Methodology

Every paper the human review put on the shelf (tracked or noted), independent of triage status, full corpus on every download. headline_claim, the_test, effect_size, limitations, counterpoint, econ_implication, and who_cares come from the model extraction jsonb; all null until the paper has been analyzed. thread_slugs only lists CONFIRMED thread placements. review_note, the reviewer's private why for tracking a paper, never ships in this or any dataset. rigor_prior rides here only because this dataset requires an access key, the same scoped exception the signals-export dataset uses for touch_details; it stays banned from every guest-reachable dataset.

Schema

Column	Type	Definition
id	text	Stable UUID of the paper; the internal join key.
arxiv_id	text	arXiv identifier, when the paper came from arXiv.
url	text	Link to the paper.
title	text	Paper title.
review_status	enum	tracked (actively followed) or noted (worth knowing).
published_on	date	Paper date (YYYY-MM-DD), when known.
reviewed_on	date	When the human review decision was made (YYYY-MM-DD).
rigor_prior	number	The reviewer's 0 to 100 methodological-rigor prior, when set.
citation_count	number	Citation count at last check.
author_hindex	number	Max author h-index at last check, a prior, not a verdict.
headline_claim	longtext	What the paper actually asserts, one falsifiable sentence.
the_test	longtext	What was actually measured, on what data or systems, at what scale.
effect_size	longtext	How big the result is, and where it holds or breaks.
limitations	longtext	What the authors themselves concede, plus obvious unstated ones.
counterpoint	longtext	What a fair skeptic would say.
econ_implication	longtext	What this means for the economics of AI in the next 12 to 24 months, stated with restraint.
who_cares	longtext	JSON array of objects with lens and note: the audience lenses this paper speaks to, and why.
thread_slugs	text	Research thread slugs this paper is confirmed placed in, joined with a semicolon.
advisory_claim_touches	text	Argument-map codes the paper may bear on, joined with a semicolon; advisory only, never evidence.
promoted_signal_id	text	Signal ID when the paper was promoted to the Signal Board.
analyzed_by	text	Model that produced the extraction, when one exists.
abstract	longtext	The paper's abstract.
full_text	longtext	The retained full text, uncapped. Null for a paper never successfully fetched (abstract-only).

Research export, firewall

Request access Unlock with an access key

/api/datasets/research-export · 23 columns · access key required

The firewall handoff of the reviewed arXiv shortlist: every tracked or noted paper with its extracted finding and retained full text.

Methodology

Schema

Column	Type	Definition
id	text	Stable UUID of the paper; the internal join key.
arxiv_id	text	arXiv identifier, when the paper came from arXiv.
url	text	Link to the paper.
title	text	Paper title.
review_status	enum	tracked (actively followed) or noted (worth knowing).
published_on	date	Paper date (YYYY-MM-DD), when known.
reviewed_on	date	When the human review decision was made (YYYY-MM-DD).
rigor_prior	number	The reviewer's 0 to 100 methodological-rigor prior, when set.
citation_count	number	Citation count at last check.
author_hindex	number	Max author h-index at last check, a prior, not a verdict.
headline_claim	longtext	What the paper actually asserts, one falsifiable sentence.
the_test	longtext	What was actually measured, on what data or systems, at what scale.
effect_size	longtext	How big the result is, and where it holds or breaks.
limitations	longtext	What the authors themselves concede, plus obvious unstated ones.
counterpoint	longtext	What a fair skeptic would say.
econ_implication	longtext	What this means for the economics of AI in the next 12 to 24 months, stated with restraint.
who_cares	longtext	JSON array of objects with lens and note: the audience lenses this paper speaks to, and why.
thread_slugs	text	Research thread slugs this paper is confirmed placed in, joined with a semicolon.
advisory_claim_touches	text	Argument-map codes the paper may bear on, joined with a semicolon; advisory only, never evidence.
promoted_signal_id	text	Signal ID when the paper was promoted to the Signal Board.
analyzed_by	text	Model that produced the extraction, when one exists.
abstract	longtext	The paper's abstract.
full_text	longtext	The retained full text, uncapped. Null for a paper never successfully fetched (abstract-only).