LoginSign Up
← Back to all sparks
T

taxifydb

INFRA · APIS
Velocity5.0

Build Backbones and Enrichments for the 'taxify' Package

github.com/gcol33/taxifydb ↗

taxifydb found that its coverage audit was checking a hand-maintained list, not reality.

taxonomyreference datagbifbuild integrityr packages
◆Current state
taxifydb packages taxonomic backbone databases for R, publishing prebuilt registers alongside the underlying data. The 0.1.21 release is a self-audit: the coverage checker was walking a hand-kept BACKENDS vector, so a backbone wired into the build but never added to that list was audited by nothing and reported clean, which had been the state of one backbone since it was added. The check now parses the real registration source, and refuses to run rather than audit a guessed set.
◆Where it's heading
The visible work is about closing gaps between what the build actually contains and what the checks believe it contains. Alongside the coverage fix, register artifacts are now compared against the backbones recorded in their own metadata sidecar, adding drift states the weekly workflow can open issues on. A curl bug in the same release is the same theme at a different level: setting an Authorization header replaced the header set rather than adding to it, silently dropping the Accept header that requests the asset bytes.
◆Prediction
Expect the drift detection to keep expanding into the parts of the build that currently have no signal, since each release so far has found another place where a check was passing without looking at anything.

◆Recent moves

  1. 7d ago

    Coverage audit rebuilt to read the real backbone set

    The coverage audit stops trusting a hand-maintained list and parses the actual extractor registration, having silently ignored one backbone since it was wired in. Register artifacts are now checked against their own metadata sidecar, adding two new drift states, and a curl header bug that dropped Accept and User-Agent from authenticated requests is fixed.

    View source ↗
  2. 8d ago

    GBIF backbone rebuilt at 6.4 million rows

    A new GBIF backbone build of 6,404,001 rows at 1.5 GB. Routine in form, but these builds are the package's actual deliverable, and the audit work in the following release exists to catch exactly the case where a build like this fails to reach the published register.

    View source ↗