I see your points - I'll certainly use clpfd more, in future.

Also, I was mistaken - my peano-ish solution worked in both directions, though it had probable performance issues (perhaps because I didn't know `dif/2` was more flexible than `\=`) and then had trouble when I tried to map back from peano to naturals. I used a fun and perhaps misleading form to try for brevity:

  rle([], []).
  rle([X], [[1, X]]).
  rle([X,X|XS], [[N+1,X]|ZS]) :- rle([X|XS], [[N,X]|ZS]).
  rle([X,Y|XS], [[1,X]|ZS]) :- rle([Y|XS], ZS), X\=Y.

And so:

  ?- rle([7,7,7,8,8,9], Y).
  Y = [[1+1+1, 7], [1+1, 8], [1, 9]];
  false.
  ?- rle(X, [[1+1+1, 7], [1+1, 8], [1, 9]]).
  X = [7, 7, 7, 8, 8, 9];
  false.

Edit: If one cared for safety/compatibility, each [1, X] would be [0+1, X].