We built a game that weighs your halves. After 76 recorded runs, the highest score is 282 out of 300, and the shape of the failures is always the same.
Halfsies is the casual daily split-in-half cutting game. It gives you three objects a day and one cut each. Both halves go on a scale, and your score is how close the lighter half comes to the heavier one, out of 100 per cut.
A perfect day is 300. Across the 76 runs recorded so far, from 46 players over 14 days, nobody has scored it.
| Out of 300 | |
|---|---|
| Best run recorded | 282 |
| Average run | 197 |
| Median run | 206 |
| Worst run | 24 |
| Runs at 270 or better | 3 of 76 |
| Runs at 300 | 0 |
That is a small sample and we will publish a bigger one. But the shape has been stable since the first week: a fat middle around 200, a thin tail of good days, and nothing at the top.
Two hundred out of 300 means the average cut splits an object about 67/33 by weight. People are not missing by a sliver. They are missing by a third.
The obvious explanation is a shaky hand. It is not, because the game removes the hand. The blade lands exactly on the line you draw, every time, with no timer on the daily mode. You can stare at the object for a minute before committing.
The scores stay in the 200s anyway.
A cut through the visual middle of a shape is a cut through its middle length. The scale measures its mass. Those two numbers come apart the moment an object stops being a uniform rectangle, and they come apart fast.
Stand a cone on its base and cut halfway up. The top piece is one eighth of the whole, because volume scales with the cube of height. To get two equal masses the blade has to land about a fifth of the way up from the wide end, leaving a short fat stump that weighs as much as the tall cone above it.
Area grows with the square of the radius. Volume grows with the cube. Your intuition is running a linear model on a cubic problem.
Nobody looks at a cone and guesses one fifth. That gap, between where the middle looks and where the mass actually divides, is most of the missing hundred points.
There is a second, much smaller effect. Ask people to mark the midpoint of a drawn line and they land slightly left of true, by a percent or two. It is called pseudoneglect and it has been replicated since the 1980s. It is real, and next to the geometry it is a rounding error.
Four players have five or more recorded runs. Their first runs average 184. Their most recent three average 212.
That is 28 points from experience alone, and it is not steadier hands. It is learning to aim at the fat side of centre instead of at the middle, on the kinds of shapes that hide their weight: anything that tapers, anything that swells, anything with a pit or a bone in it.
Find the widest cross section first, then ask which side of it holds more body. Cut toward the fat.
Cut nearer the wide end than looks right on anything that tapers. Cones, pears, drumsticks, bottles. The narrow end holds far less than its length suggests.
Turn the object ninety degrees and judge again. If your two answers disagree, the truth sits between them, and you have measured your own bias for free.
Assume a hidden core. A peach is a sphere with a rock in it, and the rock is off-centre.
282 is the record. 197 is the average. If you want to know which one you are, the scale does not care what the cut looked like.