MeniPorat
Thank you for posting your solution. I found the same here before that: https://stackoverflow.com/questions/76065555/how-to-use-textsplit-on-a-column-from-a-dynamic-array-spill-range-in-excel (posted on April 20, 2023) (Some other workaround options can also be found there) and from: https://stackoverflow.com/questions/76985094/return-split-values-in-a-dynamic-range-and-include-blanks (Posted on Aug 26, 2023).
So I had that solution with me when I posted here. My suggestion was that either MAP-LAMBDA should support what REDUCE is already doing, so that DROP, VSTACK, etc. (which is a workaround solution) can be avoided. Therefore the fallacy is, if REDUCE supports the formula why not SCAN and MAP-LAMBDA? They work in a similar way. If TEXTSPLIT does not allow arrays as input why that behaviour is not replicated when using REDUCE?
Best solution of course would be not even using MAP-LAMBDA or SCAN but native support of arrays in TEXTSPLIT (and other similar functions), going forward, like = TEXTSPLIT(array,";") [";" refers to delimiter]