Balance is a statement about probabilities
When designers say a game is balanced, they are usually making a claim about probabilities. In a symmetric two-player game, balance means neither side should win more often than the other for reasons that have nothing to do with player decisions. In a game with many characters, classes or factions, it means that no option is so strong that picking it is simply the correct choice, and none is so weak that picking it is a handicap.
The most common way to express this is the win rate: the share of games a given option wins. It is a probability estimate built from past results, and like any estimate it carries noise, bias and context. Most of the confusion about balance comes from treating a win rate as a fact about the game rather than as a measurement of it.
Why 50 percent is a reference point, not a goal
In a two-sided matchup, one side’s win rate is the other’s loss rate, so the average across all options is pulled toward 50 percent by construction. A character sitting at 50 percent is therefore just average, not necessarily fair, fun or well designed.
Several things can make a 50 percent figure misleading:
- Averages hide matchups. An option can win half its games overall while beating some opponents heavily and losing badly to others. Whether that is healthy depends on how those matchups are distributed.
- Pick rates matter. A strong option that almost nobody plays can have a modest win rate simply because its few players are experimenting. A very popular option often has players of mixed skill, which pulls its rate toward the middle.
- Skill changes the picture. An option that is hard to execute may win less in casual play and more at the top of the ladder, or the reverse. A single global number blends these groups together.
Designers therefore tend to look at win rate alongside pick rate, skill bracket and matchup spread, rather than treating a single figure as a verdict.
How chance enters the measurement
Even if a game were perfectly balanced, observed win rates would not sit exactly at 50 percent. Each game is a trial with an uncertain outcome, and a finite number of trials gives a noisy estimate.
A rough guide: for a true 50 percent rate, the standard error of the measured rate is the square root of 0.25 divided by the number of games. With 100 games that is 0.05, or 5 percentage points, so a range of roughly plus or minus 10 points (about two standard errors) is unremarkable. With 10,000 games the standard error is 0.005, and the same two-standard-error range shrinks to roughly plus or minus 1 point. This is why sample size matters so much when judging any change: a 54 percent result from a few hundred games may say very little, while the same result from a very large data set may be a real signal.
Selection effects add another layer. Players choose what to play, and they often choose based on what they think is strong, so the people using an option are rarely a random sample of the player base.
Small edges and how they compound
Chance also shapes how a per-game edge turns into results over a longer contest. If one side wins each game with probability 0.52, the chance of winning a best-of-three is p squared times (3 minus 2p), which works out to about 0.53. The series amplifies the edge only slightly. That is one reason a modest imbalance can stay hidden in individual results while still showing up in large aggregates. How series length changes outcomes is a topic of its own, but the underlying point here is that small per-game edges are easy to miss and slow to confirm.
Randomness as a balancing tool
Designers do not only fight randomness; they also use it. A game with substantial chance, such as critical hits, shuffled decks or randomised map elements, compresses the gap between strong and weak players, because the better player’s decisions have less deterministic influence. That can make a wide range of matchups feel winnable. It can also frustrate players who feel their choices should decide the result.
Chance can also hide balance problems. If outcomes are noisy enough, a real 3 or 4 point imbalance is harder to detect and easier to dismiss. The mix of luck and skill in a game affects both how fair it feels and how easy it is to measure. How much of a result is skill is a separate question, discussed under luck and skill.
Symmetry, asymmetry and the cost of variety
The easiest case to balance is a mirror matchup, where both sides have identical options. The first-move or starting-position advantage, if any, can often be offset by simple rules. Once sides differ, balance becomes a set of trade-offs between different strengths, and no single number captures it fully. Players also adapt: counter strategies emerge, pick and ban rules shift what is available, and the strongest option changes as the community learns. Balance is therefore a moving target rather than a fixed state.
What good balance work looks like
Most teams that handle balance well share a few habits, and none of them guarantees a perfect outcome:
- They look at several signals together: win rate, pick rate, skill bracket and matchup spread.
- They account for uncertainty and avoid reacting to small samples.
- They prefer smaller adjustments they can measure over large swings, since overcorrection is common.
- They treat player feedback as a source of hypotheses about how options feel, which data can then test.
- They accept that some imbalance is tolerable, since a game where everything is exactly equal might also feel flat.
Probability does not tell a designer what the right balance is. It tells them how much they can trust what they are seeing, and how large a change might be before it can be told apart from noise.