A repository with 100,000 stars that gained them over 10 years is fundamentally different from one that gained 50,000 stars in the last month. Raw star count measures accumulated history, not current relevance.

Traditional GitHub trending lists use a 24-hour or weekly star window — useful, but gameable and noisy. We take a different approach.

Attention Score combines multiple signals into a single 0–100 score, normalized within each category so you're always comparing like with like:

The score is computed fresh on every query — it never goes stale even as new repos enter a category and shift the distribution.

It's not a quality score. A repo can have a high Attention Score because it went viral for controversy, or a low score because it's a mature, stable project that no longer needs stars to prove its value (like Linux). The score measures current attention, not merit.

It's also not gameable by buying stars — star velocity spikes from bought stars look exactly like organic viral growth, but they don't come with corresponding fork/watcher increases, making them easy to spot in the full signal picture.

We pull fresh data from the GitHub API weekly, storing snapshots in our database to power the star growth charts on each repo page. Category rankings update on the same schedule.

All data is sourced directly from the GitHub public API. We don't scrape, infer, or estimate any metrics.