Uncertainty estimation with a small number of measurements, part I: new insights on the t-interval method and its limitations