Backtracking problems - my approach that works every time

After struggling with backtracking for weeks, I developed this mental framework:

1. What are my choices at each step?
2. What constraints must I satisfy? (when to prune)
3. What does the base case look like? (when am I done)

Then the template is always:

def backtrack(current_state, choices_left):
    if is_solution(current_state):
        save_result()
        return
    for choice in choices_left:
        if is_valid(choice):
            make_choice()
            backtrack(new_state, remaining_choices)
            undo_choice()

This worked for: N-Queens, Sudoku, Permutations, Subsets, Combination Sum, Word Search, and Palindrome Partitioning.

Hope this helps someone!