Data, API and code

Everything behind the index is open: the ratings of every task, the scores of every occupation and country, the rubric, the prompts and the code that builds them. The archived release is on Zenodo; the files below are the same tables.

Zenodo · doi:10.5281/zenodo.23552918 Hugging Face GitHub

Table	Contents	Files
Occupations	433 ISCO-08 unit groups: ADI with its 95% interval, ADI3, ADIs, the four barrier shares and the number of tasks.	`adi_isco.csv` 45 KB `adi_isco.parquet` 38 KB
Countries	146 countries: the three indices, total employment, income group, ILOSTAT year and the ISCO resolution used.	`adi_pais.csv` 10 KB `adi_pais.parquet` 12 KB
Tasks	All 18,796 task statements: consensus level, each model’s level, share of time, human minutes and the four barriers.	`adi_tarea.csv` 3.5 MB `adi_tarea.parquet` 771 KB
Claims	Every number in the paper, the report and this site, with its identifier and the script that produced it.	`claims.jsonl` 28 KB
Dictionary and checksums	Column definitions for every table, and SHA-256 sums of the release.	`DATA_DICTIONARY.md` 9 KB `SHA256SUMS` 50 KB

The raw ratings of each model (one JSON line per task, with the rubric hash and a timestamp) are in the Zenodo record, under release/v2026.1.

Static JSON API

One file per record, with its citation

No key, no rate limit, CORS open. Paths are versioned, so a link never changes meaning.

Path	Returns
`/adi/v2026.1/index.json`	List of occupations and countries with their headline values.
`/adi/v2026.1/isco/{ISCO-08}.json`	One occupation: the three indices with 95% intervals, barrier shares, rank and a `cite` field.
`/adi/v2026.1/country/{ISO3}.json`	One country: indices, intervals, income group and a `cite` field.

Python

adi-index

pip install "adi-index @ git+https://github.com/Cespial/delegation-divide#subdirectory=herramientas/adi-index"

adi lookup 2512
adi country COL
adi top -n 10
adi cite bibtex

Command line and library. The tables travel inside the package, so it works offline.

MCP server

adi-mcp

pip install "adi-mcp @ git+https://github.com/Cespial/delegation-divide#subdirectory=herramientas/adi-mcp"

claude mcp add adi -- adi-mcp

Four tools for AI assistants: lookup_occupation, country_profile, top_occupations, cite.

How to cite the data

Text

Espinal Maya, Cristian, and Santiago Jiménez Londoño. 2026. ADI v2026.1: Agent Delegability Index data, code and MCP server. Zenodo. https://doi.org/10.5281/zenodo.23552918

BibTeX

@misc{espinal2026adidata,
  title     = {{ADI} v2026.1: Agent Delegability Index data, code and {MCP} server},
  author    = {Espinal Maya, Cristian and Jim{\'e}nez Londo{\~n}o, Santiago},
  year      = {2026},
  publisher = {Zenodo},
  version   = {v2026.1},
  doi       = {10.5281/zenodo.23552918},
  url       = {https://doi.org/10.5281/zenodo.23552918}
}

For language models

/llms.txt is a plain-text map of the index; /llms-full.txt adds definitions, column schema and the tool reference.

Licences: data and text CC BY 4.0; code MIT. O*NET content is used under CC BY 4.0 from the U.S. Department of Labor; ILOSTAT data under CC BY 4.0.