One real dataset carries the whole module: the UCI "Diabetes 130-US Hospitals" file, 101,766 visits, 50 columns. It opens with pandas confidently reporting zero missing values in a column that is 97% empty — and every later idea (cleaning, describing, seeing, chance) is the next tool needed to answer one running question: which diabetic patients come back within 30 days, and what does the data actually say about them?
This hospital-readmission dataset was chosen for its verified messiness and real stakes, not by request — swap in your own dataset for the project's optional stretch if you'd rather work with something else.