That's one important takeaway. But we don't actually know what (synthetic) D2 and (food derived) D3 are doing differently in the body. This is exacerbated by: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4210929/

That there's this much nuance just with this one ingredient is a microcosm of the overall flaw in Soylent's approach.

Optimizing for just the legible dimensions of a poorly understood domain will not result in good outcomes.

For mealsquares we take consilience seriously and combine the inside and outside view (see: reference class forecasting) on human nutrition to try to establish reasonable confidence intervals around nutrition models, and even then we aren't stupid enough to think or claim that mealsquares are just as good as fish, fruit, and vegetables.