The goal is to have a complete set of tracking metrics.

• Pipelines should fail early and often.
• Each algorithm should be tested/checked if possible.