Blog

Notes on building the benchmark, what the runs keep showing, and where agentic Apple development actually breaks.