A prediction may turn out to be incorrect without being pointless. Its value also depends on the possibility of reconstructing why the result changed. Keeping the initial version makes it possible to distinguish the error of estimation from the events that took place after the decision.
To start, I would choose a precise horizon and the same unit of measurement. Orders, pieces and sales value answer different questions. Even the timing of the forecast counts: an estimate formulated before knowing a promotion is not comparable to a later one.
A hypothetical example concerns requests for a service on the weekend. The forecast can systematically underestimate days with a local event. Annotating this element makes it possible to improve reading, without attributing any difference to the ability of the people involved.
I propose a periodic review of the larger errors and those repeated in the same direction. The purpose is to correct recognizable assumptions. An estimate expressed as an interval can help when variability is high, as long as its meaning remains understandable to those who organize the work.