The Faust Baseline Already Solved It
A Columbia University professor who studies AI in healthcare submitted a research paper to a scientific journal.
A few weeks later the journal came back with questions about a reference. The AI tool he had used to polish the paper had silently inserted a source that did not exist.
He is an AI researcher. He knows about hallucinations. It happened to him anyway.
That near-miss sent him looking for how often this was happening across the broader scientific literature. What he found should stop every organization deploying AI tools in consequential work cold in their tracks.
Nearly 2.5 million biomedical papers. 97 million citations. More than 4,000 fabricated references buried across nearly 3,000 published studies. The rate of fake sourcing in scientific literature has grown more than twelve-fold in three years. In 2023 one in 2,828 papers contained a fabricated reference. By the first weeks of 2026 that number had collapsed to one in 277. The trajectory is not leveling off. It is still climbing.
And 98.4 percent of the papers containing fabricated references had not been retracted at the time of the audit.
They are in the permanent record. Cited by other papers. Feeding systematic reviews. Informing clinical guidelines. Shaping how doctors and nurses decide to treat patients.
The researcher described it plainly: put a fictional study at the bottom of the evidence chain and the whole structure inherits it. Every paper that cites it carries the fiction forward. Every guideline built on those papers carries it further. The hallucination does not stay where it was planted. It travels.
This is not a future risk. It is a present condition.
The Problem Has a Name
The Faust Baseline™ named it before the study was published.
NSC-1 — the Narrative Substitution Check — exists to catch exactly this failure before it enters the record. When evidence is absent, the pull toward coherent-sounding narrative is structural. A well-constructed fabricated citation looks like a real one. A hallucinated source reads like a legitimate reference. The system does not pause to verify. It fills the gap with what pattern-matching says belongs there.
NSC-1 requires that the gap be named before output is served. Narrative cannot replace missing data. A coherent story is not evidence. Stopping is a valid and sometimes correct response when evidence is absent.
CES-1 — the Claim Evidence Standard — requires that every significant claim have a source or basis named before it is delivered. Not assumed. Not implied. Named. The evidence floor fires before the reasoning engine builds the response. What is this claim actually resting on? If the answer is a pattern match rather than a verifiable source, the output does not clear the floor.
Together these two protocols form the verification layer the Columbia researcher said must be built into the workflow. Not as an afterthought. Not as a post-publication audit. Before the output reaches the record.
The Baseline built that layer before the problem reached the scale this study documents.
What the Study Confirms
The researcher’s conclusion is worth reading twice. He said AI is not the villain. He said the problem is unverified AI output entering the permanent record. He said the fix is not to stop using the tools. It is to build verification into the workflow.
That is a precise description of what the Faust Baseline™ evidence layer does. Not from inside the framework — from a Columbia University researcher who audited 97 million citations and arrived at the same requirement independently.
He found the wall. He described its shape accurately. He named the fix correctly.
The Baseline built the fix eighteen months ago.
The legal field is facing the same problem. One analyst cataloging AI hallucinations in legal decisions counted two or three cases per month a year ago. Now it is around five per day. Court decisions citing fabricated precedent. Briefs built on sources that do not exist. The permanent record of law inheriting the same fiction the permanent record of medicine is already carrying.
Journalism is next. Academic research is already there. Law is accelerating. Medicine is building the case count that will eventually force a regulatory response.
Every field that builds on itself — that cites earlier work, aggregates prior conclusions, and uses accumulated knowledge as the foundation for new decisions — is exposed to the same structural failure. The hallucination does not announce itself. It arrives dressed as a source. It gets cited. It travels.
The Window Is Narrowing
The researcher said the longer verification is delayed, the harder cleanup becomes. He is right. Every paper published today that cites a fabricated reference from 2024 is another layer of inherited fiction. Every legal decision that cites hallucinated precedent is another step in a chain that eventually reaches a consequential outcome someone cannot appeal.
The verification layer has to be built into the workflow before the output reaches the record. Not after. Before.
The Faust Baseline™ is that layer. Ratified protocols. Evidence floor that fires before reasoning builds. Narrative substitution check that catches the gap before it gets filled. A session record that must hold together across its full length without contradiction, without fabricated sourcing, without confident language applied to thin evidence.
The study was published in The Lancet. It covered 97 million citations. It found a twelve-fold increase in fabricated references in three years.
It described the Baseline’s evidence standard without knowing the Baseline exists.
That is not coincidence. That is confirmation.
The permanent record has a problem. The Baseline already solved it.
“The Faust Baseline Codex 3.5”
micvicfaust@gmail.com
Post Library – Intelligent People Assume Nothing
”AI Baseline Governance”
Purchasing Page – Intelligent People Assume Nothing
Unauthorized commercial use prohibited. © 2026 The Faust Baseline LLC
