the subjective framework is exactly why its good prior bms relied mostly on unit tests or synthetic judges which are easily benchmaxxed, which leads to nobody

the subjective framework is exactly why its good prior bms relied mostly on unit tests or synthetic judges which are easily benchmaxxed, which leads to nobody trusting benchmarks we need people manually checking the data for good code quality