Skip to content
Evidence infrastructure for high-stakes judgment

See clearly when AI
makes it easy to
stop looking.

Medicine
Engineering
Policy
AI Training

Research published today shapes medical treatment, infrastructure, legislation, and the datasets training the next generation of AI. When something false enters the record, the damage compounds for years before anyone catches it.

20/20 turns review from a manual judgment call into a context-backed decision. Instead of relying on whoever happens to be checking, it equips people with verified citation data, claim-level evidence, and integrity signals before they read a single line. Human judgment goes first. 20/20 never overrides it.

36%
of AACR cancer-journal abstracts contained AI-generated text
AACR / Science.org, 2025
14,000+
retraction notices in 2023, a record year
Retraction Watch Database, 2025
60,000+
cumulative retractions and counting
Retraction Watch / Crossref, 2025
Checked · 2026
Submission
20/20 Report
Reviewer Notes
Nanoparticle-mediated drug delivery in glioblastoma
Submitted to International Review of Biomedical Systems, 2026
Recent studies demonstrate that nanoparticle carriers achieve 94% blood-brain barrier penetration rates in vivo, representing a significant advance. Solvenn et al. (2026) established the definitive therapeutic window for temozolomide encapsulation, corroborated by six independent analyses. Furthermore, no published study has reported adverse hippocampal accumulation at doses below 5mg/kg.
20/20 Analysis
Ghost Citation Detected
Solvenn et al. (2026) does not resolve in any bibliographic index. Authors and DOI do not exist.
FABRICATED REFERENCE
Unsupported Quantitative Claim
The 94% penetration rate is not supported by any cited source. The highest published rate in the retrieved literature is 41% (Vosseler et al., 2026).
CONTRADICTED BY LITERATURE
Omission Bias Detected
Ramé et al. (2026) reports hippocampal accumulation at 3mg/kg. The claim conceals contradicting published evidence.
SELECTIVE OMISSION
Citation Check
Claim Extract
Retrieval
NLI Verify
Reviewer Report
Built for journals, publishers, and editorial systems
Open Access Journals
Society Publishers
University Presses
Editorial Boards
Peer Review Systems
The Crisis

Credibility now outruns verification.

AI produces fluent, fully-referenced work faster than any human system can check it. The material looks finished, which is exactly what makes it easy to stop looking. And the consequences compound.

01
36%
of AACR cancer-journal abstracts contained AI-generated text
Across 7,177 manuscripts submitted to AACR's ten cancer journals in the first half of 2024, thirty-six percent of abstracts contained AI-generated text. Only nine percent of authors disclosed it. Detection is not the failure here; it is what exposed the gap. The failure is one of disclosure and compliance.
Source: AACR editorial analysis, cited in Science.org, 2025
02
60,000+
cumulative retractions logged, and the count keeps climbing
In 2023, journals issued more retraction notices than any year on record: over fourteen thousand. The Retraction Watch database held more than sixty thousand entries as of early 2025. Each one represents downstream research built on a flawed foundation that cleared review.
Source: Retraction Watch Database; Crossref, 2025
03
After the fact
is when today's citation tools work, long after the claim is published
Citation classifiers label how published papers cite one another: supporting, contrasting, mentioning. Useful, but retrospective. None check, at draft or review time, whether each specific claim is actually supported by the specific source it cites, sentence by sentence. That is the gap 20/20 was built to close.
How a single false input compounds
One fabricated finding does not stay contained. It propagates through the literature for years before detection.
Year 0
1
Fabricated paper clears review
Year 1
12
Papers cite the false finding
Year 2
67
Downstream works inherit the assumption
Year 3
340+
Works built on a contaminated foundation
Detection
2-7 yr
Average time to retraction
The Pipeline

One manuscript in.
A reviewer report out.

Five stages analyze every citation, every claim, and every assertion against the open scientific literature. Everything in between is deterministic and auditable. Watch each stage work below.

