens to sell fake trust around code.

None of this is allowed. GitHub’s Acceptable Use Policies prohibit “rank abuse, such as automated starring or following” and “creation of or participation in secondary markets for the purpose of the proliferation of inauthentic activity” [[11]]. Baddhi Shop’s product page, which in April promised “No bots or fake accounts”, now asks buyers to review those policies before ordering [[9]]. Enforcement exists, but it arrives late. GitHub “does ban suspect accounts in many cases”, Check Point found, yet “the operations run uninterrupted for a long time before those measures are taken” [[12]]. The stars WIRED bought were gone within a month [[6]]. A month after Dagster’s purchase, all 100 GitHub24 stars still stood and a quarter of the Baddhi Shop stars had gone; within 48 hours of its blog post, every account that had starred its test repository was deleted, by “either the GitHub team or both fake star vendors” [[7]]. The stars bought for the experiment described below seem to have gone the same way, eventually [[13]].

That banality is the point. If laundering repository popularity required elite access, the security implications would be narrower. If it is a cheap checkout flow, the problem is structural.


What The Evidence Says Stars Actually Do

This is where the piece needs one explicit concession.

The strongest version of the popular argument would be: buy stars, get users. That turns out to be too neat. Lucas Shen, Gaurav Sood, and Daniel Weitzel’s 2026 paper Social Proof is in the Pudding ran field experiments on GitHub and the Python Package Index [[13]]. Of 622 repositories behind Python packages first listed in late April 2023, 100 chosen at random received stars from the authors’ network, and 25 of those got 50 more bought from Baddhi Shop, every one from an account created around 20 April 2023. At the end of the intervention the two treated groups’ median gains over the control group were 20 and 69 stars. Three months later there was no discernible effect on package downloads, forks, pull requests, issues, or other developer-engagement measures; a second experiment that inflated download counts found nothing either. If your claim is that stars are a reliably profitable conversion funnel all by themselves, that paper is bad news.

The authors mark the limits themselves. The dose was modest by design and given once, where real campaigns are sustained, and the bought part did not last: a later check found the high-dose group’s median down from about 70 stars to 22, consistent with the purchased stars being removed. A single organisation adopting a malicious package could matter in practice, and the experiments are, in their words, “underpowered to detect anything so subtle”. They add that “it is entirely plausible that a more intense treatment would persuade developers”.

He and colleagues’ observational data point the same way [[4]]. Fake stars were followed by a small gain in real stars that faded within two months, about a fifth of the gain that real stars bring, and over the longer run a history of fake stars went with fewer new stars, not more. Only 78 of the 18,617 campaign repositories (0.42%) appeared on GitHub’s Trending page, although the archive the authors used covers just five languages, so that is a lower bound. And 738 published packages link back to 229 of the repositories (1.23%), most with no other package depending on them; since such links are optional, that count is low as well.

The null result is useful news all the same, because it sharpens the argument.

The case against fake stars does not need to rest on a simplistic theory that developers see a number, lose all critical faculties, and start running whatever glitters. Stars can matter without mechanically driving downloads. They can matter at the earlier and sloppier stage: first impressions, shortlisting, perceived legitimacy, investor decks, media coverage, conference talk slides, recruiter heuristics, and the general “this seems real” layer that precedes more careful scrutiny.

Among the sources cited here, that stage is never measured on its own (Fig. 1). He and colleagues’ regression watches real users respond to star counts, but the response it records is more stars, not a choice of dependency. Shen and colleagues run the whole chain from stars to downloads, so their null result covers shortlisting as well, at their dose and power. The only direct evidence on shortlisting itself is what developers told Borges and Valente about themselves [[1]]. The experimenters say their study “allays but does not dispel” the concern [[13]], and I read it the same way: the null result narrows the plausibility argument without settling it.

Figure 1. In these sources, shortlisting is reached only through self-report and through one experiment’s end-to-end null result at two modest doses. The experiment’s group medians are at their peak, at the end of the intervention; the high-dose median later fell to 22 stars, consistent with removal of the bought stars [[13]]. The Trending count is a lower bound [[4]]. The band marks where supply-chain harm happens (next section); no source counts victims per purchased star.

In other words: fake stars may be bad at producing commitment, but still good at laundering plausibility.

And plausibility is often enough. Attackers do not need every viewer. They need some viewers to stop being careful.


Where This Turns From Embarrassing To Dangerous

He and colleagues’ paper is useful precisely because it does not leave the issue at vanity. Its most important result is not that some founders may be inflating ego metrics. It is that the majority of fake stars in the authors' analysis were used to promote short-lived phishing and malware repositories [[4]]. Part of that is inference, and the paper says so. By the time the authors checked, 90.42% of the campaign repositories had been deleted; their names point to pirated software, game cheats, and cryptocurrency bots. Of the repositories still online, the authors classed 31.1% of a random sample as spam or active phishing, and 22.6% of all those they coded.

