The important distinction is shieldin-family identity versus an intact DNA-end-protection module. Human SHLD2 has experimentally distinct long/short splice products; a short product can still recruit to damage sites and assemble with shieldin while failing to suppress homologous recombination.
The 883-residue horse model retains the insertion absent from human canonical SHLD2, but its distal OB-domain region is divergent/gapped in the long-isoform alignment. Nuclear and damage-site recruitment are plausible; the complete NHEJ/resection-suppression mechanism is UNC for the exact model. Human Protein Atlas XML supports an additional human actin-associated pool, so the actin prediction is unresolved in horse rather than refuted as incompatible with nuclear localization.
The reproducible alignment, source paths and hashes are in the paired comparison. Current UniProt sequences have not been proven identical to the original ProtNLM input sequences. Sequence anomalies are therefore recorded as model/transfer limitations, not as proven wrong-input pipeline errors.
Obtain a transcript-supported horse C terminus and inspect the OB folds. Distinguish long/short human isoforms in any assay transfer. Inspect the HPA antibody/cell-line context before transferring secondary actin localization.
Every seeded annotation receives a current assessment. UNDECIDED marks unresolved source-specific or biological evidence; these are initial reviews rather than a claim that every original experimental assay has been independently reproduced or verified. The human Edison report is retained as a research synthesis and source-finding aid; decisive YAML excerpts cite primary publications, source records or reproducible analysis. The validator advisory to cite the deep-research file is deliberately not satisfied by citing AI prose as biological proof.