Could anyone from Agisoft clarify the intended confidence intervals of the 'accuracy' terms for camera and marker geolocations?

(some users seem to assume that these are 1 sigma (68% confidence interval), while others consider them to be 2 sigma (95% confidence interval)).