BFS needs a FIFO queue: append to the end, popleft from the front. A Python list supports append in O(1) but pop(0) in O(n) because every other element must shift. deque makes both O(1), which is what makes a BFS on a million-node graph feasible. Always reach for deque when the algorithm needs a real queue; using a list 'because it works' is one of the most common Python performance bugs in graph code.

Setting maxlen=N on a deque turns it into a fixed-size ring buffer: appends past the limit silently drop the oldest entry from the other end. This is the cleanest implementation of 'last N values' for monitoring dashboards, moving averages, and chat-history caches. Both appendleft and append honour maxlen. The deque is bounded, so memory usage stays predictable regardless of how long the producer runs.

rotate(n) cyclically shifts every element: positive n moves entries to the right (with the tail wrapping to the front), negative n moves them to the left. Both directions are O(min(n, len)) which is essentially constant for the typical small rotation. This is the right primitive for round-robin scheduling, calendar shifts, and any sliding-window structure where the boundary is a wrap-around. Slicing a list to achieve the same effect would copy the whole list every iteration.