This paper introduces a method for identifying and ordering landmarks in planning tasks. Landmarks are facts that must be true at some point in every valid solution plan. The authors extend the concept of reasonable orders between top-level goals to include sub-goals (landmarks) that arise during planning. They show how to find landmarks, approximate their reasonable orders, and use this information to decompose planning tasks into smaller sub-tasks. The methodology is domain- and planner-independent and has been implemented to improve the performance of state-of-the-art sub-optimal planning systems like FF and LPG. The key idea is to use landmarks to guide the search process, helping the planner focus on the most important steps first. Landmarks are facts that must be true at some point in any solution plan, and their reasonable orders can be approximated using a directed graph called the landmark generation graph (LGG). The LGG is used to decompose the planning task into smaller sub-tasks, which can be solved more efficiently. The authors show that deciding whether a fact is a landmark or determining the reasonable orders between landmarks is PSPACE-complete. They propose approximation techniques to handle this complexity, including the use of a relaxed planning graph (RPG) to estimate reachability and identify landmark candidates. The RPG is used to find the earliest actions that can achieve a landmark, and to determine the necessary orders between landmarks. The authors also introduce the concept of obedient reasonable orders, which are reasonable orders that are only valid if certain constraints are already in place. These orders are used to further improve the performance of sub-optimal planners by guiding the search process more effectively. The paper demonstrates that using landmarks and their reasonable orders can significantly improve the runtime performance of sub-optimal planners, even though it may result in slightly longer plans. The approach is shown to be effective in a variety of domains, including the Blocksworld and Logistics domains. The results show that the method can lead to significant runtime improvements, especially in complex planning tasks.This paper introduces a method for identifying and ordering landmarks in planning tasks. Landmarks are facts that must be true at some point in every valid solution plan. The authors extend the concept of reasonable orders between top-level goals to include sub-goals (landmarks) that arise during planning. They show how to find landmarks, approximate their reasonable orders, and use this information to decompose planning tasks into smaller sub-tasks. The methodology is domain- and planner-independent and has been implemented to improve the performance of state-of-the-art sub-optimal planning systems like FF and LPG. The key idea is to use landmarks to guide the search process, helping the planner focus on the most important steps first. Landmarks are facts that must be true at some point in any solution plan, and their reasonable orders can be approximated using a directed graph called the landmark generation graph (LGG). The LGG is used to decompose the planning task into smaller sub-tasks, which can be solved more efficiently. The authors show that deciding whether a fact is a landmark or determining the reasonable orders between landmarks is PSPACE-complete. They propose approximation techniques to handle this complexity, including the use of a relaxed planning graph (RPG) to estimate reachability and identify landmark candidates. The RPG is used to find the earliest actions that can achieve a landmark, and to determine the necessary orders between landmarks. The authors also introduce the concept of obedient reasonable orders, which are reasonable orders that are only valid if certain constraints are already in place. These orders are used to further improve the performance of sub-optimal planners by guiding the search process more effectively. The paper demonstrates that using landmarks and their reasonable orders can significantly improve the runtime performance of sub-optimal planners, even though it may result in slightly longer plans. The approach is shown to be effective in a variety of domains, including the Blocksworld and Logistics domains. The results show that the method can lead to significant runtime improvements, especially in complex planning tasks.