The stubs currently add DtypeLike as "anything that can be coerced by np.dtype. However, with new user dtypes, the dtype classes will be the important building-blocks probably. For these I currently use the spelling DType (capital T).

This does not need to affect the stubs, however, it may be that the stubs basically end up using the DType class as the category (i.e. byte order and string length rarely make a difference probably). In this case keeping the spelling between the DType class and the DTypeLike synched up, may make additional sense. Otherwise, being intentionally different may be an argument.

I guess I use the DType capitalization as a short form of camel-case DataType, but I am open to reconsider that as well.