Increases the base count limit to 3000. By increasing the history, the change decreases the variance observed by the miners due to fluctuations in data. This also ensures that any outlier in both the prompt dataset and answers have less effect on future responses
Currently, at the limit of 1000 responses, we observe fluctuations are heavily impacted by outliers, which has lasting effects and increases the variance in rewards. This change will decrease the effects of outliers by 3x.
3000 responses at 80% response rate can be achieved most validators within 200-300 blocks; well within the expected lifetime of the validator. If we were to further increase the count limit, a saving mechanism should be implemented for the reward models to enable faster consensus.