Personally I don’t use a random method at all. Such techniques will cause Refinery to be unable to optimization as there is no correlation between seed 1 and seed 2.
Instead use number sliders to shift the distances and directions rather than building a list of all possible points, then move objects from the origin.
Alternatively you can look into list permutations which at least maintain some similarities between permutation 1 and 2. Eventually these do fall off a cliff with no correlation at permutation
(n-x)!. As a result this method may not always find a global optimum, but you generally finds a local optimum or two in most decent runs. Limiting your lists to 12 items or less certainly helps in that front.
Note that in either case of list alteration (random or permutations) you are exploring contents in factorial design space, so the numbers will rapidly eclipse the limit of integer sliders in Dynamo (happens in design spaces as small as 13 items long), but with some creative python work you can circumvent that limit.