J

If one line of code needs five lines of comment, I'm not sure how much of an improvement the "expressive power" is! But I learned how to use J's group-by operator (/. or /..) and a trick with evoke gerund (`:0"1) to transform columns of a matrix separately. It might have been simpler to transpose and apply to rows.

data_file_name =: '11.data'
data =: ". > cutopen fread data_file_name
NB. split splits an even digit positive integer into left digits and right digits
split =: ; @: ((10 & #.) &.>) @: (({.~ ; }.~) (-: @: #)) @: (10 & #.^:_1)
NB. step consumes a single number and yields the boxed count-matrix of acting on that number
step =: monad define
   if. y = 0 do. < 1 1
   elseif. 2 | <. 10 ^. y do. < (split y) ,. 1 1
   else. < (y * 2024), 1 end.
)
NB. reduce_count_matrix consumes an unboxed count-matrix of shape n 2, left column being
NB. the item and right being the count of that item, and reduces it so that each item
NB. appears once and the counts are summed; it does not sort the items. Result is unboxed.
NB. Read the vocabulary page for /.. to understand the grouped matrix ;/.. builds; the
NB. gerund evoke `:0"1 then sums under boxing in the right coordinate of each row.
reduce_count_matrix =: > @: (({. ` ((+/&.>) @: {:)) `:0"1) @: ({. ;/.. {:) @: |:
initial_count_matrix =: reduce_count_matrix data ,. (# data) $ 1
NB. iterate consumes a count matrix and yields the result of stepping once across that
NB. count matrix. There's a lot going on here. On rows (item, count) of the incoming count
NB. matrix, (step @: {.) yields the (boxed count matrix) result of step item;
NB. (< @: (1&,) @: {:) yields <(1, count); then *"1&.> multiplies those at rank 1 under
NB. boxing. Finally raze and reduce.
iterate =: reduce_count_matrix @: ; @: (((step @: {.) (*"1&.>) (< @: (1&,) @: {:))"1)
count_pebbles =: +/ @: ({:"1)
result1 =: count_pebbles iterate^:25 initial_count_matrix
result2 =: count_pebbles iterate^:75 initial_count_matrix