As AI coding agents approach human-level performance on existing benchmarks, the research community faces a critical challenge: how do we continue measuring progress when current evaluation suites are...