🍕What should you eat tonight?

Only things you would eat tonight without complaint. The most common mistake is loading it with aspirational options — the salad nobody wants, the restaurant that closed — which produces results that get vetoed and a wheel nobody trusts. Six real choices beat twenty theoretical ones, and a shorter wheel is also far easier to read on a phone.

That is usually the better list, because a restaurant name answers the ordering question as well as the craving question. Paste the eight places within delivery range, spin, and you are done. Some people keep two bookmarked wheels — one of cuisines for cooking at home, one of takeaway names for when nobody is cooking at all.

Give it more lines. Pizza on two lines of eleven is 18.2% rather than 9.1%. This is also the honest way to run a household wheel where one person genuinely eats more of something: the weighting is on screen, in slices anyone can count, rather than hidden in a setting somebody could have quietly changed.

Tick remove winner and spin seven times. Each winner leaves the wheel, so you get seven different meals rather than pasta four nights running. Start with ten to twelve options so the last spins still have something to choose between — a two-slice wheel on Sunday is not much of a decision.

Your entries are carried in the page address, so bookmarking the wheel after you have edited it saves the list, and sending the link shares it. Nothing is stored on a server and nothing is uploaded — which also means clearing the tab without bookmarking loses the list, so bookmark it once and reuse it for years.

Every slice is the same size and the winner is drawn from crypto.getRandomValues with rejection sampling, so there is no bias and no memory. Getting tacos twice in a row is not the wheel being lazy — with ten options it happens about one time in ten, exactly as it should. Full details are on the methodology page.