Every real decision eventually comes down to trying two things and comparing them. The trouble is usually time: standing up a second version of anything takes an afternoon you don't have.
Spin up a workspace per idea, point each at the same repos, and let them run in parallel — separate worktrees, separate threads, no risk of one idea stepping on another's changes. By dinner you're not debating which approach is better; you're looking at two working versions and picking one.
It's the same trick behind A/B testing a feature, except the variants are entire implementations, not just a button color.