The scenario splits users between different assistant versions and answers each with their version, to fairly compare which one performs better.