A Worked Example Illustrating How Background Choice and Gene-Family Correlation Can Erase Apparent Reverse Vaccinology Enrichment: Barbervax in Haemonchus contortus

Purpose: RV enrichment claims often test whether a shortlist recovers a reference antigen set more than chance, using whole-proteome background and assuming independence among its members. This cautionary worked example uses Barbervax, the licensed Haemonchus contortus hidden-antigen vaccine, to examine both defaults, since its components are known membrane proteases including a five-member paralog family. Methods: Eight Barbervax components (Set A) were mapped to the H. contortus proteome; recovery in a blind filtered pool was decomposed by filtering stage, and within-family correlation was calibrated with a beta-binomial model across 40 comparison families. Results: A naive hypergeometric test against the full proteome appears significant (p=9.98×10⁻⁴), largely reflecting Set A's known biology rather than a Barbervax-specific effect. An illustrative test of pool survival is compatible with three independent backgrounds (56–70%), though at n=8 it cannot rule out a true advantage (minimum detectable effect about 98%). Treating the H11 paralogs as fully correlated – the appropriate bound given their over 90% identity – gives corrected recovery probabilities of 0.33–0.54, still non-significant. A third apparent signal (p=0.0012, against Set A's own homology family) traced to a documented tool bias against cysteine proteases, not a genuine effect. Conclusion: This self-corrected case does not show blind RV recovers Barbervax beyond its known biology, but its small sample size limits that claim. It illustrates specific ways naive enrichment claims mislead – tautological recall, background choice, unmodeled correlation, and composition-driven false positives – offered as a diagnostic checklist for one worked example, not a validated general result.