EvalRouter now charts every benchmark run. Two models can post the same score and fail completely different tasks. A single number cannot show that. Each results page now opens with: - the outcome of every task, model by model - each score with its interval - score against