Chess Engine
Theoria is a chess engine based on Stockfish's search architecture with an NNUE evaluation network trained on Leela Chess Zero data. Instead of learning from iteration on Stockfish's evaluation, Theoria's network learns positional understanding from Lc0's self-play data.
Theoria combines Stockfish's efficient search with evaluation patterns that emerge through exploration of chess positions, with the aim to clarify the strategic value of chess positions in a way that is more transparent.
Theoria applies principles from information theory and complexity reduction to chess analysis. We treat the game tree not just as a space to be searched, but as a complex system where we can identify and prioritize the most elegant, low-entropy strategic pathways. It's about finding the signal in the noise. This leads to a design priority we call 'narrative coherence'. We believe the most valuable analysis for a learner isn't the most computationally complex, but the most conceptually parsimonious—the line that reveals the core, governing idea of the position.
This reflects a deliberate philosophical approach: that chess evaluation should use more coherent internal representations of chess knowledge, and not merely be a "black box" optimized for maximum play strength against other engines.