← Journal

June 6, 2026

A developer ran a severity-weighted permutation test across every rsync release to measure whether Claude-assisted commits introduce more bugs than the baseline. The methodology is worth reading if you've wondered how to actually evaluate AI coding quality at small sample sizes. Elsewhere, Vercel Sandbox now supports persistent Drives in private beta, solving the stateless problem for workflows that need storage to outlive a single run. LlamaFactory v0.9.5 also landed with Qwen3, Gemma4, and Transformers v5 support for anyone running fine-tuning pipelines.

June 6, 2026 · wwwatch