This is not true. The chain of logic models are able to check their work and try again given enough compute.

They can check their work and try again an infinite number of times, but the rate at which they succeed seems to just get worse and worse the further from the beaten path (of existing code from existing solutions) that they stray.