India’s only antivenom is raised on venom pooled from four species collected almost entirely from one facility in Tamil Nadu. Venom composition varies across the country. That is covariate shift, and covariate shift is measurable. This model treats neutralisation as a limiting-reagent problem over a finite antibody budget — not as geometric similarity — and every number here is computed in your browser from the published proteomes.
Click a district for its per-family breakdown, provenance and flags.
Tick venoms in and out. A vial holds a fixed mass of antibody, so antibody raised against one more venom is antibody not raised against the toxins a given bite delivers. Past the optimum, coverage falls.
Bars where adding a venom lowered coverage are shown in the deficit ramp.
Four published antivenom-failure findings were held out. The calibration code raises an exception if a held-out row reaches a parameter fit, and tests assert it raises. A holdout study’s venom compositions are legitimate model inputs; its antivenom results are not.
| Criterion | Pre-registered statement | Computed | Threshold | Outcome |
|---|
Neutralisation is a limiting-reagent problem. A vial holds a fixed antibody mass B, partitioned across toxin families in proportion to the immunising mixture. Nothing below was fitted to produce the dilution effect — it follows from the supply equation.
Non-monotone in |S| — adding a venom moves supply away from the families a bite actually delivers. Dose-dependent — demand is linear in D, supply is not. Saturating — excess antibody against a family is wasted.
One parameter. Everything else is fixed a priori or unidentifiable.
Fixed before any data was ingested, with a written rationale per family. Never fitted.
| Population | Dominant | Source |
|---|
It predicts a risk ordering of Indian districts and venom populations by antivenom deficit; which toxin family drives each deficit; the shape of the coverage-versus-mixture-size curve; and a ranked answer to where new venom collection sites should go.
It does not predict clinical neutralisation percentages, ED50 values, vial requirements, or patient outcomes. It is not validated against clinical data and must not be used to guide treatment.