Random numbers generated on a quantum generator require postprocessing to eliminate hardware distortions. The quality of sequences is assessed by NIST tests, where the key parameter is the threshold value, exceeding which determines the randomness of the sequence. The paper examines the effect of the threshold value on the percentage of passing NIST statistical tests. Postprocessing is carried out by the method of lexicographic block numbering. It is found that increasing the threshold tightens the criterion for estimating the randomness of the sequence, shifting the optimal values of the size of the postprocessing block and the size of the evaluation window. It is shown that using a standard evaluation criterion may not be sufficient. By ensuring that tests pass for most sequences, a small evaluation criterion can mask weak correlations in quantum noise, leading to an overestimation of randomness and reducing reliability in cryptographic applications.
Mavkov et al. (Mon,) studied this question.