Laugh, I just hit reply on my last posting, got an invalid authentication, or some such, refreshed paged, and saw David's posting hitting some of the very issues I was going to bring up, like the high auto reference cost especially along with using manual costs of 10K.

I was going to also ask about EXACTLY what has been appearing, or not, in route tables, along with what looked incorrect about the OSPF database.

BTW, David, "The max metric of an OSPF link is 65535. So it wont install in the RIB if it hits the max metric.", in the past, I recall (?), when I've actually bumped into hitting the max cost for multiple paths, all such were considered equal and were installed.  (I also recall using an IOS version that when you exceeded the number of max OSPF ECMP routes, it would incorrectly compute what should be installed [in my case, though, IOS had been fixed].)

Anyway, I just set up a very small PT environment, with a topology of one spoke area only having one router, and another spoke having two routers.  Did not muck with costs and/or auto bandwidth costing, but, of course, sh ip route and show ip os database look "different" from each of the spoke L3 switches.  Hence, probably worthwhile to fully understand EXACTLY what you believe OSPF is doing incorrectly.

Don't know if it's of any use, but I've attached the PT file.