LiveNerf: open benchmark watching Claude Opus 5.5 for post-launch nerfs
LiveNerf is a frozen-panel, append-only eval (Inspect AI + headless Claude Code) that scores Opus 5.5 daily for 30 days to detect quiet post-launch regressions. As of Sept 29 it’s on day 6 of 10 in the baseline window; first statistical call lands around Oct 24.
