Stage 1 calculates a freshness weight for each of the seven Rank Factor Classes based on how few times it has been tested, probabilistically selecting an under-explored class to prevent the optimizer from getting trapped in a single mechanism.