Built-in mixtures: named compositions of several molecules at fixed dry-basis mass fractions.

Each member is stored in the JSON as nothing but a molecule id and a mass fraction, and is RESOLVED HERE against molecule_database.json so the response carries that molecule's own molecular weight, density, pI, solubility and log P. That is the whole point of the endpoint: a mixture must never reach a client - or a mass balance - as one averaged pseudo-component. The members span the real property range (WPC80 runs 14 kDa to 150 kDa; citrus peel 180 Da to 162 kDa), and every separation model in the app is written against a single species, so it has to see the species rather than their average.

A member whose id is not in the molecule database is DROPPED rather than served with empty properties, and the mixture records what was dropped. A component with no molecular weight or density cannot close a mass balance, and serving one silently is worse than serving a shorter list.