Chess

NOTE: MoveGen has been tested up to depth 7, and achieves perfect accuracy in perft.

Benchmarking:

• luau tests/miniMax.test.luau --codegen (codegen nearly doubles performance by applying native luau)
• luau tests/perft.test.luau --codegen For any of the above commands you can add --profile && python3 tools/perfgraph.py profile.out > profile.svg to save flamegraph as SVG.

Train Magic Numbers:

• g++ -O3 -march=native -std=c++20 -o tools/magicNumbers tools/magicNumbers.cpp
• ./tools/magicNumbers

UCI

Run rokit add luau-lang/lute@0.1.0-nightly.20260228 to get Lute.

NOTE: With my experience, Lute does not have Native Luau, performance will be worse.

• lute tools/uci.luau

You could not use go depth 5 and get a eval, bestmove, and move sequence.