That is the real story.

GitHub has been a security surface for years. Beyond the Surface checked 47,285 GitHub repositories of proof-of-concept exploits for vulnerabilities discovered from 2017 to 2021 and found 899 (1.9%) with indicators of malicious intent: exfiltration, malware installation, trojanised binaries, hard-coded reverse shells [[14]]. Unveiling A Hidden Risk had ChatGPT label the metadata of 35,200 repositories that declared themselves educational and flagged 9,294 as malicious, across 14 malware families [[15]]; those are a language model’s labels, not code analysis. Kaspersky’s 2025 reporting on the GitVenom campaign described hundreds of fake projects with polished README files, inflated commit histories, and malicious payloads hidden across Python, JavaScript, C, C++, and C# code [[16]]. Check Point’s reporting on the Stargazers Ghost Network was even more direct: a network of more than 3,000 GitHub accounts used stars, forks, watches, and other activity to make phishing and malware repositories appear legitimate [[12]]. The service was advertised on dark-web forums at 100 stars for USD 10, the rate of the cheapest storefront He and colleagues listed. They later used that network as a test case: StarScout found 81% of its 847 repositories.

That is why fake popularity on GitHub matters differently from fake popularity on Instagram. On Instagram, the counterfeit product is attention. On GitHub, the counterfeit product is trust in code.

The same basic bundle keeps recurring:

  • a plausible repository topic
  • a polished README
  • tags chosen to catch search traffic
  • signs of activity or apparent history
  • inflated engagement signals
  • some external lure or malicious payload waiting behind the facade

Not every malicious repository needs fake stars, and not every repository with fake stars is malicious. But the overlap is strong enough that treating stars as mere vanity misses the supply-chain and malware angle completely.


The Structural Problem

What makes this irritating is not the moral quality of the buyers. There will always be people willing to cheat a public metric. That is the least surprising fact in the world.

The real problem is that platforms keep exposing low-cost trust signals and users keep over-reading them.

GitHub did not invent this pattern. Every platform that publicly ranks, recommends, or summarises social approval acquires an economy devoted to counterfeiting that approval. Campbell’s law and Goodhart’s law are waiting in the lobby. The difference is that on a code-hosting platform, the downstream object is not a post or a lifestyle brand. It is software, scripts, dependencies, proof-of-concept exploits, build tooling, installers, and copied snippets that may end up running inside real systems.

Once you see it that way, the question “Do fake stars boost downloads?” is too narrow. The more important questions are:

  • Do they help a malicious repository survive first contact with a human reader?
  • Do they help it look non-ridiculous long enough to earn one click, one clone, one copied command, one meeting, one investor glance, one journalist mention?
  • Do they make a fraudulent project blend into the background noise of normal open-source activity?

That threshold is much lower than mass adoption. It only has to work sometimes.


The Joke Is Still Wrong

So yes, buying GitHub stars is pathetic. But that is still the least interesting thing about it.

The embarrassing part is real. The boring part is the actual story. There is a public, low-friction market for faking a trust signal that developers visibly use as a shortcut. The latest evidence suggests that signal may not reliably translate into downstream downloads, which is worth saying plainly. It does not need to. Counterfeit social proof is still useful for laundering credibility, and on GitHub credibility attaches to things people execute.

That is why this is not just a story about insecure founders buying vanity metrics. It is a story about code hosting inheriting the same fraudulent attention economy as every other platform, except here the object being sold is trust in executable artifacts. That is not ridiculous. That is a security problem.

Literature and mutable vendor evidence checked through 2026-10-28. The cited field experiment does not test every first-impression or targeted-attack pathway, and no source here attributes a specific compromise to a specific purchased star count.


References

[1] Borges, H., & Valente, M. T. (2018). What’s in a GitHub Star? Understanding Repository Starring Practices in a Social Coding Platform. Journal of Systems and Software, 146, 112–129. DOI: 10.1016/j.jss.2018.09.016. Preprint: arXiv:1811.07643

[2] Borges, H., Hora, A., & Valente, M. T. (2016). Predicting the Popularity of GitHub Repositories. In Proceedings of the 12th International Conference on Predictive Models and Data Analytics in Software Engineering (PROMISE 2016), pp. 1–10. DOI: 10.1145/2972958.2972966. Preprint: arXiv:1607.04342

[3] He, R., Ye, H., & Zhou, M. (2024). Revealing the value of Repository Centrality in lifespan prediction of Open Source Software Projects. Preprint, arXiv:2405.07508

[4] He, H., Yang, H., Burckhardt, P., Kapravelos, A., Vasilescu, B., & Kästner, C. (2026). Six Million (Suspected) Fake Stars on GitHub: A Growing Spiral of Popularity Contests, Spams, and Malware. In Proceedings of the IEEE/ACM 48th International Conference on Software Engineering (ICSE ‘26), pp. 1469–1481. DOI: 10.1145/3744916.3764531. Preprint: arXiv:2412.13459; its first version (December 2024) was titled “4.5 Million (Suspected) Fake Stars in GitHub”.

