Assign one role before combining
Multi-component development begins by defining why each component exists. A component without a distinct hypothesis adds complexity without adding interpretability. For NOVA-4, the abstract roles are enkephalin-preservation research, FAAH-related lipid-signaling research and oxytocin-receptor-context research. These are target assignments, not assumed human outcomes.
Each assignment expands into measurable predicates: exact chemical identity, potency, selectivity, exposure, tissue distribution and a component-specific contribution endpoint. A component advances only when its data can be distinguished from the behavior of the other components. This prevents an attractive total score from hiding a nonfunctional or counterproductive element.
Enumerate before selecting
The current NOVA-4 design lattice contains six C1 candidates, two C2 candidates and two C3 candidates. Their Cartesian product produces twenty-four exact triplets. Enumeration makes the selection universe explicit. A candidate cannot disappear because it produced an inconvenient result, and a preferred story cannot introduce an unregistered combination after analysis.
Within each triplet, a concentration matrix samples low, medium, high and control levels across the three axes. That creates sixty-four concentration cells before replication and controls. Bliss independence and highest-single-agent comparisons can describe different forms of interaction, while confidence intervals and replicate agreement determine whether an apparent difference is stable enough to interpret.
Selection must be deterministic
A deterministic selector states its ordering before seeing the result. It can require assay quality, component qualification, minimum combination effect, bounded antagonism and safety counterscreens, then break remaining ties with an explicit stable rule. The same input dataset must always produce the same winner and the same replay hash.
Determinism does not turn synthetic or simulated observations into physical measurements. It guarantees that the analysis is reproducible once authenticated observations exist. That distinction is central: software closes ambiguity in data handling, while laboratories close uncertainty about material behavior. Both are required for a complete combination program.
The integrated product gate
After selection, the exact combination becomes a new test article. Interaction pharmacokinetics, impurity profiles, formulation stability, dose ordering and integrated safety may differ from the components considered separately. Human evaluation must use the same defined combination rather than an approximate blend or a substitute from the same chemical family.
The evidence compiler keeps these obligations connected. The combination ID binds its three component identities, source records, assay packet, statistical analysis and public communication state. A public page can explain the lattice today, while any future product-specific statement must point to the later evidence object that actually supports it.