“The Claude Code team maintains its own set of evaluations, separate from Anthropic's company-wide model evals, to ensure product-specific performance.”