SciNet

A task-level database of scientific research — a comprehensive map of what researchers actually do, broken down by domain, field, subfield, and topic. SciNet enables rigorous, task-level analysis of scientific work by mapping the granular activity structure of science across 5 domains, 34 fields, 320 subfields, and 4,516 topics, with 5,044 released task statements (27 universal, 45 domain, and 4,972 subfield-level).

Website: anatomyofscience.com · Repository: github.com/lukasalthoff/scinet

Overview

SciNet organizes research work into a hierarchy aligned with OpenAlex (domains, fields, subfields, and topics). For each level, we use large language models to generate O*NET-style task statements describing what researchers in that area regularly do.

The files in data/ are released for replication and downstream research.

Data files

All files are UTF-8. CSVs use comma separators. See data/README.md for a standalone description.

File	Description
data/tasks.csv	Every task in the hierarchy (universal, domain, and subfield levels) with category labels
data/openalex_topic_subfield_mapping.csv	Maps each OpenAlex topic to its SciNet display domain, field, and subfield

Data dictionary

tasks.csv

Column	Description
task	Task statement text
category	Task category (e.g. "Ideation & Hypothesis Generation", "Data Gathering")
level	One of universal, domain, or subfield
domain	Domain name, e.g. "Social Sciences" (empty for universal tasks)
field	Display field name, e.g. "Economics" (empty for universal/domain tasks)
subfield	Display subfield name, e.g. "Labor Economics" (empty for universal/domain tasks)

openalex_topic_subfield_mapping.csv

Column	Description
topic_id	OpenAlex topic identifier
topic_name	Topic display name
domain	SciNet display domain
field	SciNet display field
subfield	SciNet display subfield

Methodology

1. Hierarchy: OpenAlex domains, fields, subfields, and topics define the taxonomy.
2. Task generation: Large language models produce O*NET-style task statements at field, subfield, and topic levels using a top-down hierarchical approach.

For a complete description of every pipeline step — including prompt design, coverage thresholds, O*NET calibration results, and protocols.io validation — see METHODOLOGY.md.

For visual summaries of the data — task distributions, verifiability rankings, AI adoption — see DATA_OVERVIEW.md.

For the research paper when available and project updates, see the Stanford project page.

Citation

If you use this dataset, please cite the SciNet project and this repository, for example:

@misc{scinet_data,
  title        = {SciNet: The Anatomy of Science},
  author       = {Althoff, Lukas},
  year         = {2026},
  howpublished = {\url{https://github.com/lukasalthoff/scinet}},
}

License

Data and documentation in this repository are licensed under CC BY 4.0 — see LICENSE.

Changelog

2026-07-13

• Refreshed tasks.csv to match the live site: 5,044 task statements across 34 fields and 320 subfields (previously 4,981 / 30 / 315).
• Adds four fields that were introduced after the March release: Public Health & Epidemiology, Nutrition & Dietetics, Veterinary Medicine, and Library & Information Science.
• Corrected the headline counts in README.md and DATA_OVERVIEW.md, which still described the March hierarchy (and quoted a task count that included topic-level statements not present in tasks.csv).
• Realigned the domain column with the site's five display domains. Previously the released file used an older four-domain grouping in which Arts, History, Languages & Linguistics, Literature, Philosophy and Religion were labelled Social Sciences, and Neuroscience was labelled Health Sciences — neither matching what the website shows. All 34 fields now carry the same domain as anatomyofscience.com.

Known gaps (not yet released):

• Arts & Humanities has no domain-level tasks. The 45 domain tasks cover only four domains; the humanities fields currently inherit none. TODO: author them.

• Substeps. Each subfield-level task decomposes into substeps; these are not yet part of the public release. TODO: publish once the decomposition is regenerated.

• Substeps for the four new fields. The substep decomposition predates them, so Public Health & Epidemiology, Nutrition & Dietetics, Veterinary Medicine and Library & Information Science currently have no substeps. TODO: generate.

2026-03-20

• Replaced generated_tasks.csv, openalex_topics.csv, and catalog.json with two simpler files:
  • tasks.csv: flat task file with category, level, domain, field, and subfield columns.
  • openalex_topic_subfield_mapping.csv: maps OpenAlex topics to SciNet display domains, fields, and subfields.
• Added METHODOLOGY.md: full pipeline documentation covering taxonomy construction, hierarchical task generation, O*NET-style rating and filtering, AI exposure scoring, O*NET calibration, and protocols.io validation.