EIGRP Loop Prevention Logic
Hamid sent me the following question:
I have already memorized (bad idea, BTW) that a loop can occur if FD < RD. Could you please tell me how a loop could occur assuming FD < RD and we ignore the feasibility condition.
I’ll use a simple three-router network (see the following diagram) to illustrate why EIGRP cannot figure out whether an alternate more expensive path could lead to a loop or not.

Based on the link costs (I’m using scalar metrics for simplicity), C advertises the LAN IP prefix with cost=5. B adds its own incoming link cost (12), reporting the total cost of 17 to A (RD = 17).
A receives a similar update from C and adds its incoming cost (10). The total cost of getting from A to LAN prefix behind C is 15 (FD = 15).

However, A does not know whether B has an alternate path to C or whether it reports the path through A back to A.
Assuming EIGRP split horizon is disabled on B (which might be required for NBMA networks like DMVPN), B could increase the total cost to C and report the increased cost back to A.

We know that’s not the case because we can see the whole topology of the network. A doesn’t have that much visibility (that’s the main difference between EIGRP and OSPF); the only safe course of action it can take is not to use alternate paths with RD > FD.
There might be a post explaining that behavior somewhere on this web site (don't remember whether I wrote it or not).
Shouldn't it be (FD=17) ?