Hello Walid Ahmed

In the case of a prediction : At time t, you make a single prediction for time t+1. At time t+1, you compare sensor (algorithm here) result with your prediction and start by updating your state and covariance so your next prediction will converge.

In the case of smoothing/denoising : You start by taking the t-1 data and predict the value at time t. You immediately compare the predicted value with the value of your algorithm.