I’d like to clarify one thing I said above. Yes at the moment I think it’s reasonable to implement a species group using the ‘complex’ taxon, only because ‘species group’ doesn’t exist in iNat. I think the ideal is that both should exist. I know of at least one large genus where we felt the need to implement ‘species group’ using ‘section’, because the genus is divided into well defined subgenera, which contain well defined species groups, which contain well defined complexes, all of which are useful to the identifier! (I speak of Platycheirus) If you agree that ‘species group’ should exist, there is a feature request for it here.
On naming, the curator guide gives clear examples - the name should be written in full with only the genus capitalised. I’m not sure what you mean with the reference to subgenera - these are a much more ‘formal’ regulated rank, I don’t think a complex or species group should be implemented as a subgenus. The example you give is not really appropriate - the subgenus is called Fuscopolistes and is a recognised subgenus. ‘Fuscatus-group Paper Wasps’ is given as a ‘Common Name’. I’m not sure of the history. I would hazard a guess that it was called a species group before being made a formal subgenus and that specialists commonly refer to it in that manner - but we should not be raising ‘groups’ to ‘Subgenera’ ourselves.
What it means for a complex to be ‘recognised in the literature’ (as per the curator guide) is a bit fuzzier I think: it can be mentioned almost in passing that such-and-such a species is part of a complex with A, B and C, without a ‘Species such-and-such complex’ being formally declared anywhere (the curator guide even gives instructions on determining a name for complexes), sometimes there is simply an awareness that two species are indistinguishable e.g. apart from genitalia. I think this should count as evidence of ‘recognition’ if specialists on the site agree. They are rarely formally declared and named in the same way as e.g. Subgenera. So I think there is discretion - what we should absolutely not be doing is creating complexes left, right and centre for everything we think looks a bit similar, and I think that is the point. There should be evidence of the general recognition of a complex, and it should be implemented sparingly, and only if useful to identifiers.