Statistical Analysis Finds No Evidence Claude Introduced Bugs into rsync

After a Mastodon post in late May 2026 ignited a viral campaign accusing rsync’s maintainers of degrading the tool through Claude-assisted development, an independent researcher ran the empirical test that critics never bothered to perform. Using severity-weighted bugs per 10 commits across 36 releases from v2.4.6 to v3.4.3, the analysis examined where the two Claude-assisted releases fall within the historical distribution of release quality. The methodology was designed in consultation with a statistician, and all figures are templated directly from a reproducible Python pipeline to rule out LLM hallucination of numbers.

The results undercut the outrage narrative. The two Claude releases bracket the interquartile range in opposite directions — v3.4.2 sits below it, v3.4.3 above it — and neither qualifies as an outlier. An exact permutation test returns a p-value of 46%, and Fisher’s exact test gives 74% with an odds ratio of 1.06, meaning Claude-assisted releases are statistically indistinguishable from any other pair of releases. The historical mean severity rate is actually 1.8× higher than the Claude releases’ mean. A pre-Claude release, v3.4.1, is the real outlier in the dataset.

The piece also documents the social dynamics: an evidence-free post spread to Hacker News and culminated in a GitHub issue titled “Please Do Not Vibe Fuck Up This Software” with 350+ comments, including deleted threats and violent imagery directed at maintainers. Calls on Lobste.rs for someone to actually chart the regression data went unanswered until this analysis. The data does not support the claim that Claude made rsync worse.