Introduced on day 1
Prediction
A best guess for an outcome you have not observed yet (or for a unit you have no label for), judged by how close it lands on new data. It says what will happen, not why or what would happen if you acted.
In detail
A prediction answers "what will the value be?" for a specific case: how many passengers will not show up on ZX104 next Friday, or whether this booking will be cancelled. It is judged by its error on data the model has not seen, evaluated the way it will be used - usually forward in time.
A good predictor can rely on any signal that is available at decision time, including signals that are not causes. Suppose the booking channel predicts no-shows well in your data: the channel is a marker of who the passenger is, not a lever. That is fine for prediction and wrong for a what-if question.
A prediction is not a decision. The no-show forecast feeds an overbooking limit, and the limit also depends on the cost of a denied boarding versus an empty seat.