It works way simpler than you may think.

Lets' have:

; The Collection will purge the excessive records.
; 0 means 100000
collection_capacity = 100

; Sort the Collection by one of the following metrics:
; NetBalance, Profit, ProfitFactor, ReturnToDrawdown, RSquared, Stagnation, WinLossRatio
sort_by = Profit

When the Generator find new suitable strategy, it pushes it to the Collection.

The Collection puts in within the other strategies by finding its place according to the sorting criteria ("Profit" in this case).

Collection checks if the new strategy correlates with any of the other strategies. If it finds a correlation, it removes the one of the correlated strategies with the lower Profit.

If the count of the strategies reaches 101, the Collection removes the strategy with the lowest Profit. It maintains the maximum capacity of 100 strategies in that way.

If the Generator continues generating and pushing strategies, the Collection will continue sorting them per Profit and pruning the excessive once with the lowest Profit.

You see that the actual Collection Capacity doesn't matter too much because the strategies we want are near to the top.

...

I'm currently generating and validating strategies in three steps (With Express Generator).
1) - generate up to 1000 new strategies per market in a new Collection.
2) - pass this new collection via Monte Carlo validation and merge it the previous validated one. This collection has capacity of 30 (recently increased to 40) strategies.
3) - validate this collection with data from different broker (Eightcap, MetaQuotes, or Premium Data). I keep between 10 and 30 strategies of these collections.