It sounds like you want to combine the phrase into a single coherent concept or term. Here are a few possible interpretations:

DNGs can be used to filter outputs from hardware random number generators (HRNGs). Hardware sources (like thermal noise or avalanche noise) can sometimes produce "stuck" bits or predictable stretches due to physical interference. By implementing a discard protocol, the system can reject values that exhibit statistical anomalies or fail "health tests," ensuring that only high-entropy data reaches the user.

Args: num_discards (int): The number of discard numbers to generate.

A DNG solves this by discarding the "overflow" values that cause the bias. If the generator produces a number that falls into the "leftover" range, it throws that number away and tries again. While this reduces the total yield of numbers, it ensures that the resulting distribution is mathematically perfect.