Level Completion Rate: How to Read It
A completion rate is the share of players who start a level and win it, and it means nothing until you compare it with what you wanted. If 41 of 100 finish, split up the other 59, because the people who left are not one group and each kind needs a different fix.
What does 41 out of 100 mean?
Fox Ring Trials is an invented action game, and these numbers are made up for this example. In one week, 100 players started level 3 and 41 finished it.
Read that in three steps.
First, how exact is it? The true share is very likely between 32% and 51%. That is the range you get from 100 players. From 10 players, 4 of 10 would give you 17% to 69%, which is nearly useless. From 1,000 players, 410 of 1,000 would give 38% to 44%.
Second, compared with what? The maker planned for 70%. Even the top of the range is far below that, so the level is harder than she meant. Level 2 had 88 of 104 finish, which is 85%. So the fall comes at this level and did not build up slowly. If level 3 were a planned boss with a 40% target, the same number would be good news.
Third, who are the 59? This is the step that most people skip. A rate tells you how many left and nothing about why. The log can split them by what they did before leaving.
| What the 59 did | Players | What it suggests |
|---|---|---|
| Lost 3 times or more, then left | 31 | Too hard, or it feels unfair |
| Lost once or twice, then left | 16 | Losing costs too much |
| Left without losing once | 12 | Lost, bored or confused |
These groups need opposite treatment. Making enemies weaker helps the 31 and does nothing for the 12, who never met an enemy that beat them.
The maker played the level with those groups in mind. The 16 made sense straight away: a loss sends you back to the very start, four minutes away. The 12 fit a long corridor with two identical doors. The 31 all reached the same fight at the end.
Which fix comes first?
She had three ideas, and picked the cheapest one first.
- Add a checkpoint before the final fight. This should win back many of the 16, and some of the 31, who now get more practice.
- Mark the right door. This speaks to the 12.
- Weaken the fight, only if the first two are not enough.
She changed one thing at a time, and counted each time. Fixing all three at once would raise the number, but she would never learn which one did it. Our guide to comparing versions covers that.
Two honest limits. A session is one sitting, so a player who gives up today and wins tomorrow counts as a loss today and a win tomorrow. And completion does not say how hard a level is. A level that everyone finishes after 30 tries has a 100% completion rate and may be miserable. So read it next to retries and first-try wins.
Paste this to your AI
Explain the completion rate of one level in my game: [level]. The event log is in [file or folder]. The share I want to finish is [for example: 70%].
1. Read the level and draw how a player gets through it as a Mermaid state diagram, including every way of losing and every way of leaving without losing.
2. Count sessions that started and finished the level. Show the rate with its 95% range, and the same for the level before. Use get_recipe with difficulty-curve if Spielerando is connected.
3. Split those who did not finish into groups by what they did before leaving: lost many times, lost a few times, never lost. Show the counts, and which part of the diagram each group ended in.
4. For each group, say what the level does there, with the exact place in the level, and what would help that group.
Put the ideas in order, cheapest first, and tell me which group each one is for. Do not guess why people left if the log cannot show it. Say what I should watch for when someone plays. If a group has under 30 players, treat it as a hint only.
Or skip this and connect the MCP: your AI then finds all of this by itself.Connect the MCP
How to check it worked
- Add up the groups. Finishers plus each group must equal everyone who started.
- Play as each group. Lose three times. Lose once. Then try to leave without losing. Each one should show you something.
- Compare the range with your target, not the bare number.
- Measure after each change. If a change was meant for one group, that group should shrink.