Skip to content

Lovat Dashboard

How predictions work

How Lovat estimates each alliance's score and its chance of winning a match.

Lovat predicts every match on the schedule. You’ll see the win chance on Now and Matches (for example, Red 64%), and the full prediction on each match page. After a match is played, the page keeps showing what Lovat predicted before the match, so you can see how it held up.

Predictions use the same sources as everything else. Each robot needs at least two scouted matches; until then, Lovat says which teams don’t have enough data yet.

Lovat starts from each robot’s total points per match: its average and its standard deviation , which says how much its score varies from match to match.

To find win percentage, it first builds each alliance’s score distribution from the three robots’ averages and standard deviations:

Treating each score as a normal distribution, combining three robots into one alliance might look like this:

0102030405060708090100
Robot averaging 8 pointsRobot averaging 20 pointsRobot averaging 40 pointsAlliance

The horizontal axis is points scored in a match, and the height of each curve is how likely each score is. The alliance’s curve is wider than any one robot’s, because three robots’ ups and downs add up.

Next, Lovat subtracts one alliance’s distribution from the other’s to get the distribution of the point difference between them:

-40-20020406080100120
Blue allianceRed allianceRed points minus blue points

Any point to the right of zero on the difference curve is a red win, because red’s points minus blue’s points is positive. Any point to the left is a blue win. So the chance that blue wins is the area under the curve to the left of zero:

-40-20020406080
Blue wins: the shaded area, about 3.3%

Lovat finds this area by working out zero’s z-score on the difference distribution and converting it to a probability:

Here is the standard normal cumulative distribution. In this example blue has about a 3.3% chance of winning, and red about 96.7%.

  • A prediction is only as good as the scouting behind it. Early in an event, two or three matches per robot can swing a lot.
  • A close prediction, near 50%, means the alliances look evenly matched, not that Lovat doesn’t know.
  • Breakdowns, notes and defense don’t go into the number. Check the team pages before you plan around it.