MX Anticheat (originally Mineland Xcalibur)

Advanced entropy analysis powered by local Recurrent Neural Networks (RNN).

MX is a free, open-source, Zero-Dependency Machine Learning anticheat for Minecraft (1.8 - 1.21). Powered by powerful heuristic algorithms combined with machine learning on a self-written RNN (Millennium 5) basis.

---

🧠 The Millennium 5 Engine

MX is built upon the Millennium 5 self-written open-source library. It is a plug&play fully custom Deep Learning framework written in Java for CPU.

Key Technologies:

• Local Processing: No external Python scripts, no GPUs, no cloud API calls required. Everything runs asynchronously on your server's CPU.
• Architecture: Stacked Bi-LSTM (Bidirectional) with Attention Mechanism, Layer Normalization, AdamW.
• Model: quark-e-1.0-56k-public (56,000 parameters) — capable of detecting many 'clever' hacks.
• Advanced Heuristic: From basic to advanced heuristics, perfectly complemented by machine learning.

---

🚀 Getting Started

Installation

1. Download the latest release .jar.
2. Drop it into your plugins folder.
3. Restart the server.
4. Done. The pre-trained weights are loaded automatically from JAR.

Training Your Own AI (Optional)

MX allows you to train the neural network directly on your server to adapt to new cheats.

1. Collect Data:
   • Record a cheater: /mx dataset cheat <player>
   • Record a legit player: /mx dataset legit <player>
   • Play for ~30-60 minutes to gather samples.
2. Train:
   • Run /mx train <model_index> <epochs> (e.g., /mx train 7 16).
   • The plugin will run the AdamW optimizer (Backpropagation) in the background.
3. Deploy:
   • The new weights are saved automatically.

---

💻 Commands

Command	Permission	Description
/mx	mx.admin	Main help menu.
/mx alerts	mx.admin	Toggle violation alerts.
/mx stats	mx.admin	View global ban/flag statistics.
/mx activity <player>	mx.admin	Generate a visual graph of player rotations (Pastebin).
/mx dataset <mode> <player>	mx.admin	Start recording samples for ML training (legit/cheat/off).
/mx train <index> <epochs>	mx.admin	Start async training of the neural network.
/mx ml <index> <param> <val>	mx.admin	Tweak ML hyperparameters (Learning Rate, Dropout, etc.) live.

---

📂 Configuration

MX is highly configurable.

• checks.yml: Enable/Disable specific checks, tune statistical thresholds.
• config.yml:
  • prevention: Set the harshness of lag-backs (0 = Silent, 3 = Aggressive).
  • ignoreCinematic: Reduce false positives for cinematic camera users.
  • rotationsContainer: Enable storing movement history for the /mx activity command.

---

🤝 Contributing

MX is open-source under the Unlicense. You are free to fork, modify, sell, or do whatever you want with the code.

We need your help:

• Submit datasets of new private cheats.
• Share trained .dat model weights.
• Improve the math logic.

---

Created by Kireiko Oleksandr (pawsashatoy)