Verification infrastructure for science

Proof,
not promises.

Fabrication got cheap. Fluent fake papers, hallucinated citations, AI-written reviews — the old signals of trust are broken. PaperBadger verifies the evidence behind a paper and issues a record anyone can re-check.

On the list — we'll reach out before the next conference cycle. ✓

See how it works
Verification recordvr_8f3a2c
Citations resolved0/41
Citations support their claim41/41
Claims grounded in evidence0/58
Numbers match tables & figuresok
No hidden-prompt tamperingclear
Verifiedre-checkable · signed
The fire

The trust layer of science is failing.

Fabrication is cheap now — and it lands on publishers, program chairs, and the researchers doing honest work.

$35–40M
lost by a single publisher to paper mills in one year — journals now risk delisting and losing their impact factor.
21%
of a top ML conference's 76,000 peer reviews were AI-generated.
~98%
success rate of hidden prompts planted in papers to flip AI reviewers positive.
Your name
is on every citation. One unchecked reference can become a correction — or a retraction that follows you for years.
Your edge
submit with proof your evidence holds, and clear the avoidable gaps that get good papers desk-rejected.
The principle

We don't detect AI. We verify the evidence.

Detection is a dead end

The wrong question.

"Was this written by AI?" can't be answered reliably — as models improve, no detector beats chance. And real work written with AI assistance still counts. Detection is a losing arms race.

Verification is checkable

The right one.

Are the claims grounded in real, correctly-cited, consistent evidence? A fabrication fails regardless of who wrote it; a genuine paper passes regardless of tooling — and you can re-check the answer yourself.

What PaperBadger checks

Four checks. One record.

Citations

Every reference exists, isn't retracted, and actually supports the claim it's attached to — not just that it's real.

Claims

Every assertion is linked to in-paper evidence or a real source — anything ungrounded gets flagged.

Numbers

Reported results are reconciled against the paper's own tables and figures. "+12%" had better match the data.

Tampering

Hidden, invisible, and prompt-injection text — the attacks now used to game AI reviewers — caught and surfaced.

Every result ships with the evidence behind it, in a portable verification record a third party can re-run.

Two ways to use it

Before you submit. And at the door.

Author Verify

Before you submit.

Verify your manuscript, fix the things that get papers desk-rejected or retracted, and attach a verification record that proves your work is grounded.

For researchers & labs
The Gate

At the door.

Screen every incoming paper and review for fabrication, hallucinated citations, and gamed AI reviews — before they cost you your indexing.

For conferences & publishers
The edge

Built to be checked.

01

Verifiable by design

Our output is checkable evidence, not a "trust us" claim — the test that privacy promises and AI detectors both fail.

02

A graph that compounds

Every paper sharpens a one-of-a-kind claim–evidence–citation map. It gets more valuable, and harder to copy, with volume.

03

The standard at the gate

Like the plagiarism check before it, a verification record becomes a record authors want and venues require.

Write with AI. Publish with proof.

Join the early access list. We're onboarding CS/ML researchers and venue partners around the next conference cycle.

On the list — we'll reach out before the next conference cycle. ✓