Data downloads
Every paper and researcher profile published in this guide, with its annotations, as CSV for spreadsheets and JSON for code. Rebuilt every day. Free to use, no account or key.
| File | Contents | Records | Size |
|---|---|---|---|
| papers.csv | Annotated papers: one row per paper, for spreadsheets | — | — |
| papers.json | Annotated papers: full records with explainers | — | — |
| authors.csv | Researchers: one row per researcher, a position per claim | — | — |
| authors.json | Researchers: profiles with positions, evidence and sources | — | — |
Files are dated with the day they were built.
What is in the files
Papers
id · title · year · date · authors · institutions · venue · theme · doi · source_url (the paper itself) · page_url (its page here) · open_access · cited_by · lab_report (an AI lab's own research, not peer reviewed) · summary (AI-written) · question · approach · findings · limitations · method_type (the AI-written explainer, when there is one) · abstract
Researchers
id · name · role · bio (AI-written, documented facts only) · profile_url · links · sources · for each claim (productivity, jobs, inequality, regulation, pace) a position (agree, lean_agree, mixed, lean_disagree, disagree) and its confidence · positions_coded (n of 5) · updated. The JSON adds each position's rationale and the evidence it cites.
Reusing the data
Summaries, explainers and researcher bios are AI-written, and every record says so
(the ai_written column, or labels in the JSON). Researcher positions
are documented positions with cited evidence, not characterizations; when you count them, give
the denominator (“n of N coded researchers”). Keep the source links when you quote. Only
published records are included; the internal scores used to curate the collection are not.
How the guide is made is on the About page.
To query the same content live, one record at a time, use the API & MCP server.