Dynamic Difficulty

Dynamic difficulty watches how the player is doing and moves the game one small step easier or harder, aiming for a steady win rate. It judges only after enough results, waits after every change, stays inside a floor and a ceiling, and leaves a gap where a player who is doing fine is left alone.

Also called: Adaptive difficulty, Rubber banding, Difficulty scaling, Auto difficulty.

Paste this to your AI

Copy it into Claude, Cursor or ChatGPT. It already has good values and the usual mistakes in it.

Add "Dynamic Difficulty" to my game.

What it should do: Players struggle and coast at different times. Nudging the challenge toward what each player can handle keeps a stuck player in the game and a strong player interested, without asking them to pick a setting.

Build it exactly as this state machine. First check that it fits my game, then build it like this:

```mermaid
stateDiagram-v2
  [*] --> settling
  settling: Gathering results
  watching: Watching
  easiest: At the easiest
  hardest: At the hardest
  settling --> watching: enough results
  settling --> easiest: enough, at the floor
  settling --> hardest: enough, at the ceiling
  watching --> settling: changes, or returns
  easiest --> settling: toughens, or returns
  hardest --> settling: eases, or returns
```

Use these starting values:
- target: 0.7
- tolerance: 0.2
- alpha: 0.2
- minResults: 6 results
- maxEase: 3 steps
- maxToughen: 2 steps
- stepSize: 0.1 x

Watch out for these common mistakes:
- The game swings up and down: Tell your AI: after a change, set the average back to the target and wait for several new results before judging again.
- Easing off with no floor: Tell your AI: set an easiest level that still has some danger, and stop easing there.
- One bad night remembered forever: Tell your AI: when the player returns, keep the level but wipe the recent results, so the game starts watching afresh.
- Unfair deaths count as lost fights: Tell your AI: only report a result for fights that were fair, and leave out ones that ended for outside reasons.
- A quiet zone is missing: Tell your AI: only react beyond a band on each side of the target, and do nothing inside it.
- The change hits in the middle of a fight: Tell your AI: read the strength once when an encounter starts and keep it until the encounter ends.

Do not change anything else in the game. Do not claim anything without proof: model what it takes, and show me what your solution rests on and how I can see that it works. If you cannot show something, say you do not know. When you are done, tell me in plain words how to try it.

Or skip this and connect the MCP: your AI then finds all of this by itself.Connect the MCP

What it does

Players struggle and coast at different times. Nudging the challenge toward what each player can handle keeps a stuck player in the game and a strong player interested, without asking them to pick a setting.

How it runs, as a diagram

Everything it can be in, and what moves it on. Your AI builds from exactly this.

Good starting values

What usually goes wrong

How to check it works

• A new game with the starting values. The player loses 5 fights in a row: The difficulty has not changed yet.
• A new game with the starting values. The player loses 6 fights in a row: The difficulty goes one step down.
• A new game with the starting values. The player wins 6 fights in a row: The difficulty goes one step up.
• The difficulty has just gone one step down. The player loses 5 more fights, then a sixth: It holds for the first five and goes another step down on the sixth.
• A player who wins seven fights in ten, over 300 fights. You follow the difficulty: It never leaves normal, and the average stays between 0.5 and 0.9.
• A player who loses every fight. 200 fights have passed: The difficulty sits at the easiest level and goes no lower.
• A player who wins every fight. 200 fights have passed: The difficulty sits at the hardest level and goes no higher.
• A game that has eased down to the easiest level. The player then wins 30 fights: The difficulty comes back up from the floor.
• A game one step down and 4 losses into the next stretch. The player comes back to the game: The level is still one step down and the streak is wiped.
• A step size of 0.1. You ask for the enemy strength at levels 0, 1, -1 and 2: It is 1, 1.1, a little over 0.909 and 1.21.

Works well with

• Health and Damage: Scale the enemies' damage and health by the strength, read once when the encounter starts.
• Bad Luck Protection: Both smooth out what the player feels. Keep them on different things, the fights for one and the drops for the other.

Clashes with Difficulty Settings: A player who chose hard may find the game quietly easing off, and a player who chose relaxed may find it toughening up. Let the adaptive part work only inside the chosen setting, and tell the player when it is on.

Clashes with Meta-Progression: Upgrades earned over many runs make the player stronger, and an adaptive game answers by making the game harder, so the upgrades seem to do nothing. Pick one to lead. Base the adaptive steps on a level that already includes the upgrades, or keep the steps small.

Common questions

Where it comes from

We could not find a reliable origin for this one yet.

Where this mechanic comes from has not been researched yet.

Games are named only as references. Names and marks belong to their owners, and nothing here is endorsed by them.

What was checked

• 15 tests pass on the reference code.
• The state diagram was checked against the code, at the default values and at every listed low and high value.

This is what separates the blueprint from a catalogue entry: the values on this page come from code that runs.

Your AI gets the code

Tested example code comes through the Spielerando connection, so your AI can use it right away. How to connect it