Beyond single-shot evaluation: Measuring LLM capabilities through collaborative iteration - View it on GitHub