- Research
- Data
On May 7 the pipeline that builds the phenotype dataset was released as open source under Apache 2.0: Agaveis/phenotype-catalog-pipeline. Anyone can read the code and rebuild the dataset.
A methodology paper documents the construction end to end, and the v1.0.0 release that carries it received a Zenodo DOI. Every citation block, on Hugging Face, on GitHub, in the paper and on Zenodo, carries the DOI, the dataset and the repository together.
Five reference pages launched on the site the same day: a glossary, a methodology page, a taxonomy overview, a taxonomy FAQ and an ethics page. Together they explain in plain language how the catalog is built and what it is for.
Under the hood
- The repository holds six stages: notable-people extraction from Wikipedia, image discovery, vision analysis, per-group aggregation, phenotype-profile drafting and the dataset build, plus the analysis prompt with design notes and a schema excerpt for the core tables.
- Exports are deterministic: the same inputs and scripts produce the same CSV and JSONL outputs, timestamp fields aside, and a manifest records the schema version, generation time, row counts and per-config coverage.
- The paper, about 3,700 words, covers construction, schema, coverage, sampling bias, intended uses, ethics and reproducibility. It sets the dataset beside fairness benchmarks that sort faces into four to seven coarse categories, and works instead at the grain of 239 ethnic groups.
- Zenodo archives each GitHub release under its own DOI, and a concept DOI always resolves to the newest version.
- The methodology page runs in eight sections drawn from the paper, the FAQ answers ten questions from scientific validity to mixed heritage and source provenance, and the ethics page sets out intended and out-of-scope uses.
- open pipeline stages
- 6
- words in the methodology paper
- ~3,700
- new reference pages
- 5
See it on the site
Sources and open data
Related build notes
- May 6, 202613,094 notable people and 5,668 photos read by a vision modelGroup pages gain Notable People linked to Wikipedia, and a vision model turns 5,668 reference photos into structured trait observations, each traceable.
- Aug 31, 2026A corpus of 28,756 facial and body observations over 39 peoples2,681 portraits read by a vision model against 22 controlled vocabularies now yield 28,756 structured observations that feed the trait pages.
- May 7, 2026196 ways to describe a human body, each tied to a published scale22 controlled phenotype vocabularies: 196 dimensions, 853 defined terms and 113 references, with every dimension anchored to a published scale.