Today a blend is read component-wise: each molecule's profile is predicted independently and combined arithmetically (see mixture_to_molecule). Real mixtures do not work that way — components suppress and enhance each other, which is most of the craft in formulation.
This needs panel data on MIXTURES (perceived profile of a blend vs its components), which no public dataset provides — hence data-gated, like odour intensity. Worth stating the requirement precisely so it can be scoped if a customer brings their own panel data, and so the current arithmetic combination is not mistaken for a mixture model.
Today a blend is read component-wise: each molecule's profile is predicted independently and combined arithmetically (see
mixture_to_molecule). Real mixtures do not work that way — components suppress and enhance each other, which is most of the craft in formulation.This needs panel data on MIXTURES (perceived profile of a blend vs its components), which no public dataset provides — hence data-gated, like odour intensity. Worth stating the requirement precisely so it can be scoped if a customer brings their own panel data, and so the current arithmetic combination is not mistaken for a mixture model.