ChainMap looks up keys by walking its maps in order, returning the first match. So ChainMap(request, user, defaults) reads as 'request takes priority, then user, then defaults'. This is the cleanest expression of layered configuration in the Python standard library. Iterating the chain returns each unique key exactly once, in first-occurrence order. The chain holds references to the underlying dicts (no copy), so updating any layer is immediately visible through the view.

Mutations on a ChainMap (write or delete) only affect the FIRST map in the chain. Deleting a key that exists deeper raises KeyError. This is what makes the chain safe to use as a per-request scope: writes inside the request never accidentally update shared defaults. The new_child() helper (config.new_child(extra)) creates a fresh chain with the new dict prepended, which is the canonical way to push a temporary scope without mutating the existing chain.

ChainMap's new_child() and parents attribute model nested scopes the way a programming-language interpreter does. Entering a scope is scope = scope.new_child(); leaving it is scope = scope.parents (which returns a chain of all but the first map). Writes inside a child scope shadow the parent without touching it; leaving the scope discards the child entirely. This is also how Python's own scope rules can be modelled in a teaching interpreter or a small DSL.