I’ve followed a similar pattern before in large features as well. During each iteration my mental model improves dramatically, and I have no qualms about deleting everything and starting over, because I _know_ the next iteration will be that much more correct, performant, and/or maintainable.