Did Claude increase bugs in rsync?

Recently, controversy arose in the rsync community that commits using AI assistance tools like Claude were undermining the software's stability. In response, a statistical verification was performed based on data from 36 releases, and no evidence was found that the introduction of Claude significantly increased the bug rate.

Key Analysis Results:

• Absence of Statistical Significance: According to the exact permutation test (p=46%) and Fisher's exact test (p=74%), the probability that Claude-assisted releases had more bugs than random releases was not statistically significant.
• Bug Severity Assessment: Using Qwen 3 35B, bug severity was scored on a 0-100 scale, and analysis was conducted using the resulting 'severity-weighted bugs/10 commits (sev/10c)' metric.
• Comparison with Historical Data: The v3.4.1 release, which predates the introduction of Claude, actually recorded the highest bug rate among all the data. Claude-assisted releases (v3.4.2, v3.4.3) fall within the middle range of the historical distribution.
• Scale of Changes: Claude-assisted releases had a much higher average number of changed lines (LOC), but neither the number of commits nor the severity-weighted bug count was abnormally high.

In conclusion, the claim that Claude degraded the quality of rsync is not supported by the data, and current AI-assisted development is statistically indistinguishable within the existing software development cycle.