Your question is Benchmarking Reasoning Improvements. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How do you design evaluation benchmarks to ensure a model’s reasoning capabilities are actually improving?