Keep the level data as a simple 1‑D array of tile structs. Each tile only stores its type and maybe a collision flag. On load, walk the array once, build a flat list of active colliders and store the world start/finish coordinates. During the game loop, just loop over that list, skipping any tiles that are already behind the player. That means you’re only touching O(n) tiles per frame, no dynamic allocation or pathfinding. Use a tight update loop: update position, check collisions, then render. Keep the physics simple—just a fixed timestep and a single velocity vector. If you need to add optional “doorway” checks, put them in a separate small function that only runs when the player is near the doorway index. No extra logic for the rest of the level. Result: linear flow, no wasted checks, and you avoid runtime hiccups because the system never goes back over the same tiles.