What to measure
Learning paths through this topic
Which kinds of players get knocked down most, and are they usually rescued in time?
In a game where a teammate can be revived, the story is told by how often rescues succeed and how long they take. This entry shows how to count them so that people who were still down when the match ended are not mistaken for failures.
Did players leave the match because of a bad connection or because they wanted to?
A player who disappears from a match could have been thrown out by lag or could simply have had enough. The fixes are completely different. This entry shows what to log around every exit so the two can be told apart afterwards.
Players' events arrive days late or carry a date in the future
Games played offline report later, and some devices have the wrong date. If you trust every timestamp, yesterday's chart keeps changing and some events land in next week. This entry shows what to do with each kind of strange timestamp.
The market price of an item is wrong because two friends keep trading it at a crazy price
Two players passing an item back and forth can drag the recorded market price anywhere they like. This entry shows how to spot such pairs in the trade log and how to compute a price that ordinary traders agree on.
Do new players who lose their first online match come back less often?
It is tempting to compare how many winners and losers returned, but the newest players have not had time to return yet and make everything look worse. This entry shows who to leave out and how to avoid drawing a conclusion from too few people.
Two players' games drifted apart, but when did it start?
When two screens show different worlds, the interesting question is the first moment they differed and who was the odd one out. This entry shows a small fingerprint to log that answers it afterwards without storing the whole game.
Do players with friends stay longer, or are friends just a sign of keen players?
Groups that play together nearly always stay longer in the raw numbers, but that may only mean the keen players are the ones who bring friends. This entry shows how to compare like with like before you build features around it.
How often does a group finish a match and straight away play another one together?
The best sign that a co-op match was fun is that the same group starts another one. This entry shows how to recognise a group in the logs even when the players join in a different order, and how to avoid counting the same follow-up twice.
Report where players died without making a single player recognisable
Counting deaths per spot is useful, but a spot visited by two people can point at a person, and hiding just one small spot can still be undone with simple subtraction. This entry shows how to publish the counts so no one stands out.
Gold piles up in the game, but is everybody richer or just a few players?
The average balance can rise sharply because a handful of players hoard, while the typical player has not gained anything. This entry shows the three figures that together tell whether prices should change.
One player got the best drops three times in a row: luck or a bug?
People notice streaks and complain, but a streak of some length is expected in a long enough run of fair draws. This entry shows how to work out how likely a streak of that size really is before you touch the loot code.
12 of 40 testers quit at the bridge: is that a real problem or just chance?
With a small group of players, a percentage jumps around a lot, and two levels with different numbers may be equally hard. This entry gives your AI an honest range for the number, and a rule for when two numbers really differ.