The editorial argues that trending is downstream of social media visibility and zero-cost click behavior, not actual usage or dependency. Stars don't expire, don't reflect usage, and are the easiest signal in the developer world to coordinate or bot — so any account with a few thousand followers tweeting 'star this' can saturate the page for an afternoon.
The editorial contends that production engineers who rely on critical software don't star things in coordinated bursts — they file tickets and move on. As a result, the genuinely useful infrastructure (e.g., a well-chosen Postgres pooler) is invisible to trending, while creator-economy pushes dominate the leaderboard.
A personal profile README repo trending at a 225-score position — cited by the editorial as evidence that the trending mechanic rewards social-graph visibility over software substance. It's a real person, not a scam, but it is not software anyone would install or depend on.
birds.cafe is a personal site repo trending at 786 — the editorial uses it as another example of trending surfacing personal projects rather than dependable software. The repo is real and lovely, but it's not something you'd install or learn from in the way trending implies.
learn-ai-practice is a Chinese-language AI tutorial collection trending at 176 — the editorial points to it as an example of trending elevating curated link lists and tutorial dumps. i5ting has shipped real OSS, but this particular trending repo isn't installable software, it's a learning resource for a specific linguistic community.
GitHub's trending page today is a museum of why the trending page is broken. Three of the top entries: PMTraderAdam/PMTraderAdam — a personal profile README — sitting at a 225-score trending position. kanavtwtgg/birds.cafe, a personal site repo, at 786. i5ting/learn-ai-practice, a Chinese-language AI tutorial collection, at 176. Yesterday's leaderboard had freeCodeCamp at 437.9k lifetime stars and a repo called 'openclaw' — a 283.1k-star 'personal AI assistant 🦞' that nobody in actual AI tooling has heard of — sharing the page with free-programming-books.
None of these are scams, exactly. PMTraderAdam is a real person. birds.cafe is a real (lovely) site. i5ting has shipped real OSS. But none of them are software you would install, depend on, or learn from in the way 'trending' implies you should. They're trending because trending is a function of star-velocity inside a follower graph, and any account with a few thousand followers who tweets 'star this' can saturate the signal for an afternoon.
The core mechanic: GitHub /trending ranks by stars-per-time-window, weighted by recency. That metric is downstream of exactly two things — how visible the repo is on social media, and how willing the audience is to click a button that costs them nothing. Stars don't expire. They don't reflect usage. They don't survive abandonment. And they are, demonstrably, the easiest social signal in the developer world to bot, swap, or coordinate.
Discovery has gotten worse, not better, in the last five years. The repos you actually want to find — the maintained, the load-bearing, the quietly-excellent — almost never trend, because the people who use them in production don't star things in coordinated bursts. A senior backend engineer who finally figured out the right Postgres pooler doesn't tweet 'star this.' She files a Jira ticket and moves on.
What trends instead is whatever a creator-economy GitHub account pushed this week. The incentive structure is the same one that turned product listings, app stores, and Amazon search into curated chaos: when a free, prominent ranking exists, optimizing for it becomes a job. There are now Fiverr listings for GitHub star packages. There are 'star-for-star' Discord servers. There's a cottage industry of 'launch your repo on Trending' newsletters. The signal-to-noise floor has been negotiated downward for years.
The community has noticed. The HN comments under repos that 'trend' overnight are increasingly dominated by some version of 'how does this have 40k stars and zero commits in 18 months?' The answer is usually: it doesn't, in any meaningful sense. It has 40k clicks on a button, distributed across an audience that will never run the code.
The fix isn't fixing trending. The fix is routing around it. Below are five signals that are harder to game because they're closer to actual use, actual money, or actual editorial judgment.
1. Package-registry install graphs. npm-stat, crates.io's recent-downloads chart, PyPI's BigQuery dataset, RubyGems' API. An install is a non-trivial action by a real machine; a star is a vibe. A library going from 200k to 2M weekly npm downloads in three months is telling you something real about what production systems are pulling in. Star counts can mislead by a factor of 100; install counts mislead by 2x at the outside.
2. Maintainer RSS, not project RSS. Subscribe to people, not projects. Simon Willison's feed will tell you about every interesting AI tool worth knowing about before /trending does. Drew DeVault's blog tracks the parts of the OSS ecosystem GitHub doesn't index well. Julia Evans surfaces debugging tools nobody markets. fasterthanlime, Matt Pocock, Dan Abramov, Rich Hickey's rare posts — pick fifteen maintainers in your domain and your RSS reader becomes a better trending page than GitHub's.
3. 'Show HN' and Lobsters /new. Not the front page — the new page. Show HN gets ~50 submissions a day; the ones with substantive comments from people who actually understand the domain are higher-signal than anything on Trending. Lobsters' tag-filtered feeds (lobste.rs/t/rust, /t/databases, /t/ai) are run by an invite-only community that flags low-effort content within minutes. The cost of submitting requires either karma or invite-graph standing, which makes the spam economics worse than just buying stars.
4. GitHub Sponsors recipients. People who voluntarily route money at a maintainer have made a judgment that stars cannot capture. Browse github.com/sponsors/explore filtered by category. The top-sponsored Rust maintainers, the top-sponsored security researchers, the top-sponsored Python ecosystem contributors — that's a list of people whose work is load-bearing enough that other engineers gave up dollars to keep it going. It's also a list that's nearly impossible to inflate, because Stripe is not as generous with refunds as GitHub is with star removals.
5. Domain-curated lists with editorial judgment. Papers With Code for ML. This Week in Rust for Rust. Awesome-Selfhosted (with the strict inclusion criteria) for infrastructure. The Pragmatic Engineer's deep-dives for industry tooling. State-of-JS / State-of-CSS surveys for frontend. A human gatekeeper with a domain reputation to protect is, empirically, the most spam-resistant ranking algorithm we have built. It's also the oldest. Hacker News' moderation, Lobsters' invite graph, and the editorial discretion of a single trusted curator have all outperformed engagement-weighted star counts by a wide margin for over a decade.
There is no programmatic fix here. The pattern repeats: any ranking that's free to influence and visible to a large audience becomes a marketing channel. The only durable countermeasures are gatekeeping (editorial, invite-graph, paid), or measurement of behavior that costs the rater something — installs, dollars, time-spent. Stars cost a click.
GitHub will not fix /trending. It's a feature that drives engagement and signups, and the gaming is a feature for some accounts, not a bug. Expect the next round of degradation to be AI-generated repos with AI-generated README diagrams trending alongside profile READMEs and Notion-templates-disguised-as-repos. The practitioners who got value out of /trending in 2015 have already moved on. The practitioners still relying on it in 2026 will need to as well — and the five signals above are a reasonable starting point.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.