Iteratively updating the global proof plan when individual lemmas fail, rather than backtracking within a single proof path.