PyPI census: 24.8% of popular packages have yanked a release

A pre-registered study by Munim, Inc. sampled 3,000 top-download and 3,000 random PyPI projects: 24.8% of popular projects had at least one yanked file versus 3.15% in the long tail. Yanked-file share was 1.50% versus 0.66%.

• Popular packages yank releases about 7.9x more often than the long tail
• Yanked-file share: 1.50% for popular projects, 0.66% for the tail
• Mean files per project: 370.9 in the head vs 24.1 in the tail
• Three of four pre-registered predictions held; the under-1% threshold failed