An original paper which suggests a way of thought for semi-linear credibility theory development, founded on analysis of the functions of the observable random variables. This line of thought fits perfectly within the framework of the greatest accuracy credibility theory.