First for the actual data.
As Damgaard linguistic supplement pointed out the earliest evidence for IA in Syria refers to mariannu in a tablet from Tell Leilān from 18th century BCE. This suggests IA were alreay making presence felt in region by then. The early Aryan names from West Asia are predominantly Indo-Aryan not Iranian suggesting that this branch was not undivided I-IR but already divided IA. So we have to place IA separation after that.
Narasimhan et al& Damgaard et al’s data clearly indicates that there is east Asian admixture from 1500 BCE onwards. This type of admixture is seen in Eastern Iranics like kuShANa and shaka but not at all in Indics. This means that the Aryan group which invaded India should have done so before 1500 BCE.+++(5)+++ Given when the BMAC outliers have steppe admixture and when Harappan civ collapse we can say this is medially at 1700 BCE but could be 1900 BCE. This is quite possible given that matches devolution of mature IVC.
So from multiple lines of evidence IA-IR split could have been at least 1900 BCE. The Lubotsky’s detailed study points out that the loans of Aryan into Finno-Ugric are not merely proto-I-Ir or Ir but specifically show dialect developments of IA. There is no evidence for FU close to Indian subcontinent. Hence, I-Ir split should have happened further north even as these groups were operating in the Uralic zone. They were already nearing BMAC by ~2000 BCE; so this split was earlier.
The RV suggests that the term asura is not all negative: in fact more often positive than negative and used in positive sense similar to what later continued in avesta. maNDala 8 of RV and some paippalAda AV sUkta-s show parallels to avestan material. So early IA material was composed in milieu where Iranics were part of the circle. If IA they had already settled in the greater Panjab there is little evidence for Iranics in their proximity. In contrast in yajur brAhmaNa-s we do see strong deva-asura polarity mirrored in avesta. Hence, that split reflects a later period, which is more in line with a time of schism and separation.
Now, admittedly pushing this a bit. There are parallels between IA and Balto-Slavic not present in Iranic so again suggests intial dialect separation in their vicinity which must be far from India hence. Hence, the idea is already there was dialectal divergence in the steppes leading towards IA when earliest layers of RV and AV-P were composed.
The linguistic developments in India and compilation merely “updated” this old composition to a new register. This has happened even later: e.g. paippalAda in kAshmIra vs Odisha or the case of mAdhyaMdina. I believe something along these lines was mentioned by @RangaTheDude the other day.
Now methodological quibbles: glottochronology while a valid idea lacks the calibration which its relative in biological phylogenetics has: neither good fossils nor a molecular clock has been worked out. In fact those who did it and published in nature and science did it wrong. Chang et al did a better job with a lot of statistical smoke screens though & they got dates as early as 2100 BCE for the split of I-Ir
Problems with Mitanni superstrate. THere is no doubt about the IA nature of the Mitanni superstrate but to read details into it are dangerous. Many are reconstructions of sound values and forms embedded in non-IA languages. Hence, to claim primitiveness based on those e.g. the very form of the cognate of mILha is to be taken with caution. Moreover consider how mILha is pronounced in RV recitation. That itself points to memory of archaism as opposed to how it is rendered in later Skt.