Imagine you're considering a change to the architecture of your system.

The change is unproven, but could significantly simplify the system.

The first question to ask yourself is: if this were a bad decision, how could I find that out as quickly as possible?

Then seek to falsify it as quickly as possible.