One good default beats ten options

A few years back, a couple of researchers at Carnegie Mellon put a website online called Spliddit. It splits things fairly: rent between roommates, an inheritance between siblings, credit on a group project. Not fair in the hand-wavy sense. Each tool runs an algorithm with an actual mathematical guarantee behind it.

The interesting part, to me, isn't the math. It's a design decision they made, and wrote down.

One algorithm, on purpose

For each problem, Spliddit offers exactly one method. No menu. And this is a group of academics who know perfectly well that the research literature often holds several different “fair” algorithms for the same problem, ones that are genuinely incomparable, each fair in its own way. They could have exposed all of them behind a settings screen. They deliberately didn't. Their stated reason was usability, and wanting to be able to explain the guarantee to a normal person in plain language.

Most apps do the opposite

Most money apps go the other way. They pile on options. Ten ways to split, toggles for everything, an advanced mode tucked behind a gear icon. It looks generous. Give people control, let them choose. In practice, a screen with ten split modes and a wall of switches isn't handing you control. It's handing you a decision you didn't want to make, while you're standing at a table trying to pay and leave. Every option you add is a small tax on the person who just wanted the thing to work.

Even “fair” has flavors

There's a second piece of research that makes the case sharper. A team that included the same Spliddit people looked at rent splitting and asked what “fair” should even mean. The obvious answer is envy-free: nobody would rather have someone else's room-and-price combo over their own. That sounds airtight, but it's not enough on its own. There can be many envy-free splits, and some of them feel a lot fairer than others.

So they picked a specific one, the split that lifts the worst-off person as high as it can while staying envy-free, and they tested it. They showed 46 real people their own actual rent situations, once with that split and once with a plain envy-free split. People rated the considered one as clearly fairer. They could feel the difference between technically fair and actually fair.

That's the whole argument for a strong default, backed by data. A default isn't the thing you fall back on when you don't choose. It's the single most important product decision, because most people will never change it, and a good one is something they can feel without knowing why.

How this shows up in dvup

This is more or less the bet dvup makes. There's a small number of ways to split, and a default that tries hard to be right. Scan a receipt and it just splits the thing and shows you the result. You can change it, some tables need a custom split, but you're adjusting a sensible starting point instead of assembling the answer from parts. Fewer knobs, a default that holds up.

I nearly went into design instead of engineering, and this is the kind of thing I can't let go of: the decision to not add the option is usually harder, and usually better, than adding it.

Questions