Download the data

Taxonomy v1.0.0 · released 2026-08-06

The taxonomy is published as data, not only as a website. Every profile is machine-extracted from the manuscript and cross-checked against the authors' analysis notebook.

Pin the version you cite. Profile values are the dataset's contract: a change to any of them moves the major version, so 1.0.0 will always mean the same 49 profiles. The site is versioned separately — a redesign never moves the data version. Prose revisions move the minor version only. See the changelog.

taxonomy.json

49 records with full six-dimensional profiles, definitions, and provenance.

Download JSON

taxonomy.csv

The same table flattened for spreadsheets; set-valued cells joined with |.

Download CSV

dimensions.json

All six vocabularies with definitions and meta-groups.

Download

derived.json

Gower distances, nearest profiles, separators, silhouette.

Download

How it was verified

All 294 profile cells compared between the manuscript table and the authors' notebook; 294 agree.
Every dimension value is checked against its published vocabulary.
Nothing is re-typed by hand.

Adjudicated differences

fsl / D3 — resolved in favour of the manuscript. Appendix C states FSL relies on either 1-5 examples for metric-based support sets or tens of examples for gradient-based updates, so both values belong. The notebook was corrected 2026-08-06 and now reproduces the manuscript's reported raw-Gower silhouette of +0.0173.

Using the identifiers

The slug is the stable identifier — use it for cross-references and URLs. Row numbers appear in the paper but depend on how the document is built, so they are display metadata only.

Every dimension value is an array, including single-valued ones. Set-valued cells are the normal case.

Citing the dataset

Afdideh, F., Seoane, F., & Abtahi, F. (2026).
A Six-Dimensional Taxonomy of Post-Training Adaptation Techniques
with Applications in AI Governance. arXiv:2608.06246
Taxonomy version 1.0.0 (2026-08-06).

Released under CC BY 4.0.