What a writing record cannot prove

The short answer

A writing-provenance record is positive evidence of a process. It can show that a named person worked in a document over time, what grew by typing and what arrived in chunks, and that the record has not been changed since it was sealed. It cannot show what the writer meant, what they did in another window, or that a missing record is suspicious.

That limit is the point. A detector fails by accusing the wrong person. A process record fails by not existing, which proves nothing. The definition and the four properties that make a record checkable are in what writing provenance is.

Five things the record does not say

Intent. A paste is a paste. It might be a quotation, a sentence moved from the writer's own notes, a citation, or text from somewhere else. The record can mark the arrival. It cannot say why.

Work outside the recorded document. Notes in another file, a conversation, a book, a draft written before recording started — none of that is in the record. Absence of those steps is not evidence they did not happen.

That every keystroke was original thought. Someone can type from a source in front of them. A gradual history, including a Google Docs version history, can be produced by a person transcribing. Version history is still useful; its ceiling is covered in Google Docs version history as proof. Google's own history is a recovery feature, not an authorship seal.

That no tool was used anywhere. Spellcheck, a dictionary, a search, a calculator, an editor's suggestion on a different document — the record of this document does not inventory the writer's whole desk.

Guilt, when there is no record. Recording is off until the writer turns it on for a document. A student who never started a record is in the same position as every writer before these tools existed. They still have the ordinary layers — drafts, history, a conversation about the work — set out in how to prove you didn't use AI. Turnitin itself says a detector score should not be the sole basis for a decision; a missing provenance record is even less than that. It is not a score. It is a blank.

What you can still say, precisely

If the seal checks, you can say: this is the sealed record, and it was not altered afterward. How to read that result, including the honest reasons a check fails, is in what checking a seal actually shows.

If the record shows a passage arriving in one insertion, you can say: this passage was not typed in character by character inside the recorded session. You cannot say "therefore AI" without something else.

If the record shows hours of edits by a named person who consented to the recording, you can say: this person did this work in this document, and the record is checkable. You still cannot say they had no help of any kind. No document system can prove a negative that large.

Why the limit protects the writer

Tools that claim to see everything become surveillance, and surveillance records are easy to overread. A consent-based record is smaller on purpose. The writer turned it on, the writer can hand it over, and anyone can check the seal without an account at the public verify page.

An integrity process that treats "no record" as a confession has inverted the tool. The record is something a writer may bring. It is not something an accuser may demand as the price of being believed.