Calling a diplotype from consumer DNA — and refusing when we can't

Why the caller sits upstream of every other guard

The lookup that turns a diplotype into a metabolizer phenotype is deterministic and well tested, which is exactly the problem: hand it a mis-called *2/*2 and it will faithfully return a confident "poor metabolizer". No downstream check can catch that, because nothing downstream can see the genotype any more.

So the caller owns its own gate and defaults to producing nothing. Silence is a supported outcome; a plausible guess is not.

Three things that make it return nothing

Strand ambiguity: genotypes are resolved against both strands, and palindromic SNPs — A/T and C/G, where the two strands are indistinguishable — are refused rather than assumed.

Phasing: heterozygous at two loss-of-function SNPs is genuinely ambiguous from unphased consumer data. It could be *2/*3 (poor metabolizer) or *1 carrying both (intermediate). Without phase information those cannot be told apart, so the caller declines instead of picking the likelier one.

Absence of a variant is not evidence of the reference allele. An all-reference genotype yields no phenotype at all, because the alleles that were never tested for — *17, for instance — could still be there. Reporting "*1/*1, normal" from a panel that only looked at two positions would be the most dangerous output the system could produce.

Hardened by trying to break it

The caller went through an adversarial review pass whose only job was to find inputs that produced a confident wrong answer. Two changes came out of it: reported markers now cite only the SNPs actually measured rather than the gene's full panel, and biologically impossible input — homozygous variant at two distinct alleles at once — is refused outright instead of being quietly reduced to something plausible.