I see three different changes to the splitter behavior. 1] The memory is reduced from [-5,5] to [0,1], I approve. 2] The memory is now separate for each lane instead of shared, this alone makes splitter sorters impossible and also forces compressed inputs to alternate between lanes on each output belt, I approve. 3] Alternation between output belts is now independent of item type, I disapprove.
And, initial output still defaults to the same belt every time as it did before, I disapprove though this is mostly aesthetic.
"To simply alternate every item to each output belt while considering all items to be the same type, that is the worst behavior, there's no reason to do that. ... Either it should be entirely random, or it should alternate each type between each output belt (then only the first of each type should be random). I'd rather have the latter; why introduce this random element when you can easily alternate between outputs, producing an even distribution? Ultimately, a random output will trend toward an even distribution anyway, but is more, well... random in the short term.
I'm okay with the way it works now. Though, a memory of 5-6 per type instead of 1 per type is not a very consequential aspect."
viewtopic.php?f=18&t=19114&hilit=belt&start=40#p169462
It looks like I reached a similar conclusion long ago, that the intended benefit would be achieved if the number of items remembered for each type was only 1 for each outgoing belt. And what you've changed it to, alternating without respect to item type, is what I once said was even worse than just making it random. (Because, at least random still prevents any patterns in a mixed input from manifesting as artifacts in the belt split.) Sure you still have the same number of items on each belt, but it's inferior for it to be independent of type like that. It was made this way in the first place for a reason. In particular, it is superior when you have mixed items on a lane. And making it random, simply trends in the long term to what having a memory of 1 for each type would
additionally achieve in every short term.
There is also another, unrelated change with this patch which "fixes" this somewhat cosmetic artifact:
viewtopic.php?f=18&t=19114&hilit=belt#p123107
but it's a totally separate change and though it was an inconsequential problem, this fix can be applied to the old per-type behavior too. This change, is making each lane have separate memory, instead of a shared memory for the entire belt. This also makes splitter sorters impossible, even if it were to retain the type-specific alternation.
So in summary, my vote is for reinstating the old splitter behavior, but have it only remember which belt the last item of each type was sent to, rather than the last 5 to 6 items of each type (depending on which output belt). And randomize between the two belts separately for each type when the splitter is first placed, instead of defaulting every type to the same belt. And, make this memory separate for each lane instead of shared between the two, so that we don't get aesthetically offensive behavior as in the final link (nor splitter sorters). *(That is, this makes splitter sorters impossible, DESPITE the sorters remembering which belt each type was last sent to.)
Relative to the new behavior, this means alternating between output belts for each separate item type, and randomizing the first choice for each one (the latter isn't important, it's just more pleasing).