VCF Record Types

Genotypes MUST store per-allele indices (0=REF, 1+=ALT, None=missing) and per-separator phasing (true=phased |, false=unphased /). The phase bit on the first allele separator is ignored by convention but MUST be written as unphased.

Alleles MUST be represented via a type-safe enum with variants for common VCF allele patterns: reference-only (REF with no ALT), SNV (single-base REF and ALT), insertion (anchor base + inserted bases), deletion (anchor base + deleted bases), and complex (arbitrary REF/ALT strings for MNVs, symbolic alleles, or mixed multi-allelic sites). Construction MUST validate structural invariants: SNV alts must differ from ref, insertion/deletion sequences must be non-empty.

The reference length (rlen) MUST be derived from the alleles variant: 1 for Reference, SNV, and Insertion; 1 + deleted.len() for Deletion; ref_allele.len() for Complex.

VCF REF/ALT text serialization MUST be handled by the alleles type. Each variant produces the correct VCF encoding: Insertion emits anchor as REF and anchor+inserted as ALT; Deletion emits anchor+deleted as REF and anchor as ALT; Reference emits the ref base with . as ALT.

n_allele counts ALL alleles including REF. A record with REF=A and ALT=T has n_allele=2. A record with no ALT (monomorphic/reference site) has n_allele=1.