livenerf open source: uses pre-registered benchmarks to long-term detect whether Claude Opus 5.5 is weakened after release
A long-running, deterministic-as-possible benchmark for detecting whether a frontier model gets quietly worse after launch. !
Return to original text
A short introduction is retained here; the full text from the source website is not reposted. For specific wording and subsequent updates, please refer to the original article.
Open github.com original