That's a cool project. Amazing that the bytecode interpreter is slower than the AST interpreter also here. Wasn't there a paper by S. Marr's team which came to the same conclusion?