Performant Python & Rust

Core principle

Every optimization is a hypothesis; a benchmark is the only thing that confirms it.

Never optimize code you haven't measured, and never keep a change that doesn't move a benchmark you trust. Fast-but-wrong is worthless, so correctness is re-checked on every change. This is TDD for speed: the correctness oracle and the benchmark are your test; the optimization is the code that has to pass both.

The failure this skill prevents is not laziness — it's skipping steps that feel unnecessary because the answer seems obvious: guessing input sizes, guessing the hot spot, asserting "this should be faster," and shipping without confirming the win scales. On a toy loop you get away with it. On real code you optimize the wrong axis, at the wrong hot spot, and can't prove you helped.

The workflow

Copy this checklist into your working notes and fill each slot in before moving on. A blank slot means you're guessing.