Dataset walkthrough

A retraction graph you can audit line by line

Retraction is not a boolean that arrives from one authority. Crossref records an update relation, Retraction Watch records a notice, OpenAlex carries an index flag — and they disagree. This dataset keeps every assertion as its own row, with the source that made it, so a downstream judgment can show its work instead of asserting a conclusion. Published under CC0-1.0, snapshot ri-20261025-1526.

98,868 papers144,814 citation edges572 status rows3 source tiers

Schema

Table	Rows	Fields	Why it is shaped this way
papers	98,868	paper_id · doi · title · publication_year · openalex_is_retracted · cited_by_count	One row per work. paper_id is the OpenAlex ID and stays stable across snapshots, which is what makes incremental builds possible.
cites	144,814	citing_id · cited_id · cited_doi · source_tier · source_uri	Directed edges: citing_id cites cited_id. cited_doi is null when the reference has no DOI, so do not join on DOI alone.
retraction_status	572	paper_id · doi · source · source_tier · is_retracted · retraction_date	One row per (paper, source) assertion. A paper can have three rows that disagree; the judgment layer reports the disagreement instead of picking a winner silently.

Source tiers: A = a registry that records retractions directly (Crossref update-to, Retraction Watch); C = index-derived flag (OpenAlex). Retraction Watch reason and nature free text is excluded from the release, so no row can be misread as a misconduct finding.

The graph, rendered

12 retracted papers, 781 distinct works that cite at least one of them, 788 edges. 6 papers cite two or more retracted works — the yellow nodes — which is exactly the population where “is my claim downstream of a retraction?” stops being a lookup and becomes a graph question.

● big ring = retracted paper (seed)● small dot = paper that cites a seed● yellow = cites more than one seedclick a seed to isolate its citers

Rendered client-side from a 115 KB nodes/edges payload built from the published JSONL, no external chart library. Hover a node for its title; click a seed to isolate the works that cite it.

The twelve seeds

Retracted paper	Year	DOI	Recorded retraction	Citing works
RETRACTED ARTICLE: Pluripotency of mesenchymal stem cells derived from adult marrow	2002	10.1038/nature00870	2024-06-17	122
RETRACTED ARTICLE: A specific amyloid-β protein assembly in the brain impairs memory	2006	10.1038/nature04533	in record, no date	62
RETRACTED: Primary Prevention of Cardiovascular Disease with a Mediterranean Diet	2013	10.1056/nejmoa1200303	in record, no date	62
RETRACTED: Ileal-lymphoid-nodular hyperplasia, non-specific colitis, and pervasive developmental disorder in children	1998	10.1016/s0140-6736(97)11096-0	in record, no date	61
Visfatin: A Protein Secreted by Visceral Fat That Mimics the Effects of Insulin	2004	10.1126/science.1097243	in record, no date	61
Retracted: Predictive Validity of a Medication Adherence Measure in an Outpatient Setting	2008	10.1111/j.1751-7176.2008.07572.x	in record, no date	60
RETRACTED: MicroRNA signatures of tumor-derived exosomes as diagnostic biomarkers of ovarian cancer	2008	10.1016/j.ygyno.2008.04.033	in record, no date	60
RETRACTED: Recent progress in processing and properties of ZnO	2004	10.1016/j.pmatsci.2004.04.001	in record, no date	60
Retracted: An enhanced transient expression system in plants based on suppression of gene silencing by the p19 protein of tomato bushy stunt virus	2003	10.1046/j.1365-313x.2003.01676.x	in record, no date	60
A Comprehensive Review on Metabolic Syndrome	2014	10.1155/2014/943162	in record, no date	60
Hydroxychloroquine and azithromycin as a treatment of COVID-19: results of an open-label non-randomized clinical trial	2020	10.1016/j.ijantimicag.2020.105949	in record, no date	60
RETRACTED: 6-month consequences of COVID-19 in patients discharged from hospital: a cohort study	2021	10.1016/s0140-6736(20)32656-8	in record, no date	60

Check a DOI yourself

Free · no key · no account

Check any DOI for retraction

One HTTP GET against the live snapshot. The response lists every source that covers the paper, its source tier (A = registry-grade, C = index-derived) and the retraction date when a source records one. Nothing is generated by a model, so the same DOI returns the same answer.

What the live lookup returned for 20 DOIs

Queried 2026-10-28T06:19Z. Median round trip 1595 ms, slowest 5318 ms. Retracted cases below; the edge cases that matter are underneath.

DOI	Year	Sources	Retraction date	ms
10.1001/jama.293.14.1727	2005	A+C	2018-09-19	2423
10.1111/j.1365-2702.2011.03951.x	2012	A+C	2016-06-01	1447
10.1002/14651858.cd003041.pub2	2003	A+C	2012-02-15	1526
10.4049/jimmunol.176.3.1534	2006	A+C	2015-06-15	1415
10.1111/j.1442-2042.2009.02429.x	2009	A+C	2010-10-01	3484
10.1182/blood-2008-01-133710	2008	A+C	2012-02-16	3321
10.1182/blood.v98.9.2615	2001	A+C	2009-03-05	943
10.1083/jcb.200501040	2005	A+C	2007-05-07	1509

The three answers people misread

not retracted 10.1038/nature14581 — Glypican-1 identifies cancer exosomes and detects early pancreatic cancer (2015)
no status rows 10.1126/science.aac4716 — Estimating the reproducibility of psychological science (2015)
no status rows 10.1056/NEJMoa2034577 — Safety and Efficacy of the BNT162b2 mRNA Covid-19 Vaccine (2020)
abstained 10.9999/fake-not-a-real-doi-42 — (n/a)

Full benchmark, method and limitations →

Get the data

Hugging Face: ashesborn6/integrity-citation-retraction-graph — three .jsonl.gz files plus a manifest with row counts and SHA-256 per file.
graphrag-ts — the engine that builds the snapshot and answers from it.
Hosted judgment: POST https://pay.ashesborn.cloud/api/v1/judge ($0.02 propagation, $0.10 evidence audit via x402).