Consider the following example. The operational precipitation type algorithm predicts a 4-hour period of reezing rain (bottom line of the 3rd trace). The alternate NCEP algorithm (middle, the Ramer algorithm (top) predict all snow. The explicit % of frozen precipitation from the model microphysics (green line) is at 100% during the period in question.
Three of the four members indicate snow. (Note: the % of frozen precip only distinguishes snow/sleet vs. rain/freezing rain.) Why is the operational algorithm an outlier? It is designed to have a high probability of detection of freezing rain at the expense of missing some wet snow events. There have been documented cases in which freezing rain was observed even though no temperature in the observed sounding was above 0 degrees C; supercooled water processes are likely responsible. The scheme therefore rules out snow if there is substantial area in the sounding with a wet-bulb temperature greater than -4 degrees C. Consider the forecast sounding for 07/12z, when the operational algorithm predicts freezing rain.
There is no temperature in the sounding above 0, but there is clearly a deep layer in which the wet-bulb temperature will be greater than -4, so the algorithm rules out snow! The alternate algorithm (middle) uses an area check based on 0 degrees C instead of -4. It will likely detect more wet snow events but also miss some freezing rain/sleet events. Information on the Ramer scheme can be found here. The % of frozen precip at 100% strongly suggests snow (or sleet) as the dominant type. In any case in which there is disagreement among the algorithms, close inspection of the forecast soundings as well as vigilant monitoring of local observations is clearly warranted.