Last-conflict search and conflict-ordering search can be used with a fall-back heuristic such as Min-Dom.

In conflict-ordering search, once all the variables have been branched on at least once, the fall-back heuristic will never be used again.

In last-conflict search, once all the variables have been branched on at least once, the fall-back heuristic will never be used again.

Last-conflict search and conflict-ordering search are value selection heuristics.

Conflict-ordering search always generates a smaller search tree than last-conflict search.

Select a value that has the smallest lower-bound propagation impact.

Select an unfixed variable involved in the largest number of constraints.

Select an unfixed variable with the smallest domain size.

Select an unfixed variable with the largest domain size.

Select an unfixed variable having the lowest propagation activity.

Select a random unfixed variable.

Select an unfixed variable involved in the smallest number of constraints.

If the variable selection heuristic is a dynamic first-fail Min-Dom but the value selection heuristic is static (it selects the minimum value of the selected variable on the left branch and removes it on the right branch), then by switching from a forward-checking filtering for AllDifferent to a domain-consistent filtering the number of nodes explored can increase, when searching for all the solutions (although it is unlikely with a stronger filtering).

If the variable selection heuristic is static (it selects the leftmost unfixed variable in [X1,X2,...,X10]) and the value selection heuristic is also static (it selects the minimum value of the selected variable on the left branch and removes it on the right branch), then by switching from a forward-checking filtering for AllDifferent to a domain-consistent filtering the number of nodes explored never increases, when searching for the first feasible solution.

If the variable selection heuristic is static (it selects the leftmost unfixed variable in [X1,X2,...,X10]) and the value selection heuristic is also static (it selects the minimum value of the selected variable on the left branch and removes it on the right branch), then by switching from a forward-checking filtering for AllDifferent to a domain-consistent filtering the number of nodes explored never increases, when exploring the complete search tree.

If the variable selection heuristic is static (it selects the leftmost unfixed variable in [X1,X2,...,X10]) and the value selection heuristic is also static (it selects the minimum value of the selected variable on the left branch and removes it on the right branch), then by switching from a forward-checking filtering for AllDifferent to a domain-consistent filtering the number of nodes explored can increase, when searching for the first feasible solution (although it is unlikely with a stronger filtering).

If the variable selection heuristic is a dynamic first-fail Min-Dom but the value selection heuristic is static (it selects the minimum value of the selected variable on the left branch and removes it on the right branch), then by switching from a forward-checking filtering for AllDifferent to a domain-consistent filtering the number of nodes explored never increases, when exploring the complete search tree.

If the variable selection heuristic is static (it selects the leftmost unfixed variable in [X1,X2,...,X10]) and the value selection heuristic is also static (it selects the minimum value of the selected variable on the left branch and removes it on the right branch), then by switching from a forward-checking filtering for AllDifferent to a domain-consistent filtering the number of nodes explored can increase, when searching for all the solutions (although it is unlikely with a stronger filtering).

Once the variable to branch on has been selected, it is important to select as value for the leftmost branch one that has the highest chance to lead to success.

Phase saving is a learning scheme for the value selection heuristic that tries first the last value that was tried without failing.

The value selection heuristic should mimic a greedy algorithm in order to find good solutions early.

Selecting a value that has the least impact on the objective bound should lead to a good first solution.

The value selection heuristic is not important for satisfaction problems.

Once the variable to branch on has been selected, it is important to select as value for the leftmost branch one that has the highest chance to lead to failure, in the spirit of the first-fail principle.