Every patient gets a different gum.

The problem §02 identified - that dysgeusia is a family of presentations, not a single disorder - means any single-formulation gum will serve the average patient and miss many others. The algorithm addresses this by decomposing the symptom space into its mechanism components and routing each component to the ingredient strategy designed to address it.

The logic is five steps. Inputs (six patient variables) feed into base scores for each ingredient strategy. Each variable applies a delta weight to the relevant strategy or strategies. An oral hygiene multiplier then scales all four scores globally. The four scores are normalized to sum to 100 percent. The result is a relative formulation recommendation: how much of the gum's ingredient budget to allocate to each strategy for this patient.

Step 1 - Start equal. Every cancer patient starts with the same baseline: 25 points assigned to each of the four ingredient strategies. No assumptions about who they are yet. Equal allocation is the fallback if no clinical information is available.

Step 2 - Adjust for drug class. The chemotherapy regimen is the strongest signal. Platinum-based drugs (carboplatin, cisplatin, oxaliplatin) activate two mechanisms at once - metal-ion binding and lipid peroxidation - so both trigeminal stimulation and the umami bitter masker are boosted. Taxane-class drugs (docetaxel, paclitaxel) produce the most severe taste-bud apoptosis; sour/tart is the most resilient remaining signal and is weighted up. Anthracyclines (doxorubicin, epirubicin) are most associated with nausea co-morbidity, so anti-nausea support becomes the priority.

Step 3 - Adjust for bloodwork. Elevated iron amplifies the metallic signal - trigeminal and umami bitter masker both increase to compete with it. Zinc deficiency compounds taste-bud apoptosis by removing the cofactor gustatory receptor cells need - sour/tart increases as the most reliable remaining signal, and anti-nausea support gets a secondary boost.

Step 4 - Adjust for time on treatment. Early weeks (1–2): baseline. Mid-treatment (3–6 weeks): symptoms established - trigeminal and umami both climb. Late treatment (7–10 weeks): accumulated bitter signal - umami and sour take over. Post-treatment (10+ weeks): recovery arc - sour dominates as metallic signal fades and receptor regrowth begins.

Step 5 - Adjust for nausea severity. Each step up in nausea severity (none → mild → moderate → severe) adds 10 points to the anti-nausea strategy. Severe nausea adds 30 points - the largest single-input delta in the algorithm, reflecting how completely nausea can block eating.

Step 6 - Apply the oral hygiene multiplier. Oral hygiene amplifies whatever is already happening, not a specific pathway. Poor oral hygiene: all four strategy scores × 1.18. Fair: × 1.09. Good: no change. The formulation direction stays the same - everything intensifies proportionally.

Step 7 - Normalize to 100%. All four strategy scores are divided by their total and converted to percentages. The result always sums to exactly 100%: the share of the gum's ingredient budget allocated to each strategy for this specific cancer patient.

The weights in this algorithm are author-calibrated from the literature, not statistically derived from patient outcome data. Each delta weight was set by asking: given the mechanism this variable activates, and the ingredient strategy designed to address that mechanism, what is a reasonable intensity adjustment? The values are informed estimates, not regression coefficients.

Two design choices are worth making explicit. First, only relative allocations matter. The normalization step means a score of 60/40/30/20 and a score of 120/80/60/40 produce the same output. The algorithm defines direction and proportion, not absolute ingredient concentrations. Translation to concentrations is the formulation chemist's job.

Second, the algorithm is deterministic: the same six inputs always produce the same output. This is a deliberate choice. Determinism makes the algorithm reproducible, auditable, and explainable to a patient or clinician. Stochastic or ML-based approaches could potentially produce better-calibrated weights given outcome data - but they require outcome data that does not yet exist, and they trade explainability for precision.

What a proper validation study would need to do. The ideal design is a regimen-stratified Phase I/II trial with three arms (platinum, taxane, anthracycline) and roughly 20–30 patients per arm. Each patient would complete a baseline CiTAS assessment before starting treatment and receive a formulation recommendation from the algorithm at the time of enrollment. The primary endpoint would be change in CiTAS Dimension II score (qualitative metallic/bitter distortion) at 6 weeks. Secondary endpoints: appetite sub-score, nausea sub-score, body-weight stability. A platinum-specific arm would test the umami competitive-displacement hypothesis directly - comparing patients receiving maximum umami loading against a matched group receiving the trigeminal-only formulation. Given the high intra-patient variability identified in §02, an n-of-1 adaptive design may be more appropriate than a parallel-group design for the later phases: each patient serves as their own control across sequential formulation adjustments.