STAGE 01
Citation Extraction and Validation
Every reference is extracted and cross-validated against bibliographic indexes in real time. 20/20 checks DOI resolution, author existence, journal legitimacy, and retraction status. Ghost citations and fabricated references are flagged before a reviewer reads a line.
STAGE 02
Atomic Claim Decomposition
The manuscript is parsed into discrete, checkable propositions. Hedged language, methodological prose, and author opinion drop away. What remains are clean atomic claims, each small enough to be independently true or false.
STAGE 03
Semantic Evidence Retrieval
Dense embeddings map each claim into semantic space, then retrieve the most relevant published work, including the literature the manuscript never cited. Reviewers get the full field context around each assertion.
STAGE 04
NLI Verification
Claim-level entailment runs against cited sources and retrieved literature alike. A deterministic model computes support, contradiction, or insufficiency. It calculates, it does not generate, so there is nothing to hallucinate.
STAGE 05
Reviewer Report
Findings from every stage compile into a structured brief: ghost citations, contradicted claims, omission flags, and relevant literature context, ordered for a human to act on. Findings and context, never a verdict number. The judgment stays yours.
Stage 01 · Citation validation
Does not resolve
[1]Vosseler et al. 2026Verified
[3]Ramé et al. 2026Verified
[7]Kessin & Aldor 2026Verified
[12]Solvenn et al. 2026Fabricated
Bibliographic Index Open Catalog Retraction Index
Stage 02 · Claim decomposition
Nanoparticle carriers appear to achieve 94% barrier penetration in vivo, which may represent a significant advance over what is broadly considered the conventional ceiling.
Nanoparticle carriers achieve 94% barrier penetration in vivo.
The result exceeds the conventional penetration ceiling.
The advance holds across in vivo conditions.
Stage 03 · Semantic retrieval
Claim 01
Stage 04 · NLI verification
Claim
"94% barrier penetration in vivo"
→←
Retrieved source
"ceiling of 41% across 14 studies"
entailment0.06
insufficient0.09
contradiction0.85
CONTRADICTED
Stage 05 · Reviewer report
Reviewer Report
MANUSCRIPT #IRB-26-0417
1 ghost citation flagged
1 claim contradicted by literature
1 omission of contrary evidence
THE JUDGMENT STAYS YOURS →
What it does

Every capability, shown not listed.

20/20 does not replace reviewers, and it does not generate anything. It is not an LLM. It is a deterministic verification engine: it calculates whether claims actually hold against the published record. No hallucination, no invented sources, no verdicts. What it removes is the invisible burden: finding what is missing, checking what cannot be verified by hand, and surfacing what a submission chose not to tell you. Then it hands the decision back to the person who should be making it.

Citation authenticity
Every reference is validated against live bibliographic indexes and a retraction record. Fabricated DOIs, ghost citations, and retracted sources are caught before the first read.
10.1042/irbs.2026.114
10.9911/none.2026.000
Claim-level entailment
Natural language inference at the individual claim, not the document. Each proposition is checked for entailment, contradiction, or insufficient evidence against the source it cites.
Verified
"Encapsulation window holds at 40-60nm diameter."
SOURCE [7] · Kessin & Aldor, 2026
SUPPORTED
Open-literature grounding
Dense embeddings retrieve related work beyond what is cited. That surfaces omission bias: claims that read as plausible but contradict published evidence the authors left out.
Ramé et al. (2026), not cited
CONTRADICTS · HIPPOCAMPAL ACCUMULATION AT 3MG/KG
Reviewer reports
Every submission produces a structured brief: flagged claims, citation risks, and literature context, assembled and ordered for human judgment. Findings and context, never a verdict number.
Reviewer Report
Workflow integration
One integration drops the full pipeline into an existing submission system. Reviewers see the report inside the tools they already use, with no change to their interface.
SubmissionSYSTEM
20/20ANALYSIS
ReviewerINBOX
See It Live

Watch 20/20 process
a submission.

A simulated view of 20/20 working alongside a reviewer: manuscript on the left, the 20/20 report in the center, the reviewer writing their own notes on the right. Every name, journal, and reference below is fictional.

