⌥ omp leaderboard

method

• Set. Every public repository with at least 20,000 stars, re-enumerated every 6 hours through GitHub search with star-range bisection. Currently — repositories: — ship executable software, — are lists, books, courses, interview drills, prompt dumps or asset collections.
• Star counts. Batched GraphQL reads: the full set every 3 minutes, oh-my-pi and its nearest neighbours every 30 seconds.
• Rates. GitHub exposes no star timeline for these credentials, so growth is measured here: an exponentially weighted linear regression of stars against time, weighting an observation of age a by exp(−a/12h). Until enough spread accumulates the estimate falls back to a lifetime average, and the label says which is in force.
• The axis. Vertical position is star count, so the distance between two repositories is the real gap between them. oh-my-pi rises through that space at its measured rate; the camera only follows once it nears the top.
• The ticking numbers. Counters are extrapolated between reads from the last measured value and its rate, then eased toward each new measurement. Whole stars are measured or already projected past; the decimals are the projection running between reads.
• Replay. The snapshot log records every change, so any instant in the retained window is reconstructed exactly: seed each repository at the window start, then apply changes in time order. Ranks in replay are recomputed against the whole set, not interpolated.
• Passes. An overtake is recorded only when measured counts cross, and only for repositories that were genuinely above oh-my-pi when first seen.

—