Review the evidence
A useful optimisation arrives with evidence. When you review a candidate, you see the code change next to its measurements and test results. Each result belongs to one specific Version, and the Baseline is your unchanged code measured on the same Runner.
Look at each Metric on its own. Two objectives often compete: in the video's example, one Version of a particle simulation is faster but uses more memory, and another is leaner but slower. Check the spread across repeated runs as well, since small differences can be noise.
Then read the diff: what changed, why it should help, and whether the configured checks passed. The change you pick is available as an Artemis Branch for review and publishing. You decide what merges, and the Project keeps the evidence for later work.