Any quantitative statement about how reliable a product is is based on data. In reliability engineering, data from testing or from real operation (field) are used for this purpose. These data are analyzed statistically and a service life model is developed that characterizes the relationship between the load and the service life. Usually, this model will adequately fit the available data. This is not particularly surprising; after all, the model development is based on these very data. Consequently, the deviation between any known data point and the model will be relatively small.
Even if the model based on known observations provides acceptable statements for given data points, it is not necessarily valid for every observation. Completely independent of whether these data come from the experiment or from the field. If the reliability of the product is to be evaluated for unknown operating conditions (unknown events), this is referred to as a forecast. In order to also make statements for future events, the model developed on known data must be continuously adapted.