I’m not sure but I’d guess that step 1 is usually too difficult. So you can either: - produce short lists of candidates, in which case your pruning will often

I’m not sure but I’d guess that step 1 is usually too difficult. So you can either: - produce short lists of candidates, in which case your pruning will often end up with no solution at all because you’re missing the correct answer in your list of candidates for some of the clues. - produce long lists of candidates, in which case your pruning won’t converge to a unique solution. Using probabilistic approaches, I think you can - include the possibility that a given clue’s solution isn’t on your list at all (i.e. your solver will occasionally deduce a solution from crosses only) - use the probabilities of your other candidates list to have a better chance of guessing these unknown words and of choosing the right solution when your pruning doesn’t give a unique solution.