Methodology
How the numbers on this site are produced, and what they can and cannot say.
Source
Everything derives from the GH Archive public event feed, ingested hour by hour into a local ClickHouse store. No provider APIs are used at query time, and nothing is sampled: every event in the retained window counts.
Events, not stars
Rankings measure event volume — pushes, issues, pull requests, comments, forks, and the rest — weighted equally. Star counts look authoritative but are unusable here: stars stopped being captured during 2026, so any star leaderboard would describe the gaps in the archive rather than GitHub. The warm-up window is excluded from every ranking for the same reason coverage there is unreliable.
Bots and noise
Every contributor query excludes bot accounts — names ending in[bot], -bot, _bot, or matching known automation accounts — plus empty logins. Forecasts are honest about their uncertainty: the bands widen the further out the prediction reaches.
Freshness
The landing page board, categories, and organization table are precomputed on a daily refresh cycle, so they move once a day. Repository, actor, and tape pages query live data bounded to the last 30–60 days. Every page carries a coverage note showing which days are actually loaded; where days are missing, the numbers say so instead of guessing.
Editorial verdicts
The best-in-category pages assign each pick a verdict — Adopt, Evaluate, or Watch — from its 30-day momentum score normalised across every candidate. The momentum is computed; the thresholds and the category blurbs are editorial judgment, and categories with no strong candidate are left empty on purpose.