"Label Trust Estimate" or "Markov Quality Score" metric, can i know how this metric is calculated. Does the distribution of labels within the dataset plays any role in influencing this metric's outcome.