Although backtracking implementations only give an exponential guarantee in the worst case, they provide much greater flexibility and expressive power.
If the algorithm has learned the new constraint, it will backtrack from this solution, while the original backtracking algorithm would do a subsequent search.
