Here's a further explainer why this is needed. One might ask, why not just use the 'nearest' mode. Is that not the same? This block is primarily useful when doing the 'floor' rounding. You don't want it to accidentally round down say a value of 3.9999999 to the previous sample if the difference was just floating point error. If you do then depending on the situation you might get a different length in the same situation. This is intended to make it more robust against this scenario.