Rethinking Puppet Deployment

How can you possibly make an argument for not caring about failed deployments? "Skip over", "do it again", "its fast".

You can't possibly be talking about deployment here. Deployment is the result of moving tested, proven, known working artifacts into an environment in which they can execute or be accessed by an executable environment.

There is no skip, retry or anything resembling failure in a deployment model that Im aware of. So Im not sure why I would use a tool assuming it would fail to do what its intended to do.