Read Your Game's Difficulty Curve From Player Data
Write down how hard you want each level to be, then compare that with the share of players who win it on their first try. A sudden fall against the level before is what you fix first, and everything else can wait.
What does a real difficulty curve look like?
Ember Steps is an invented platformer, and these numbers are made up for this example. Before release, its maker wrote a plan: the share of players who should beat each level on the first go. Each player counts once per level, so someone who loses forty times cannot drown out everyone else.
| Level | Players | Won first try | Very likely between | Plan |
|---|---|---|---|---|
| 1 | 200 | 94% | 89% and 96% | 90% |
| 2 | 190 | 84% | 78% and 89% | 85% |
| 3 | 178 | 79% | 72% and 84% | 80% |
| 4 | 170 | 39% | 32% and 46% | 75% |
| 5 | 150 | 78% | 71% and 84% | 70% |
| 6 | 140 | 69% | 60% and 76% | 65% |
| 7 | 128 | 63% | 54% and 70% | 60% |
| 8 | 115 | 57% | 48% and 66% | 55% |
The third column is the curve you drew in your head, tested. For levels 1, 2, 3, 6, 7 and 8, the plan sits inside the range. Those levels are doing what they should, so leave them alone, even where the number is a few points off. That is what the range is for: it keeps you from fixing noise.
Level 4 is another matter. It falls 40 points against level 3, and the two ranges are nowhere near each other. People call that a spike. The plan said 75%, and even the top of the range is 46%.
Level 5 also breaks the plan, the other way: 78% when 70% was the aim. Do not make it harder yet. Only the 150 who got through level 4 are in this row. Level 4 filtered out the people who struggle, so what is left is a better group. This is called survivorship bias: you look only at those who made it, and ignore those who did not.
Why was level 4 so hard?
The numbers show where, not why. The maker played level 4 and found two new things at once: moving platforms and twice as many enemies. Level 3 had neither.
So the change was to split them. Moving platforms come first in a calm room, and the enemies arrive one level later. That is a single, small change to test.
How smooth should a curve be?
It does not have to be smooth. A dip after a boss gives people a breath. The plan is yours, and it can have bumps on purpose. What matters is that you meant them.
Game designer Jenova Chen describes why this matters. If the challenge goes beyond a player's ability, it overwhelms them. If it falls short, they lose interest. He also points out that different players have different comfort zones, so one line never suits everyone. Your numbers describe the group, not any single person.
With fewer than about 30 players on a level, do not read any of this into it. For that, use how many players you need.
Paste this to your AI
Show me how hard my levels really are. The event log is in [file or folder].
1. Ask me for the share of players who should beat each level on the first try. If I do not know, help me write it down before you look at any data.
2. Draw how one level plays out as a Mermaid state diagram: start, win, lose, try again, give up. Check that the log has an event for each of these.
3. Use get_recipe with difficulty-curve if Spielerando is connected. If not, count each player once per level. For each level give: players, first-try wins, 95% range, the plan.
4. Mark levels where the plan lies outside the range, and spikes, where the fall against the level before is 15 points or more and the two ranges do not overlap.
5. Point out where later levels only look easy because the weaker players have gone.
For the worst level, read its setup and list what is new compared with the level before. Suggest one change. Leave alone every level where the plan is inside the range. If a level has under 30 players, call it too thin.
Or skip this and connect the MCP: your AI then finds all of this by itself.Connect the MCP
How to check it worked
- Play level 3, then level 4. Does the jump feel the way the numbers say?
- Compare again after the change. Level 4 should rise, and level 5 should come down a little, as weaker players now reach it.
- Compare with your plan, not with your last result. Otherwise you end up chasing your own numbers.