Unit 10 · lab

Evidence Check: Data Science Evidence Dossier

Finish the course by defending one small data-science investigation. You may use a public dataset from an earlier lesson or a teacher-supplied synthetic dataset.

Your dossier

Include:

Question and stakeholder
Dataset and provenance
Ethics/privacy check
Preparation decisions
Analysis method
Visualization choice
Main finding
Uncertainty and limitations
Deployment/tool decision
One rejected alternative
Next question

You do not need an Azure account to complete this evidence check.

Bridge from data to AI

Open From Data to AI. Remove the battery feature or lower the dataset-quality control. The tiny neural network still returns a number. Explain why a model output does not cancel missingness, weak provenance, sampling problems, or evaluation limits.

Defense

Be ready to answer:

Which decision in your workflow had the greatest effect on the result?
What evidence would make you change your conclusion?
What can your analysis not claim?

A successful dossier is not the one with the fanciest chart. It is the one whose reasoning can survive inspection.

Done when

Every major conclusion points back to evidence, every important limitation is visible, and your next question follows from the investigation rather than appearing from nowhere.