It seems that rather than having some threshold of predictability, they are just taking the three most likely characters. So even though "ca" is a very common d

It seems that rather than having some threshold of predictability, they are just taking the three most likely characters. So even though "ca" is a very common digraph, it isn't in the top 3 in the position you had it. The difficulty, since this is intended a user aid, seems to be in balancing soundness (using a threshold) with user experience (sometimes predicting 3 letters, sometimes predicting 10, which would be overwhelming).