About Distroguess
Every round shows you a question about real data, like how rich the world's countries are, how much the stock market moves in a day or how many goals a Premier League match has. You guess how the values are spread out. Then the real data appears on top of your guess.
How to play
- Read the question and the line under it, which says what one value is: one country, one day, one home sale. The chart's x-axis shows the range those values fall in, and the height of your curve says how many values sit at each point. Some axes use a log scale, where each step multiplies the value by 10. Wealth, population and planet sizes span too many orders of magnitude to show any other way.
- Draw the shape you expect with your mouse, finger or pen. Higher means more common. Draw along the bottom to erase.
- Or pick a ready-made shape (bell curve, skewed bell, long tail, decay, flat or two peaks) and move its sliders. You can draw over a shape to fine-tune it.
- Press the tick button (or Enter). The real histogram appears with your score out of 100. The same button then moves you to the next question.
The four topic buttons at the top right (world, money, science and everyday life) switch topics in or out. A change applies at once: questions you have played stay, and the rest of the round comes from the topics you chose. If your topics have fewer than five questions, the round is shorter and does not count towards your best score.
A round is five questions, so the best possible round is 500. Only the shape counts: the height you draw at does not matter, because your curve is scaled to the same area as the data.
Keyboard: 1 to 6 pick a shape, Backspace clears, Enter guesses or moves on.
How scoring works
Your curve and the real data are both cut into the same bins as the histogram, and each is turned into a probability distribution that adds up to 1. The score compares the two.
- Score is
100 × (1 − √JS), where JS is the Jensen-Shannon divergence in bits. JS is 0 when the two are identical and 1 when they have nothing in common. Its square root is a true distance, which keeps the score fair across the range. - Overlap is the share of probability the two distributions have in common: the area under both at once.
- KL is the Kullback-Leibler divergence of the real data from your guess, in nats. It shows the "surprise" of the real data if your guess were true. It punishes hard when you give almost no weight to values that do happen. To keep it finite, your guess is mixed with 0.1% of a flat distribution. KL is shown for interest and does not affect the score.
We use JS for the score because KL is lopsided and can be infinite: one empty bin where the data has values would wreck a round.
Some datasets are small (about 150 to 220 countries), so their histograms are bumpy. A smooth guess cannot match every bump, so a score in the 80s on those is very good.
The data
All questions use real, published data: the World Bank, US government statistics, HM Land Registry, NASA, the US Geological Survey, Wikidata, the Bitcoin blockchain and others. All of it is public domain or openly licensed. Open "Source" under any result for the details and links. See the full list of data sources.
Who made this
Distroguess is made by Nodo Labs. It is free, with no ads and no tracking. If you enjoy it, you can buy us a coffee on Ko-fi.