# Sequences and annotations supporting the reference labels

- `hla_spans/`: 172 HPRC regional compressed FASTAs, each with a receipt containing coordinates, source accession and sequence/file hashes. These contain 579,008,380 bases in 167,144,360 compressed bytes.
- `preparation/annotations/`: all 172 original HPRC source GTFs (2,138,649 bytes), preserving every original feature/attribute. Source annotations used IPD 3.59.0; reference labels in this release were reassigned against pinned IPD 3.65.0.
- `preparation/ANNOTATIONS.json`, `CANDIDATE_TRUTH.json`, `DONOR_MANIFEST.json`: selected cohort metadata, gene coordinates and GTF checksum bindings. Published annotation allele names localize/support inspection; they are not the newly sequence-matched truth labels.
- `extra_regions/arab.fa.gz`, `saudi.fa.gz`: 975 retained Arab/Saudi intervals. Headers retain archived donor/haplotype labels.
- `extra_evidence/`: query JSONs containing interval sequences, gene/donor annotations and hashes; the Saudi public haplotype crosswalk; and the pinned source metadata for IPD feature annotations. No source GTF is claimed for these retained intervals.

Coordinates in GTF are one-based inclusive. Regional receipts also record zero-based half-open bounds. The generator handles strand orientation and validates hashes. Keep inputs in this relative layout for reproduction. Whole-genome reads (about 1.47 TB for the HPRC cohort) and full-assembly indexing sidecars are not distributed here.

[Release guide](../README.md) | [Reference labels](../truth/README.md)