[5] Toulas, B. (2024, December 31). Over 3.1 million fake “stars” on GitHub projects used to boost rankings. BleepingComputer. https://www.bleepingcomputer.com/news/security/over-31-million-fake-stars-on-github-projects-used-to-boost-rankings/

[6] McMahon, K. (2023, October 23). The GitHub Black Market That Helps Coders Cheat the Popularity Contest. WIRED. https://www.wired.com/story/github-stars-black-market-coders-cheat/

[7] Marlow, F. (2023, March 16). Tracking the Fake GitHub Star Black Market with Dagster, dbt and BigQuery. Dagster blog. https://dagster.io/blog/fake-stars

[8] Followdeh. Buy Github Stars (product page, Internet Archive capture of 2026-04-18) and FAQs (as served on 2026-10-28). Vendor pages; mutable. Capture; https://www.followdeh.com/en/faqs

[9] Baddhi Shop. GitHub Repository Promotion & Developer Engagement Growth Service (product page, Internet Archive capture of 2026-04-18; retitled “Buy GitHub Followers, Stars, Forks and Watchers” by 2026-10-28) and Refund Policy (as served on 2026-10-28). Vendor pages; mutable. Capture; https://baddhi.shop/product/buy-github-followers/; https://baddhi.shop/refund-policy/

[10] PlayerUp. Github Stars for Sale - Buy & Sell (archived 2025-12-06) and Middleman: Seller Payment Methods (archived 2026-01-18). Internet Archive captures. Listing; payment methods

[11] GitHub. GitHub Acceptable Use Policies, section 4, “Spam and Inauthentic Activity on GitHub”. GitHub Docs, accessed 2026-10-28. https://docs.github.com/en/site-policy/acceptable-use-policies/github-acceptable-use-policies

[12] Terefos, A. (2024, July 24). Stargazers Ghost Network. Check Point Research. https://research.checkpoint.com/2024/stargazers-ghost-network/

[13] Shen, L., Sood, G., & Weitzel, D. (2026). Social Proof is in the Pudding: The (Non)-Impact of Social Proof on Software Downloads. Journal of Online Trust and Safety, 3(2). DOI: 10.54501/jots.v3i2.286. Preprint by Shen and Sood: arXiv:2603.07919

[14] El Yadmani, S., The, R., & Gadyatskaya, O. (2023). Beyond the Surface: Investigating Malicious CVE Proof of Concept Exploits on GitHub. Preprint (first version 2022), arXiv:2210.08374

[15] Masud, M. R., & Faloutsos, M. (2024). Unveiling A Hidden Risk: Exposing Educational but Malicious Repositories in GitHub. Preprint, arXiv:2403.04419

[16] Kucherin, G., & Godinho, J. (2025, February 24). The GitVenom campaign: cryptocurrency theft using GitHub. Securelist (Kaspersky). https://securelist.com/gitvenom/115528/


Changelog

  • 2026-04-17: Initial publication. Vendor and payment-method references were checked against publicly visible pages or archived captures available on that date; those pages are mutable and may change or disappear.
  • 2026-07-11: Added the current literature cutoff and clarified that the field experiment does not establish the broader security impact of counterfeit repository credibility.
  • 2026-10-28: Corrected the citations of the two central papers: He et al. appeared at ICSE 2026, not as a 2024/2025 preprint, and its sixth author is Christian Kästner, not “Kastner”; Social Proof is in the Pudding appeared in the Journal of Online Trust and Safety with a third author, Daniel Weitzel. Corrected the storefront wording, which the cited pages no longer carry, to what their April 2026 archived captures show (“real engagement”, instant delivery, a 30-day guarantee, not “real” users and refill guarantees), and restated the payment methods source by source: cards, bank transfer and UPI appear only on a marketplace’s page for paying its sellers. Replaced “a non-trivial share” of malicious proof-of-concept repositories with the measured 899 of 47,285 (1.9%), and noted that the educational-repository labels come from ChatGPT. Added: Borges and Valente’s reasons for starring and the 567 of 777 who consult star counts; He et al.’s method and counts (six million suspected fake stars, 18,617 campaign repositories, at most 1% of monthly stars but 16.66% of repositories gaining at least 50 stars in July 2024), the first preprint as the source of the widely reported 3.1 million, the inferred part of the phishing link, the lower-bound reach figures, and StarScout’s 81% recall on the 847 repositories of Check Point’s network of more than 3,000 accounts; star prices from Dagster, He et al. and Check Point; GitHub’s Acceptable Use Policies, with evidence that removals come late; the field experiment’s design, its two group medians (20 and 69 added stars, at their peak, before the bought stars were removed) and its authors’ caveats; and Fig. 1, which maps each source to the stage of a developer’s choice it measures. References are now one numbered list in order of first citation.