Strategy guides › Eyeball
How to score lower in Eyeball: tricks by eye for all five figures
▶ Play Eyeball🎯 Try a round with a wide angle
The angle is wide, almost flat, with a shallow arc hanging under the vertex. Halfway between where the arc meets the two arms looks like the obvious spot. You lock it in, the green dot appears, and it sits a long way further down the arc. Thirteen units off, on a figure where a good eye loses one.
Eyeball has no hidden rules: five figures, one point each, and your score is the total distance between your points and the true ones, in units of a 100 × 100 box. What it does have is traps. Some figures punish a small slip far more than others, and some “obvious” shortcuts are wrong in a way that is easy to measure. We generated 20,000 rounds with the game’s own figure generator, 100,000 figures in all, and measured both.
What a score means
To get a feel for the scale: if you ignored the figures and clicked the centre of the box every time, you would score about 97 per round on average (the generator keeps every answer at least 10 units from the edge, but places it anywhere inside that margin). The top rating, Draughtsman, needs 5 or less, which is one unit per figure. On the midpoint task the lines are 36 to 64 units long, so one unit is about 2 % of the line. That is the precision you are aiming for.
The useful thing about that number is that it tells you where not to lose points. One figure 10 units off ruins a round that four perfect figures cannot save. So the tricks below are mostly about avoiding the big errors, not polishing the small ones.
Turn every figure into halving
Halving a gap is the thing human eyes do best, so the first trick is to reduce each figure to halvings. For the midpoint that is the whole task. The generator tilts every line between 15° and 75° so the screen edges cannot help you. Do the job twice: judge the halfway point once working from the top end, once from the bottom end, and settle between your two judgements. Errors from two independent judgements partly cancel.
There is one well-studied bias to know about. On horizontal lines, most people mark the middle slightly to the left of true centre, an effect called pseudoneglect; a meta-analysis of line-bisection studies found it across many experiments, and the shift is small. Our lines are never horizontal, so it is not a big factor here, but if your misses on long, shallow lines keep falling on the same side, that is the likely reason.
The circle is two halvings. Do left against right first: where is the point that leaves the same gap to the outline on both sides? Then top against bottom along a vertical through that point. Two one-dimensional judgements are easier than one two-dimensional guess.
The parallelogram: slide both sides
You get three corners and two sides; the fourth corner is where the missing sides meet. The tempting shortcut is to square the shape off. Don’t: the drawn sides meet at anything from 40° to 140°, and treating them as a right angle misses by 12.5 units on average.
The right construction is to slide one side along the other: take the side from the middle corner to one end and move it, in your head, so it starts at the other end. Its free end is the answer. You can slide either side, and the two slides end at the same point, which gives you a free cross-check.
To put numbers on it, we simulated a player whose every placement by eye is off by a random amount of about 4 % of the distance (or angle) involved. That figure is our assumption, not a measurement of anybody; only the comparison between methods matters. With one slide that player misses the corner by 1.45 units on average. Doing both slides and splitting the difference cuts it to 1.04. Using only the diagonals is worse, 2.29, because any error in the midpoint gets doubled when you extend through it. So use the diagonals to check, not to build: in a parallelogram the diagonals bisect each other, so the line from the middle corner through the middle of the gap between the two end dots should run straight into your point.
The crossing: use the whole line as a ruler
Two stubs stop short of each other; you mark where they would meet. Each stub ends 10 to 18 units before the crossing, and the generator keeps them 30° to 150° apart. The common mistake is to aim at the gap between the two near ends. That misses by 9.5 units on average, because the crossing is usually well beyond it.
What actually goes wrong for a careful player is a small tilt. If you extend one line just 1° off, the crossing slides along the other line by 0.31 units on average, and by up to 0.63 when the lines meet at a shallow angle, where a tiny tilt moves the meeting point a long way. So sight along the whole stub from its far end, the way you would check a plank for straightness, and extend it slowly into the gap. Then do the same for the other line and see whether your two extensions meet where you put the point. On a shallow crossing, spend your time here.
The bisector: don’t stop on the chord
This is the figure from the opening. The arc is centred on the vertex, so the answer is the middle of the arc, and the middle of the arc is never on the straight line between its ends. How far off the chord midpoint is depends on the angle. Across our figures it misses by 7.2 units on average and by up to 20.4 on the widest angles (the generator deals 40° to 140°). Round 128’s 140° angle costs 12.9.
The fix is to use the chord without stopping on it. Halve the chord, which is a midpoint task, then push the point straight out from the vertex until it meets the arc. That is exact geometry: the line from the vertex through the middle of the chord bisects the angle. In our simulation it cut the average error from 1.14 units (judging the half-angle directly) to 0.84.
Practice
The practice round is seed 128, which has both figures from the diagram: a 138° parallelogram and a 140° bisector, the two widest kinds of trap. Slide both sides, push out from the chord, and remember that the obvious point is usually wrong by exactly the amount the figure was designed to hide.
