Stability is a relative concept and you'll only know whether a version is stable for your use case, if there were a tool that allowed us to enter the features we need, and then the tool analyses the number of open defects against that feature. Cisco did this many years ago with routers and switches and it was a way to perform code recommendations. It was flawed in many ways because the bug id's were not always created with the required component detail to allow those tools to spit out a magic answer.
Reading release notes is also a hit and miss affair. Whoever writes the subject line of a bug ID needs to think carefully about what they are writing. If you see bug subjects like "Backups fail after upgrade" (I just mnade up a bad example) - what does that lead you to do? Avoid that version (or patch version?). It might turn out that the bug was raised for a very specific instance and that in 99% of the time you won't hit that bug. But it was a corner case and it was solved. Big deal. If you read release notes of any ISE release you'll end up never using that version because of all the doom and gloom.
In my 20 years of working with IT gear, I think it's sad that we still have to end up testing the vendor's code in our labs and production. Some vendors get it mostly right (Apple, Microsoft) but networking vendors don't have the same focus on reliability and quality and therefore push the problem onto their customers.
For what it's worth, I started my journey on ISE 2.2 patch1 and it was an utter disaster release for me. My use case was a fully distributed VM deployment using TACACS+, Radius auth, Sponsored Guest with Remember Me feature. No jokes, but I raised over 52 TAC cases in 14 months and 25 new bugs ID's were raised as a result. I didn't stay on 2.2 very long and moved to 2.3 - the same story there. Now I can report that on 2.4 patch 1 I managed to close a bunch of my ISE 2.2 issues as well as any others that came along. In my world, I am happy with ISE 2.4 patch 1 - it's a long term supported release and I think we may stay on this one unless there is a compelling feature in 2.5 etc.
