Cpython Parity

Monty exists so LLMs can write ordinary Python; silent divergence from CPython in a common idiom is the worst failure, because the model has no signal to write something else. When a change alters a builtin, dunder, method, or exception's result, type, message, or attributes, verify it matches CPython. A change that diverges (or widens an existing divergence) needs a matching entry under limitations/; a new or widened divergence with no limitations/ entry is a finding. Prefer flagging a plausibly-silent wrong result over a clean AttributeError, which is recoverable. This is diff-level review only -- the executable run-both-and-diff check against CPython is the review-usability skill and is out of scope here; do not claim a divergence you cannot see in the change.