This is a really interesting and unique approach to building an NPS. The model factors in a wide variety of research about what is and isn't healthy. So instead of being based on a model that, say, low carb diets are what really matter, they rank foods based on a variety of evidence based models (e.g., low carbs matter, lipid hypothesis, seed oil hypothesis, etc.) and foods that score well across many models or "factors" get a good Compass score.

This seems totally useful given that there is so much controversy right now about what is and isn't healthy to eat. It's an "epistemically modest" approach to nutrition labels.

The outcome of the model is that I can know chocolate covered almonds are probably the healthiest desert and salmon is probably the healthiest animal protein without having to do too much research on my own.