They write tests for it fine too, they actually work even better when you force them to.

Mind sharing your model choice and approach if you can’t get them to successfully test for counting the letters.