Modelling Tenon Matcha from the ingredients up

Hypothesis

If ingredients, preps, builds and menu products are separate records connected by reference, nutrition can be calculated automatically through the chain. A change to one ingredient should then propagate correctly to every drink that uses it, without re-entering a single number.

Method

We modelled one complete drink, Tenon Matcha, with placeholder values:

• Seven ingredients: water, ice, UniSoy, whey concentrate, soy protein isolate, ceremonial matcha and full-fat milk. Each declares nutrition per 100 g or per 100 ml, with a named source for every value.
• Three preps: the high-protein milk bases. Each states its inputs, yield, storage and procedure.
• One build: Tenon Matcha, hot and iced, with the milk base as an interchangeable slot. That gives six calculated variants.
• One menu product pointing at the build.

The data lives in plain text files in a public repository. Every record is checked against a JSON Schema and a set of cross-record rules before anything is calculated.

Results

• Changing the protein content of the whey propagates through both whey-based milk preps, the four affected drink variants and the menu product. The soy-only variants do not move.
• Doubling a prep doubles quantities, yield and totals while per-100 g values stay identical.
• Missing data is never treated as zero. A drink containing matcha (which has vitamin K data) and soy milk (which does not yet) reports vitamin K as a lower bound, not a total.
• Allergens and diet flags are derived. The soy-and-whey base can never be accidentally labelled plant-based.

Next

Replace the placeholders with real label data for each ingredient, measure the actual yields and densities at the bar, and enter the Singapore HPB reference intakes.