This discussion stems from here : #16622 (comment)

The numpy documentation says the following about instantiating the dtype object using "{'field1': ..., 'field2': ..., ...}":

"This usage is discouraged, because it is ambiguous with the other dict-based construction method. If you have a field called ‘names’ and a field called ‘formats’ there will be a conflict."

I think static typing is a good place to catch not just "wrong" types used, but also to catch early such discouraged usage in numpy. So, IMO we should have the type checker error on such behavior. WDYT ?