The Edit History That Saved William Quarterman's Degree

William Quarterman was a senior at UC Davis. He wrote a take-home midterm essay for a history class, submitted it and got back a zero. His professor's note said he'd used ChatGPT to write it.

He hadn't. He'd never used ChatGPT, for this essay or any other. The distinction didn't matter yet, because the accusation had already left the classroom. The professor referred him to UC Davis's Office of Student Support and Judicial Affairs for academic dishonesty, and a take-home midterm turned into a case file.

Where the accusation came from

The professor's reasoning, as Quarterman later described it: the essay read as too general. That feeling sent her looking for confirmation, and she ran it through an AI detection tool. The tool marked it positive. That was the whole chain, one tool confirming one feeling, and it was enough to fail him and start a disciplinary referral.

This is the part worth sitting with. The story is about what a tool's verdict does to a human judgment once it exists; one careless professor is the least of it. "Too general" is a feeling, and feelings are arguable. A detector's verdict looks like evidence, even when the thing generating it has no idea what it's talking about.

A nervous student writing a careful, unadventurous exam answer can read as machine-generated. So does a professional writer working from a tight outline. So, it turns out, does Genesis.

The Bible has a ChatGPT problem

Quarterman's family tested the detector on famous, unambiguously human writing. It flagged Martin Luther King's "I Have a Dream" speech as AI-generated. It flagged excerpts from the Book of Genesis the same way.

Sit with that for a second. A tool built to catch machine-written text couldn't clear the Book of Genesis. If your detector thinks Genesis was ghostwritten by a chatbot, the problem is your detector.

Rhythmic, unsurprising prose reads the same to a pattern-matcher whether it came from a machine, a scripture or a stressed student trying to hit a word count before a deadline. The detector doesn't know who wrote the sentence. It knows the sentence didn't surprise it.

The receipts

Quarterman had written the essay in Google Docs. Google Docs quietly logs a full edit history of every document. He pulled his up and it showed exactly what he said happened: he'd written the essay himself, over two or three hours, the ordinary way.

A real essay looks like Quarterman's: uneven and alive.

He had help fighting the case. His father is a lawyer. His mother has a legal background too. A sister who works in tech advised him on the technical side of the argument. Between the three of them, the family built the kind of case a student with a laptop and no legal help usually can't build alone, the whole packet, edit history and Genesis test, laid out for a university panel that had started from a tool's answer and a bad feeling.

It worked. Quarterman was cleared and graduated on time. But "it worked" is doing a lot of quiet lifting in that sentence. It worked because his family had legal experience and a document platform that happens to log everything by default. Most students have neither.

What he actually wanted

Before any of the evidence, before the Genesis printout, Quarterman described the moment he found out about the accusation like this: "I managed to get back to my apartment before I had what I'm now recognizing was a full-blown panic attack."

That panic was earned. A zero and a formal academic dishonesty referral, for something he didn't do, that's the kind of accusation that feels final when you first hear it. There's no appeal button on a detector's verdict.

Looking back, what he says he wanted was smaller than a lawyer or a Google Docs deep-dive. "I wish my teacher asked me, 'Did you write this? Can you show me that you wrote this? Or can you just try and rewrite this one more time?'"

A conversation. A second draft, on a whiteboard, in office hours, watched. Something a detector never asks for and a human accusation should have started with, before it turned into a case file with his name on it.

The part that doesn't get a headline

Quarterman graduated. He also came out of the experience wanting to apply to the San Francisco Police Department, a strange, almost sideways outcome for a story about being wrongly accused by a piece of software. Make of that what you will. Getting falsely accused by an unaccountable system will apparently either radicalize you against institutions or send you off to build a career inside one that has to follow due process. He picked the second.

The professor, as far as the record shows, ran an essay through a tool because it felt too competent, trusted the tool's answer over the possibility that a student can just be careful and referred a case for academic discipline on the strength of a tool's answer. As far as the record shows, nobody checked whether that tool could also nuke the Book of Genesis before it got used to nuke a senior's transcript.

Keep receipts

Here's the plain version of what happened: a detector guessed, a professor believed the guess and a student had to reconstruct two or three hours of his own life to prove a negative that should never have needed proving. He won because Google Docs happened to be logging everything the whole time, and because his family happened to have the legal chops to make that log mean something to a panel.

That's a lucky defense. Nobody should count on it repeating. Most students don't have a lawyer for a father or a sister in tech on speed dial when a professor decides a tool's answer is proof. Most students have a laptop, a deadline and a sinking feeling that nobody's going to believe the truth once a tool has already said otherwise.

We don't build detectors, and after this story you should be glad of that, a coin flip that also thinks Genesis is a chatbot isn't a coin flip anyone should be graded on. We build the boring companion piece: something that checks the sources inside your own paper against the live record, so if a professor or an editor or a suspicious classmate ever comes asking whether your work holds up, the citations do exactly what Quarterman's edit history did. They show their work, timestamped, before anyone has to have a panic attack proving it.

Keep your drafts, your version history and your sources checkable. The tool that accused you can be out-documented. That's all it takes.