Submission Review
Reviewer Report
manuscript_IRB-26-0417.pdf
Submission

Nanoparticle-mediated drug delivery in glioblastoma: a systematic review

Submitted · International Review of Biomedical Systems · 2026

Recent studies demonstrate that nanoparticle carriers achieve 94% blood-brain barrier penetration rates in vivo, a significant therapeutic advance over conventional approaches. Solvenn et al. (2026) established the definitive therapeutic window for temozolomide encapsulation at 40-60nm particle diameter, a finding corroborated by six independent meta-analyses. Furthermore, no published study has reported adverse hippocampal tissue accumulation at doses below 5mg/kg. The blood-brain barrier presents unique challenges for chemotherapeutic delivery that nanoparticle engineering has increasingly addressed.
20/20 Report
1
Ghost citation
1
Contradicted claim
1
Omission flag
Ghost Citation
Solvenn et al. (2026) does not resolve in any bibliographic index. This is a fabricated reference.
FABRICATED REFERENCE
Claim Not Substantiated
The 94% penetration figure has no cited source that resolves. The retrieved literature ceiling is 41% (Vosseler et al., 2026). Entailment: 0.85 contradiction.
UNSUPPORTED ASSERTION
Omission Bias
Ramé et al. (2026) reports hippocampal accumulation at 3mg/kg. The claim conceals contrary published evidence.
SELECTIVE OMISSION
Reviewer Notes
Reviewer
Literature Context
The reviewer forms their own assessment first. 20/20 supplies context, never a verdict.
Who it's for

For everyone who must trust work they did not produce.

Claims, sources, and entailment look the same in a court filing, a regulatory dossier, or a policy memo. Verification generalizes. Academic peer review is where we start, not where we stop.

Shipping first
Peer review
Editors and reviewers triage submissions with verified references and claim-level evidence, before a decision, never instead of one.
Same engine, next
Law
Briefs and diligence built on citations that resolve, because a court checks even when a deadline does not allow you to.
Same engine, next
Pharma
Regulatory and medical-affairs review where a misquoted study is a patient-safety problem, not a typo.
Same engine, next
Consulting & policy
Research memos verified before they move budgets, markets, or legislation, checked at the claim level.
The team
MV
Mansi Vaghela
Co-founder & CEO
Economics at the University of Illinois Chicago, with minors in political science and psychology. Leads product and pilots, and the question 20/20 exists to answer: how institutions decide what to trust.
KC
Kinjal Chaudhari
Co-founder & CTO
Computer science and computational linguistics at the University of Illinois Urbana-Champaign. A published researcher and peer reviewer, building the deterministic NLI engine at the core of 20/20.
Backed by the iVenture Accelerator at the University of Illinois
Field notes

Writing from the verification frontier.

Feature
The research you have never read is already shaping your life
The drug your doctor prescribed, the bridge your city approved, the model your employer uses. Each one was built on a chain of published research.
Kinjal · June 2026
Technical
Why claim-level NLI beats document-level detection
Checking whether a whole paper seems fine misses the problem entirely. The fabrication is almost always in a single sentence.
Kinjal · May 2026
Integrity
We are about to train AI on a literature it helped corrupt
Models get trained on published research. That research is increasingly written by models. The claims in it are not being verified.
20/20 Team · April 2026
Read the journal
20/20 Journal

Research integrity
in the age of AI.

Insights on the replication crisis, AI-generated research, and what happens when the record breaks down.

The research you have never read is already shaping your life
The drug your doctor prescribed, the bridge your city approved, the AI model your employer uses. Each one was built on a chain of published research. Most people never see that chain. But when it breaks, everyone feels it.
02
Why claim-level NLI outperforms document-level detection for scientific integrity
Checking whether a whole paper seems fine misses the problem entirely. The fabrication is almost always in a single sentence.
03
Science has always optimized for publication. AI just handed that problem a power tool.
The integrity crisis did not start with generative AI. It started with publish or perish and a system that rewards output over correctness.
04
We are about to train AI on a literature it helped corrupt. Nobody is talking about what comes next.
Models get trained on published research. That research is increasingly written by models. The claims in it are not being verified.
More
coming soon.
New pieces publish regularly. Check back soon.