Blog
Articles, technical deep-dives, and updates from the Nardex team.
Analyzing long narde games from MAT files
Long-narde games saved as MAT files can be reviewed in the browser: import .mat into Nardex and get an equity timeline, move-quality labels, PR, and cube-decision flags.
TD-Gammon to Modern AI: 30 Years of Reinforcement Learning
How Tesauro's TD-Gammon (1992) launched a research direction that led to AlphaGo, AlphaZero, and today's stochastic-game engines. The history, the architectures, what changed, what stayed the same.
Training a Backgammon Neural Network: A Practical Guide
What an actual training pipeline for a backgammon or long narde engine looks like end-to-end: position generation, rollout evaluation, supervised training, tournament gating, model promotion, reclassification of older positions. With concrete numbers and choices from our production loop.
Stochastic Games and RL: Handling Chance in Backgammon and Narde
Why randomness matters for reinforcement learning. The chance node, expected value over dice rolls, variance reduction, the credit-assignment problem in stochastic games, and what backgammon and long narde teach us about agents that play dice games.
Reinforcement Learning for Board Games: A Primer
A practical introduction to reinforcement learning specifically through the lens of board games: states, actions, rewards, policies, value functions, and why backgammon was a watershed for the field. With concrete examples in backgammon and long narde.
Why We Built a New Backgammon Engine in 2026
An origin story: why gnubg and XG aren't enough for modern apps, what we shipped instead — Rust + ONNX + CUDA, neural-net training on real games, and an HTTP API for both backgammon and long narde.