When a Mastodon post in late May 2026 alleged—without evidence—that rsync's recent regressions were caused by Claude commits, it ignited exactly the kind of internet pile-on that open-source maintainers have learned to dread. The claims spread to Hacker News and spawned a GitHub issue titled "Please Do Not Vibe Fuck Up This Software" with over 350 comments ranging from heated criticism to outright threats against the project maintainer. But now someone actually ran the numbers, and the data tells a very different story than the mob wanted to hear.

The rsync Outrage Machine

What started as an evidence-free Mastodon post pointing to a spurious correlation between regressions and Claude-assisted releases somehow exploded into a full-blown harassment campaign. The GitHub issue attached nothing but a screenshot of that original critique—no bug reports, no technical analysis, just vibes. As is typical with anti-AI sentiment online, the thread eventually escalated beyond words: one user posted drawings depicting violence against the rsync maintainer. Meanwhile, Hacker News commenters crowed about how this "finally proves" LLMs can't be used safely, ignoring the complete absence of actual data.

How One Developer Actually Tested the Claim

Rather than arguing on vibes alone, a developer decided to run an empirical analysis using bugs per 10 commits (bugs/10c) as the metric across all rsync releases from v2.4.6 through v3.4.3—a span of 46 releases with bug data sourced from GitHub issues, the rsync Bugzilla instance, and the mailing list. The methodology used an exact permutation test to determine whether Claude-assisted releases fell outside what could be expected from random chance in the historical distribution. All scripts were written by GLM 5.1, but every number was templated directly from the Python analysis script to eliminate any possibility of hallucination or data doctoring.

What the Statistical Analysis Actually Found

Only two rsync releases contain Claude commits: v3.4.2 (9 Claude commits, 0.80 bugs/10c) and v3.4.3 (28 Claude commits, 6.76 bugs/10c). Both fall inside the middle 50% of the historical distribution—the interquartile range spanning 0.65 to 6.82 bugs/10c. The exact permutation test yielded a p-value of 46%, meaning if you randomly selected any two releases from rsync's history, you'd get results as bad or worse nearly half the time. Fisher's exact test comparing Claude releases against the historical median produced an even more decisive result: p-value of 74% with an odds ratio of 1.06—essentially 1:1.

The Historical Mean Actually Favors Claude

Here's the kicker that the angry mob never bothered to check: the historical mean bug rate is 7.59 bugs/10c, while the Claude-assisted releases averaged just 3.78 bugs/10c—less than half the historical baseline. One commenter on Lobste.rs correctly identified what was needed: "It'd be interesting if someone actually did a timechart of regressions after each release." That chart now exists, and it demolishes the narrative. The runs test (p=0.123) confirmed no temporal clustering in bug rates—the sequence is statistically consistent with randomness throughout rsync's entire history.

The Pre-Claude Outlier Nobody Mentioned

The worst release in rsync's recorded history wasn't touched by Claude at all: v3.4.1 shipped with 113.33 bugs per 10 commits—102 bugs across just 9 commits—a hotfix released the day after v3.4.0 that exceeded every other release by an order of magnitude. And yet there was no GitHub issue with hundreds of comments, no death threats, no demands to fork the project. A maintainer shipped a broken release and fixed it quietly, as happens constantly in software development. The only thing that made v3.4.3 "special" was the availability of an enemy everyone had already decided to hate.

Key Takeaways

  • Both Claude-assisted releases (v3.4.2 and v3.4.3) fall inside the middle 50% of rsync's historical bug distribution
  • Exact permutation test p-value of 46% means random chance explains the results nearly half the time
  • The actual Claude mean (3.78 bugs/10c) is less than half the historical mean (7.59)
  • Fisher's exact test shows no statistical difference in above-median likelihood for Claude releases
  • v3.4.1—113.33 bugs/10c—was far worse than any Claude release and predates AI assistance entirely

The Bottom Line

The rsync community got caught up in a witch hunt based on correlation masquerading as causation, and the data exposes exactly how baseless that hysteria was. This is what happens when internet mobs decide to be angry before being informed—and it's a cautionary tale about how quickly the open-source ecosystem can turn toxic when people have an enemy they already want to blame.