The evaluator is using the `TARGET_VALUE = 2.635` from the AlphaEvolve paper which might be considered cheating. Was anyone able to reproduce the result without including this?