Research and Evaluation

Manuscripts, benchmark datasets, evaluation drivers, and historical results are maintained separately from the application checkout. The complete pre-separation source snapshot is preserved at commit ebfa635ca786796287c9ff2a3caba1c2618f016e on the research-archive-20261029 branch.

• Publication reproduction guide

• Manuscript and figures

• Evaluation scripts and evidence

OrthoBench and Quest for Orthologs are the primary benchmarks; Three Kingdoms provides supplementary conserved-family evidence. These datasets were inspected during development, so they do not independently validate settings selected using their scores. Orthogroups, root hierarchical orthogroups, and native pairwise ortholog predictions have different semantics and must not be treated as interchangeable outputs.

Shared-host timing measurements may be distorted by competing analyses in method-dependent ways. Matching CPU and memory limits does not establish isolated speedups. Historical reproduction instructions and limitations remain in the archived evidence; separating the application does not establish new biological accuracy, timing, or reproducibility claims.