Main suggestion: change some of the live eras to validation.
There’s talk of dumping thousands more columns of data on us… but just releasing some of the target data (1/4 to 1/2, is what I suggest) would have much more impact on the average user’s CORR. That’s my guess, anyway. You could test this immediately (effective on round 276) with just about no preparation, and at no cost, and then perhaps decide to pass on buying 2,000 columns worth of new data.
Supplementary suggestion: after releasing those targets, and seeing the aggregate CORR go way up, your user base will very likely expand… which will push down the payout factor! To keep that from dis-incentivizing the best models: apply the 25% earn/burn cap after applying the payout factor. I believe that actually will make a difference, once the target data becomes abundant and current.