Your question implies that you think that the solution that you need is Policy Based Routing. I do not think that you need Policy Based Routing to do what you describe.
The simple way to get load sharing is to configure two default routes and point one to the next hop address of the DSL device and point the other to the next hop address of the T1. That will give you load sharing.
The other part of your requirement is that if one carrier goes down it will not impact traffic. Depending on how the T1 is configured this may be fairly simple. If the T1 is using PPP encapsulation or is using HDLC encapsulation then if the carrier goes down the interface will go protocol down (this is based on failure of keepalives which is a feature of PPP and of HDLC). If the interface goes protocol down then the default route pointing out that interface will be withdrawn from the routing table and all traffic will be directed out the DSL.
Recognizing a failure of carrier on the DSL will be more difficult because DSL does not supply a keepalive function (as PPP and HDLC do). There is a feature in recent IOS called Object Tracking which allows you to test for reachability of some object (it could be a remote IP address). I would suggest that you set up Object Tracking and track some address in the DSL provider network. With Object Tracking if the tracked object becomes un-reachable you can configure that the default route through DSL gets withdrawn.
If this is a remote site which connects to a headquarters network there may be another alternative to consider. You could run a GRE tunnel over the DSL from the remote site to the headquarters site (or perhaps IPSec with GRE if you want to assure the security of traffic over this connection). Then run a routing protocol (like EIGRP or OSPF) over the GRE tunnel and over the T1. The routing protocol can advertise the default route to the remote site. And the routing protocol uses dynamic hello protocols which will detect failure of either the T1 or the GRE tunnel and will converge and direct all traffic over the surviving link.
I have implemented something similar to this last alternative at a customer site and it works quite well.
HTH
Rick
HTH
Rick
