Merlise achieves breakthrough results on SciFact-Open
Merlise

Case studies

The cases below rest on a filing, a court order, a retraction notice, or an editor's correction that is already public. We ran the original document through Merlise and kept the ledger it produced, sources included, so you can check it against the record yourself.

Legal

Citations checked against the reporter, the opinion, and the docket.

Teardown

Two dozen citations, and the court had already counted them

A federal appeals court catalogued the fake and misused citations in a set of briefs. We ran the same brief and reached the same appendix.

June 2026 · 5 min
Teardown

The case was real. The quote inside it was not.

A Ninth Circuit sanctions order found real cases carrying invented quotations. That failure survives a check that only confirms a citation exists.

February 2026 · 5 min
Failure mode

A source that reads like a real one and is not

Fabricated citations are no longer rare or confined to one firm. A tracked database now counts several new ones a day.

March 2026 · 4 min
Failure mode

The case exists. It just does not say that.

The hardest citation to catch is the one that resolves cleanly and still fails, because the source is real and the claim attached to it is not.

March 2026 · 4 min

Financial

Filings read against themselves and against what came after them.

Teardown

Ten days after filing, a fifth of the revenue disappeared

NeuroOne filed a 10-Q reporting 2.4 million in quarterly revenue. Ten days later, the company said half a million of it was not real.

May 2026 · 4 min
Teardown

One filing undid eighteen months of statements

Richtech Robotics told investors that five separate filings, going back a year and a half, could no longer be relied upon.

June 2026 · 5 min
Failure mode

The sentence said improved. The filing did not agree.

Prose summaries of financial results drift from the filings behind them more often than most readers expect, and the drift is rarely a typo.

May 2026 · 4 min
Failure mode

A number can be true on Monday and wrong by Friday

Public filings change underneath the content written about them. Nothing about the original article knows when that happens unless something checks.

June 2026 · 4 min

Scientific

References checked against the paper they claim to point to.

Editorial

Quotes and contributors checked against a primary record.

Teardown

An article about an AI behaving badly was written by one

Ars Technica retracted a story about an AI coding agent after the quotations it attributed to a named engineer turned out to be invented.

February 2026 · 4 min
Teardown

The byline was invented before the story was

WIRED retracted a freelance article after determining that both the story and the contributor behind it could not be verified.

2026 · 4 min
Failure mode

Quotation marks are a promise. This one had nothing behind it.

A direct quote claims exactness in a way a paraphrase never does, which is exactly why it deserves its own kind of check.

2026 · 4 min
Failure mode

The claim had already been checked. It came back anyway.

A false health claim can be debunked, go quiet, and resurface in slightly different words through a new social cycle, as if nobody had ever looked at it.

2026 · 4 min

Bring one of your own.

We will run a real document from your desk through Merlise and walk you through the ledger, claim by claim.

Request a demo →