Before going into more detailed analysis, I should however also emphasize one deeper theoretical aspect of subgrouping. Some people might look at my previous list of four methods of synchronic subgrouping, and their various weaknesses, and go “OK, so does this mean that none of these are correct and we still need to refine our methods before we can correctly define synchronic subgroups?” I would reply that this is not the right question to ask.
Synchronic subgroups have no basic existence of their own. They are abstractions, put together e.g. to help people interested in linguistics keep better track of the actual state of the territory. A giant matrix that tracks all existing dialect varieties and all relevant traits is already a complete dialectological description. But this will be unfeasible to memorize, let alone understand.
There is therefore no such thing as the objectively “right” or “true” way to subgroup data, and synchronic subgrouping will have to depend on what we want to do with it. Analyze the history of a particular feature? Provide other linguists with a taxonomy to refer to? Teach linguistics students more effectively?
To an extent, this issue propagates also back to historical subgrouping. Given two distinct varieties A and B at some time in the past, we can indeed ask what their descendants at some later time are. This much is something we can, in theory (if not necessarily in practice), determine correctly. But any means of defining the distinction between A and B in the first place is itself just another case of synchronic subgrouping — and, thus, in principle, an arbitrary choice.
In practice we don’t seem to think that all historical classifications would be arbitrary, though. Yet this seems to be less due to the existence of a theoretically absolutely basis, and more due to the fact that various divisions eventually “solidify” on a single border. A distinction such as Russian vs. Persian could be drawn solely on the basis of lexicon, or morphology, or phonology, or numerous smaller things yet like prosody or kinship terminology. By far most methods would still come to the exact same conclusion. Although we can explain this state of affairs in terms of history, it has been plainly observable long before we had any general theory about the historical development of languages.
Attempting to draw a similar boundary between Proto-(Indo-)Iranian versus (early) Proto-(Balto-)Slavic would be however less unambiguous. And it seems clear to me that there must have also once existed, across eastern Europe / western Central Asia, a whole chain of connecting varieties between these (as is suggested e.g. by some parallel developments like *ḱ > *ć > *c > *s in both Iranian and Slavic) — and moreover, if there were enough space for all of these varieties to have left as many daughter language varieties as Proto-Balto-Slavic and Proto-Indo-Iranian did, in principle it would be entirely possible for a dialect continuum of thousands of varieties to still organically bridge Russian and Persian.
(I don’t think this argument necessarily torpedoes the concept of history-based classification though — but I’ll leave continuing that tangent for later.)
It is additionally not difficult to realize that this is also the same issue as the construction of distinct “languages” from a set of dialects; and in a sense, even the construction of distinct “dialects” from a set of idiolects works like this. Surveys may normally define local dialects simply on the basis of geography rather than thru analyzing numerous idiolects, but this practice is itself derived from the observation that areal variation is under “usual conditions” probably the most prominent component of linguistic variation.
And of course this still cannot be assumed to be the case always. In modern major metropolitan areas, in our era of mass media and noncompact social networks, it would not be very surprizing at all if class, education, ethnicity etc. turned out to be more prominent dialect division lines than the precise borough or suburb where someone lives or has lived. Or, child speech is most of the time best “subgrouped” by age and acquisition level. Likely there is areal variation too, but it’s simply of no fundamental interest for the study of language acquisition.
Now suppose that we were interested in tracking something like the spread of playground slang though — and suddendly areal variation in child speech starts looking quite relevant after all, while the precise age of informants might recede to a distinctly tertiary issue. Likely no one will ask if the areal variation is “more real” than the age variation however, or vice versa.
With competing models and metrics of dialectological subgrouping, especially if employed for very similar goals, with research questions that heavily refer to pre-existing models, it may not seem as clear that approaches that are more useful are not actually “truer”. But the underlying reality is the same: they remain models, not data in itself.