readme-breakouts

"Breakout" here means a public, non-fork repo created in the 365 days up to 2026-09-30 with at least 1,000 stars that day. The search found 3,806. The dataset is a stratified sample of 100, with every detector rule, hand check and limit written down in METHOD.md. The check command reads one README and shows, next to each feature, how many of the 100 sampled READMEs have it. It needs Python 3.10 or newer and nothing else.

Population estimates, with 95% intervals: a one-line pitch within 15 lines 80% (70 to 87), an install or run command within 40 lines 13% (7 to 22) and anywhere 62% (52 to 72), badges 66% (55 to 75), a limitations heading 3% (1 to 10). The median README is 13,663 characters.

The data folder holds the population (3,806 repos with stars, forks, dates and licence), the sample, the detector output, a summary and the hand checks. README texts are not included, because they belong to their authors, and no repo descriptions are published.

It describes one sample of 100 from one day, weighted towards AI tooling, and does not show that a feature causes stars. Faster and slower climbers did not differ on any feature after correction. It does not test for fake stars. The detectors read one line at a time and recognise just package-manager style install commands.