Pity, K/D and win rate: reading game numbers honestly

Game statistics are simple to calculate and easy to misread. A pull rate does not mean what it sounds like, a kill/death ratio measures something narrower than skill, and a win rate over twenty games is mostly noise. None of this makes the numbers useless — it just means knowing what each one can and cannot tell you.

Why a 0.6% rate needs a pity system

Take a banner with a 0.6% chance per pull. The instinct is that a hundred-odd pulls should be plenty. The arithmetic disagrees: at a flat 0.6%, the average number of pulls needed for one hit is 1 ÷ 0.006 = 167. Half of all players would still have nothing after 115 pulls.

Work it the other way and it is starker. The chance of getting nothing in ten pulls is 0.994¹⁰ = 94.2%, so a ten-pull has only a 5.8% chance of hitting. Across a full 90 pulls, the chance of nothing at all is still 58% — the majority outcome. A flat 0.6% rate with no floor would leave most players empty-handed at the point the game promises a result, which is exactly why hard pity exists, and why the advertised rate is not the rate you should plan around.

Many games also apply soft pity, raising the rate sharply for the last stretch before the cap. Where that applies, the flat-rate arithmetic above becomes pessimistic — but it is still the right way to see why the mechanism is there.

Past pulls do not owe you anything

Fifty failed pulls at a flat rate do not make the fifty-first more likely. Each pull is independent, and the counter is not "due". This is the gambler's fallacy, and gacha interfaces — with their visible counters and near-misses — are unusually good at provoking it.

The genuine exception is the pity counter itself, which is stateful by design. Being 80 pulls into a 90-pull guarantee really does change your position. Being 80 pulls into no guarantee at all changes nothing. Knowing which system you are in is the whole difference between planning and superstition.

K/D measures trading, not contribution

Kills divided by deaths: 20 kills against 10 deaths is a K/D of 2.0. It is a clean measure of one thing — how favourably you trade — and blind to almost everything else. Assists, objective play, healing, utility, holding an angle so a teammate can rotate: none of it appears.

Which is why K/D and win rate frequently disagree, and why the disagreement is informative rather than paradoxical. A player who never dies because they never contest anything can post an excellent K/D while losing. Read K/D as a description of playstyle and let win rate speak to outcome.

A win rate over twenty games is noise

This is the one that costs people the most. 30 wins in 48 games is 62.5%, and it looks like evidence. Small samples are far wilder than intuition allows.

At 60% over 20 games, the 95% confidence interval is roughly ±21 points — your true rate could plausibly be anywhere from 39% to 81%. At 200 games it narrows to about ±7 points, and at 1,000 games to ±3. So a 20-game "hot streak" at 60% is statistically indistinguishable from a coin-flip player having a good night, and a hero you have played 15 times has no reliable win rate at all, no matter what the profile screen displays.

The practical version: below about 50 games, treat a win rate as a rumour. Between 50 and 200, treat it as a hint. Above a few hundred, start believing it. And when comparing two numbers, check both sample sizes before concluding anything — a 58% over 400 games is far stronger evidence than a 71% over 30.

What to actually do with them

Use pull-rate maths to set a budget before you start, not to justify continuing. Use K/D to notice how your playstyle trades, not to rank yourself. Use win rate only once the sample is big enough to mean something, and compare like sample sizes with like.