Identification Engine

A, C, G, T and IUPAC codes. Alignment gaps are accepted: a single gap is read as N, and a run of two or more is removed. Up to 100 sequences per submission, or 1,000 with an access key.

Analyst: a transformer model that, at each rank, weighs about 80 measures for each candidate taxon, from how close and how many the hits are to who identified the references and how, how independent they are, and where they come from. Rather than applying one fixed formula, it fits its own regression to each classification automatically. Weighted Consensus v2: at each rank, a score based on the proximity of the hits and the abundance of hits for each taxon, turned into a probability by a logistic function. The Analyst is the default; use Weighted Consensus v2 when submitting a large number of sequences.

Interim names are sometimes informative and sometimes not, so hits carrying them can be left out. They are recognised by patterns that break taxonomic naming rules: open-nomenclature words and punctuation (“Taygetis sp. 3”, “Agrotera cf. ignepicta”), codes or mismatched capitals within a name (“Heteroastegane flavaDS01MDG”, “uraMalaise6330”), and qualifiers or underscores in higher names (“unclassified Braconidae”, “Ovalentaria_incertae_sedis”). Every rank is checked, not only species. A hit with an interim name anywhere in its lineage is excluded entirely: it is not listed, it is not given to either model, and it is not used in any calculation.