BGP/ISIS problem

Hello Network Reddit people,

I have an issue i've been scratching my head on and have been leaning heavily on GNS3 for this testing.

Please see:DIAGRAM

In the diagram, R1/R2/R3 build BGP with their loopback interfaces that gets redistributed into isis. R2/R3 only peer with R1 since it's the route-reflector. The BGP route also follows the lowest metric as expected. But the problem I have is that whenever i loose the lower metric route I get TTL exceeded.

For Example: Link with 160 metric goes down. Bgp next-hop to 192.168.0.0/24 from R1 is 3.3.3.3, which is R3 loopback IP learned from R2 via ISIS. I also see that R2 has this route (R3 Loopback Ip) with a next-hop of 3.3.3.3 to R3 over the 300 metric link. I also see on R3 that it has a next hop of the route-reflector (1.1.1.1) to 172.16.0.0/24, with the loopback IP of R1 learned via ISIS from the R3/R2 peering.

FIX: What gets to me is that I can fix this on R2 whenever I set a static of 192.168.0.0/24 to the point-to-point ip between R2/R3. This doesn't make any sense to me, because R2 should be forwarding return traffic (dst:192.168.0.0/24) to the R3 loopback ip as the next hop, since R1 had its next-hop set as R3 and not R2.... Is this an issue with recursive route-lookup in my virtual environment and/or what am I mis-understanding with how packets should be getting forwarded?