This paper focusses on stochastic printing failures, such as microbridges in spaces or randomly missing contacts. We quantify such failures with a metric we call NOK (not OK), a metric that essentially represents the failure probability. We measure this NOK-quantity from SEM image analysis, using an in-house software package called Stochalis. We will argue that the most fundamental dependency of this failure probability is its CD dependency: NOK(CD). Using currently available CD-SEM or e-beam inspection tools, it is now possible to measure this NOK(CD) dependency down to the ppm-ppb level. This is not sufficient to prove or disprove yield, but the NOK(CD) function is an excellent tool to compare materials and conditions, i.e. to quantify sensitivities to patterning conditions and show directions for improvement. We will illustrate this with examples on the impact of dose, pitch, resist, etch and illumination mode. We will also show how CD non-uniformities, from very local to global, further affect the local failure probabilities. Finally we will argue that stochastic failure probabilities and CD-non-uniformities together, place practical resolution limits both on CD and on pitch. These limits are not absolute (which is why we call them ‘practical limits’) as they depend on the patterning settings and materials used, but they nevertheless need to be considered very carefully when setting up EUVL processes.