Complexity: sets, memoization, batching

Algorithmic wins: complexity beats micro-tuning

The biggest speedups come from changing the shape of the work:

• Membership on a list is O(n); on a set/dict it is O(1). Filtering a list against a large denylist with x in some_list is the classic quadratic accident. One set(...) call removes an entire order of magnitude.
• Memoization converts overlapping recursive work into a table lookup: naive fib(25) makes ~243k calls; the memoized version makes ~49. Same answer, different universe of work. functools.lru_cache(maxsize=None) is the standard tool; a dict keyed by arguments is the manual equivalent.
• Batching turns N round-trips into 1: fetching per-item from a database or API inside a loop (N+1 queries) collapses into one batched fetch.
• Early exit / bounds: stop scanning when the answer is decided; prefer any()/all() over building full lists of booleans.

Micro-tuning (attribute-lookup hoisting, local variable caching) yields small constant factors. Complexity fixes yield orders of magnitude. When both are available, take the complexity fix first — it also tends to be the more readable change.