Completed DogoHLA typing/retrieval/linkage experiments, release 20261021.

Full-genomic graph: 172 donor-panel arms (86 donors, two panels), each with E1/E2/E3 union and native-proposals-only calls. Original-reference retrieval ablation: 120 arms (60 donors). Linkage: independent/linked/shuffled on 10% and 25% read fractions, each 120 arms (60 donors). All planned native reference controls are complete; their 946 active genotype outputs are in the separate T1K baseline release.

Calls, decisions, sequence calls, completion receipts, frozen plans/code, scores and provenance are included. No raw reads or bulky likelihood evidence are included. Per-file SHA256 values bind this frozen snapshot.

This completes the bounded experiments, not the broader research plan. Raw-read truth certification, untouched confirmation cohort, matched H2 variant and fold-safe H3 structural endpoints remain outstanding. BASELINE_GATE explicitly records comparator completion; competitor native outputs belong to their separate baseline release.
All three runnable comparators are complete on 86 donors: OptiType/SpecHLA two-field and HLA*LA common-G endpoints. HLA*LA numeric projections are not accuracy endpoints. HLA-HD remains blocked by unavailable distribution.

Public-view provenance: private workstation/cluster path prefixes are replaced by portable logical paths. Numerical results and scientific call files are unchanged. Manifest original_sha256 binds original bytes; sha256 binds the downloaded public view. Embedded receipt hashes refer to original pre-redaction artifacts and can be matched against original_sha256.