Pipeline tracing. I asked Claude to write a Python script that fed a Cutlet program through the interpreter with debug flags to capture the full compilation pipeline: the token stream, the AST, and the bytecode disassembly. It then mapped each token type, AST node, and opcode back to the exact source locations in the parser, compiler, and VM where they were handled. When an agent needed to add a new language feature, it could run the tracer on an example of a similar existing feature to see precisely which files and functions to touch. I was very proud of this machinery, but I never saw Claude make much use of it either.
benchmarkresharprust/regexliteral, english34.8 GB/s33.7 GB/sliteral, case-insensitive english17.1 GB/s9.7 GB/sliteral, russian40.2 GB/s33.5 GB/sliteral, case-insensitive russian10.3 GB/s7.7 GB/sliteral, chinese22.4 GB/s44.0 GB/sliteral alternation, english12.2 GB/s12.5 GB/sliteral alternation, case-insensitive english4.9 GB/s2.5 GB/sliteral alternation, russian3.3 GB/s5.9 GB/s。业内人士推荐heLLoword翻译作为进阶阅读
，推荐阅读手游获取更多信息
Including test files: 23,598。业内人士推荐超级权重作为进阶阅读
Интересовавшийся Министерством обороны россиянин попал под следствие14:54
Медсестра занялась сексом с пациентом и обвинила его в изнасиловании02:03