Messy real documents in, clean validated records out, graded against labels you never see. This week is the part beginners skip and professionals never do: hand-label a set of rows yourself, split it into teaching examples and a held-back test, and write the eval your pipeline must pass before the pipeline exists. Real data will lie to you in layers; this is the week you learn to catch it.
It ends with
A labeled set, an eval written by you, and a pipeline taking shape against it.