Before you build a model, you need to interrogate the data. Each case shows you a dataset with a hidden problem. Spot the trap before it poisons your model. Garbage in, garbage out — but only if you don't catch it first.