Traditional elimination diets have poor completion rates and unreliable results. They require strict compliance, rely on informal pattern recognition, and can’t distinguish food reactions from confounders like poor sleep or high stress. A “cheat” invalidates the experiment. Results are subjective and hard to act on.
Confidente treats food sensitivity testing as a structured experiment. The core insight is that compliance is not the goal — logging is. Off-plan meals are additional data points, not failures.
Meal plans are generated using Latin square-inspired scheduling to ensure:
When a user flags multiple foods as suspected triggers, Confidente checks whether they share a sensitivity category. If A, B, and C are all high-histamine foods, the engine suggests adding D (another high-histamine food) to the test plan — but delays it until after the current washout window.
Foods are tagged with one or more sensitivity categories with severity levels:
| Category | Notes |
|---|---|
| Histamine | Fermented foods, aged cheeses, leftovers, alcohol |
| FODMAPs | Fermentable carbohydrates — onion, garlic, wheat, legumes |
| Salicylates | Many fruits, vegetables, spices — often missed because foods seem “healthy” |
| Oxalates | Spinach, nuts, chocolate, beets — common in “clean eating” |
| Lectins | Beans, grains, nightshades — relevant for autoimmune presentations |
| Glutamates | Tomatoes, parmesan, soy sauce, mushrooms — neurological symptoms |
| Capsaicin | All peppers including bell peppers — gut motility trigger |
Foods can belong to multiple categories (e.g. avocado is both high-histamine and high-salicylate).
Daily control variables are tracked to weight symptom scores:
High-stress or poor-sleep days discount food signal — the data isn’t discarded, it’s held more loosely by the model.
The model starts simple — regression of symptom scores against ingredient exposure frequency, weighted by daily control quality scores. Mixed effects modeling is the target once data volume warrants it.
The statistical logic is abstracted behind service objects so the underlying implementation can be upgraded without changing the interface.
The science is the engine. Users never see it. The UI exposes:
At-home finger-prick tests (DAO enzyme levels, IgG food antibodies, oxalate markers, gut inflammation) could close the loop between behavioral inference and biochemistry. Lab results would feed directly into the sensitivity category graph to boost or suppress hypothesis confidence.
This is explicitly out of scope for MVP but the schema is designed to accommodate it.