GitHub Trending Rewards Bookmarks, Not Code

What happened

GitHub Trending on June 26, 2026 surfaced its predictable top three: freeCodeCamp/freeCodeCamp at 437,900 stars; vinta/awesome-python at 286,700 stars; and openclaw/openclaw at 283,100 stars. The first two have been parked near the top of Trending for the better part of a decade. The third — a self-described 'personal AI assistant' project advertised with a lobster emoji and the tagline 'The lobster way 🦞' — accumulated its star count in roughly the last few weeks.

These three repositories share nothing technically. freeCodeCamp is a curriculum platform with a full LMS attached. awesome-python is a curated list of hyperlinks in a single markdown file. openclaw is, depending on how charitable you feel, a tongue-in-cheek wrapper around existing AI APIs or an outright meme. What they share is structural: none of them are libraries you would install into a production codebase. They are things you would, instead, bookmark.

That distinction — bookmark versus dependency — is increasingly the only thing GitHub Trending measures.

Why it matters

The star button has always been ambiguous. Is it an endorsement, a favorite, or a bookmark? GitHub itself has never picked one. The 2023 launch of Lists — explicit user-curated collections of starred repos — was a quiet admission that stars had drifted toward 'things I want to find again later.' The product team built a feature on top of that drift. The Trending algorithm did not get the memo.

Trending still treats each star as a vote of confidence, weighted by velocity. It cannot tell the difference between 'I'm going to come back and read this' and 'I'm shipping this in prod next sprint.' The result is a leaderboard that systematically rewards anything that resembles a saved bookmark: reference lists, learning resources, cheat sheets, every awesome-* repo ever published, and meme projects with a strong README. Actual libraries — the ones senior engineers depend on — almost never trend.

A few comparison points worth chewing on. `tj-actions/changed-files`, used in tens of thousands of CI pipelines, sits around 1,800 stars. Fastify, which serves a meaningful slice of Node.js production traffic, has 33k stars after nine years. Hono, a serious modern web framework with real adoption inside Cloudflare and Deno deploys, is around 22k. Compare those numbers to awesome-python's 286,700 — a single markdown file of links. The ratio between 'what is starred' and 'what is depended on' has been broken for years; Trending amplifies the broken signal weekly, then funnels it into hiring decisions, RFP shortlists, and 'should we use this?' Slack threads.

The openclaw case is the cleanest illustration. A repository whose entire description reads 'Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞' should not, by any defensible measure, accumulate 283k stars in weeks unless something other than software quality is being rewarded. The lobster repo trends because the lobster repo trends. It is a Schelling point — people star it because they see other people starring, and the velocity feedback loop closes. Trending becomes a meme stock chart with a green commit graph underneath.

This isn't a moral failure of GitHub users. Star-as-bookmark is a perfectly rational behavior on a platform that gives you no other native way to save a repo for later. The failure is treating that bookmark like an endorsement when surfacing repositories to twenty million developers a day. Every developer who uses Trending as a discovery surface is taking a signal that means 'people pressed a button' and interpreting it as 'people use this.'

The second-order effect is uglier. Once Trending is known to reward bookmark-bait, the supply side responds. Awesome-list inflation, README-as-marketing, screenshot-heavy repos that wrap thin libraries, and outright star-farming rings are all rational responses to a ranking system that conflates curiosity with adoption. We've watched the same arms race play out on every other reputation system on the internet. There is no reason to think GitHub is special.

What this means for your stack

Three practical takeaways.

First, stop using star count as a quality signal for dependencies. If you are doing due diligence on a library — whether it should land in your service mesh, your build pipeline, or your transitive dep graph — stars are downstream of marketing, virality, and demo-GIF quality. They are not downstream of code health. The signals that actually correlate with maintainability are boring: issue close rate, time-to-merge for outside PRs, distinct contributors in the last 90 days, release cadence, security advisory response time, and the number your package manager already gives you for free (weekly downloads on npm, pip, crates, packagist). The registry has a more honest signal than the host.

Second, discount Trending entirely as a discovery surface for serious tooling. Trending is a cultural attention feed. That has real value — it tells you what other developers find interesting enough to save. But conflating 'interesting enough to bookmark' with 'battle-tested enough to deploy' is the mistake. For serious dependencies, the better surfaces are language-specific indexes ranked by reverse dependencies (libraries.io still works), the OpenSSF Scorecard, Snyk Advisor, and the CNCF Sandbox/Incubation tracks. None are as fun as scrolling Trending. None are lying to you about what is real.

Third, if you maintain open source, recognize what the market is rewarding. Bookmarkable repos — README-heavy, screenshot-heavy, 'awesome-*' framing, learning resources — accrete stars. Production libraries do not. If your goal is genuine adoption (downloads, dependencies, deployed instances), star-chasing actively misallocates your effort toward the wrong artifacts. If your goal is hireability and personal brand, star-chasing is fine — just recognize you are optimizing for distribution, not for impact.

Looking ahead

GitHub knows about this drift. The Trending algorithm has been quietly retuned at least twice in the last decade — once after the great awesome-list explosion of 2017, once after a brief spree of AI-generated repo farms in 2024. Each tweak added more time-decay weighting and more author-diversity penalties. Each one made Trending feel less broken for a few months, then the equilibrium reasserted itself. The underlying problem isn't a bug in the ranking; it's that the input signal — the star — measures the wrong thing, and no amount of clever weighting can extract endorsement information from a button that most users now press for bookmarking.

The clean fix is structural: split the button. One action that drives Lists and personal libraries. A separate action that drives Trending and social reputation. GitHub has resisted this for fifteen years on UX grounds — adding a second button next to Star would, the argument goes, dilute the social signal and confuse new users. Maybe. But the current state, where Trending is increasingly a Pinterest board of curricula and lobster jokes, is harder to defend with each passing meme. Until the input changes, the leaderboard will keep telling you what to read on the train, not what to put in your `package.json`.