Post-selection inference, i.e. model building and doing inference on the same data set where the inference does not account for the model building stage.
Either: Given a data set and no predetermined model, a model is built based on the patterns found in the data set.
Or: Given a data set and a model, one often finds that the model is inadequate. The model is adjusted based on the patterns found in the data set.
Then: The model is used for inference such as null hypothesis significance testing.
The problem: The inference cannot be taken at face value as it is conditional on the data set due to the model building stage. Unfortunately, this fact often gets neglected in practice.