Adaptive Difficulty System

This plugin watches how your players actually perform; their reaction speed, error rate, actions per minute, and combo consistency; and uses a lightweight neural network to continuously tune difficulty in real time. No hand-tuned curves, no static tiers. The system learns what challenge level keeps each player engaged and adjusts automatically, every few levels.

Includes ready-to-drop-in support for Tetris-style and Match-3 games, with a clean extension point for any game type. An in-editor debug overlay lets you watch the agent's decision-making live during playtesting.

Features

• Reinforcement learning agent (DQN-lite) runs entirely on-device; no server required
• Targets a configurable "sweet spot" completion rate to keep players in flow state
• Works across sessions: agent state can be saved and restored between play sessions
• Built-in replay buffer for stable, low-variance training
• In-game debug overlay showing Q-values, epsilon, and live stats
• Full source included; Unity C# + Python prototype for offline experimentation

Requires: Unity 2021.3+

Note: Includes ready-to-wire integration examples for Tetris-style and Match-3 games — drop the manager component into your existing game and connect the provided hooks.  Does not Include Tetris and Match-3 games.