BF Workbench is a restricted scientific system. Records are available only to registered, authorised accounts.

Reference libraries and builds

Which libraries exist, how they are built and versioned.

Reference

Two identification paths exist on the platform, and each searches a different reference library. The workbench's Sequence Identification job searches a BLAST database built from BOLD Forensics' own records. The public identification service at barcodeid.biolytica.ca searches a library built from a public BOLD release. This page says what each library contains, how it is built and refreshed, and what is and is not recorded about the build a result used. A reviewer needs this to say what a hit list was compared against.

The workbench job's database

Source. Every day bf-data-api exports the platform's single-pane record view, one line per record with its sequence and taxonomy. The BLAST database is built from that export.

What goes in. The builder keeps every record whose sequence has at least 100 bases, for every marker. There is no marker filter, so COI-5P, ITS, rbcL and matK records sit in the same database. Duplicate headers are made unique with a __dupN suffix, and the sequences are indexed with makeblastdb. Each database header carries the record's processid, marker, BIN, taxonomy identifier and public status; the taxonomy identifier is joined at search time to a denormalised taxonomy table exported separately from the taxonomy view.

How it is refreshed. By hand. The export runs daily, but the database and the taxonomy table are rebuilt only when an operator runs the build scripts; no scheduled task does it. Between rebuilds, records added, corrected or removed on the platform are not reflected in the database.

Library choice on the form. The form's library option offers a single entry labelled "COI Full Database (500 bp)". The job does not read it. The search always runs against the one database described above, which holds every marker at 100 bases or more, not COI alone and not 500 bases or more.

Good to know: The database is built from the same records you select in the workbench, and the job has no step that excludes a query from the reference. A record that is already in the database matches itself, typically at 100% identity. Read the top hit with that in mind.

What is not recorded

  • The result package and report do not name the database build a job searched, its build date, or the number of records it held.
  • The database itself carries no version or build provenance.
  • The report does not record the taxonomy table's build either, and the taxonomy of a hit is whatever that table held at run time.

If you need to state what a job was compared against, record the build date yourself at the time the database is rebuilt. See Identification for the search itself.

The public service's library

Source. The library is built from a dated public BOLD release. The current base release is BOLD_Public.25-Sep-2026, which holds 23,684,112 records; of those, 10.24 million COI-5P sequences enter the build.

Build steps.

  1. Records are taken at 100 bases or more.
  2. Sequences are dereplicated: records with an identical marker, sequence and full lineage collapse to one representative, giving 10,240,337 representatives.
  3. Large BINs are downsampled, which reduces the library to about 8.4 million sequences.
  4. A coverage index is written alongside the library, recording what it contains.

Swapping a build in. A new library replaces the old one through a documented swap procedure with verification gates that must pass before the new build serves identifications.

Versioning. A library is identified by its base release date and the date of its build steps, and the coverage index describes the result. The decision model is tied to a build: its coefficients are fitted on a named library, and the build procedure requires a refit after any rebuild.

Good to know: The decision model's confidences are calibrated to the library build it was fitted on, so a rebuild is followed by a refit as part of the build procedure.

Comparing the two

Workbench database Public service library
Source BOLD Forensics' own records, daily export Public BOLD release, dated
Markers All, 100 bases or more COI-5P only, 100 bases or more
Processing Duplicate headers renamed Dereplicated, large BINs downsampled
Refresh Manual rebuild, no schedule Versioned build with a swap procedure and gates
Build recorded in results No Build identity held in the coverage index; not shown by the workbench

The two libraries do not overlap by design, and neither path consults the other's library. A sequence absent from the platform's records can still be identified by the public service, and a record on the platform that is not in public BOLD is matched only by the workbench job.