What a contribution record can and cannot tell you

This page exists because the gap between those two things is the easiest thing about this product to oversell, and because the pressure to oversell it will be constant. It is written for anyone who has to answer the question honestly: an academic deciding whether to trust the numbers, an integrity officer deciding whether they are admissible, and anyone from Dwixel on a call.

What Dwixel does claim

One sentence, and it is the only one on this site that describes the integrity value of the product: Dwixel keeps a record of how the document was written, which makes unusual patterns visible to the person marking it. Every word of that is supported. The record exists, the patterns are real, and the person marking it is the one who decides what they mean.

What Dwixel does not claim

• ·It does not detect contract cheating. A systematic review of the field found no reliable objective measure of it, which is why the review calls for objective behavioural measures rather than reporting one. ¹
• ·It does not detect AI-generated text, and it does not attempt to. Nothing on this site should be read as implying otherwise.
• ·It does not identify plagiarism. Similarity checking is a different tool solving a different problem.
• ·It does not distinguish text pasted from a student’s own draft from text pasted from anywhere else. The two look identical in the record, and a great many people write by drafting elsewhere and pasting in.

Why the paste signal is weaker than it sounds

Dwixel records when a large block of text arrives at once, and keeps that text out of the contribution share. It also records typing that follows a large paste, because typing on top of pasted text is one way the exclusion gets undone. Neither of those is evidence of anything on its own. A student who drafts in Word, pastes the draft in and then edits it produces exactly the same shape as a student doing something they should not, and no amount of processing separates the two. The product surfaces the magnitude and asks the marker to make a judgement; it does not make one.

Why the weaker claim is the more useful one

The stronger claim fails in the only place it matters. If a student is excluded from a programme partly on the strength of something Dwixel showed, and the claim behind it does not hold, the failure is not a marketing embarrassment; it lands on a person. The relationship between contract cheating and assessment design is also better understood than the detection of it, and the evidence points at conditions rather than catches: contract cheating tracks dissatisfaction with the teaching environment and the perception of many opportunities, with strong student–staff relationships acting as a protective factor. ² A record that helps a marker have a specific conversation with a specific student does more for that than a detector would.

If you are asked whether it detects AI

The honest answer is no, followed by what is true: Dwixel shows how a document came into being over time, and a document that arrives in three pastes on the final evening looks different from one written across four weeks. That difference is visible, it is worth a conversation, and it is not proof. Anyone who tells you a tool can do better than that is selling something this one does not do.