Should csv.DictWriter require fieldnames to be a Sequence?

I realise that according to the docs the fieldnames parameter to DictWriter should be a sequence, however code which doesn't actually pass a sequence (instead perhaps passing dict.keys()) does in fact work and may have a well-defined key ordering.

Should the typeshed record what actually works or closely match the docs?
(Should the docs be updated?)

Generally, we prefer to follow the implementation, not the docs in typeshed.