By resisting the urge to open a design tool. A blank canvas is a distraction when you don't understand the problem yet. I need to know who's using it, what they're actually trying to do, and where the friction is — before any screen gets drawn.
When I can explain why every major decision was made. If the answer is "it looked right," I'm still exploring. A committed direction has a reason behind it that holds up out loud.
Parallel directions, not sequential drafts. The goal is to see the edges of the solution space before committing to anything inside it. Once the alternatives are laid out honestly, the right one is usually obvious.
I ask the same questions a good reviewer would — "why this? what does the alternative look like? who fails with this decision?" The answers have to be honest. When they're not, I go back and explore more.
It emerges from the work. Every spacing or color decision sets a precedent. By screen 30, a system exists whether you named it or not — codifying it just makes the build consistent. The Train5D system wasn't planned upfront. It was extracted from what I'd already made.
Everything gets tighter. There's no room to draw something and say "the developer will figure it out." Every design decision has an implementation cost I'm fully aware of — which makes the design more honest, not more compromised.