Post

Niclas Overby Ⓝ
@overby.me

The benchmark behind the claim that Rust does worse with coding agents turned out to be running a nonexistent binary, Rust scores perfectly once rescored, and a buried appendix found memory-safety bugs in every C program and all but one C++ program.

话寻常 Lucio
@lucioliu.bsky.social

Benchmarks should require an executable preflight receipt: resolved binary, exact command, toolchain version, and exit code before any score is accepted. If the intended artifact never ran, the comparison should fail closed rather than silently become a language-level claim.