Nutrition@Home is a deterministic clinical triage engine. Every threshold, formula, and data source is documented here so that doctors and dietitians can audit it in full. Nothing in the engine is generated by AI at runtime.
Recipes are assembled from a walled garden of verified clinical sources — never generated by AI. The ingredient database is separate from recipes so carbohydrate sources can be swapped (brown rice ↔ quinoa ↔ barley) without breaking the math. Every ingredient carries macro and cardiovascular micronutrient values (fiber, potassium, magnesium, trans fat, omega-3).
Mifflin-St Jeor for the energy baseline. Strictest-threshold-wins across all active conditions. Recipes are dynamically scaled to the patient's exact per-meal calorie target, then true nutrient totals are aggregated from the scaled ingredient weights. Per-meal absolute caps: sodium ≤ 800 mg, glycemic load ≤ 10. Any trans fat is a hard fail. Missing nutrient data returns a hard fail — never silently defaulted to zero.
Dynamic meal division. For diabetic patients, the auto-scaler reduces carbohydrate portions in 5% increments until the glycemic load is safe, then proportionally reinvests the lost calories into the calorie-dense protein and fat ingredients already in the recipe. Escape hatches at 10 iterations or the 300 kcal cachexia floor prevent clinical underfeeding. Eating out tightens caps asymmetrically — 30% off hazard nutrients, 10% off fuel nutrients.
Ingredient and recipe values are drawn from, and cross-referenced against, the following publicly accessible, science-backed sources: