Some devs still mix up tests/evals/benchmarks, although these are different things. Wrote a small guide here, with a duck-led quiz at the end.
https://shcherbak.ai/blog/tests-evals-and-benchmarks/
Shcherbak AITests, evals and benchmarks: what each one tells you - Shcherbak AIA test checks code, an eval measures your agent on your own tasks, and a benchmark compares models on someone else's. What each of the three can tell you, what it cannot, and which one decides a release.
commit 872270f