Assessment only works if both sides can trust it

A student submits a thesis. Somewhere on the other side, an examiner runs it through a detector and gets back a number with no error rate attached. If that number is high, a year of work is suddenly a conversation about honesty — and the student has no way to know it was coming, and no way to argue with it.

That is the problem we build for. Not cheating: uncertainty. It falls hardest on the people least able to absorb it — students writing in a second language, whose plainer sentence structures these tools systematically over-flag, and careful writers whose formal register is exactly what a detector rewards.

What we do about it

We publish how often we are wrong. Every verdict carries the measured rate at which human-written academic texts receive that same verdict, broken down by language and document length. No detector we know of publishes this. It is the number that decides whether a result means anything, and we put it next to the result rather than in a footnote. The corpora and the method are on our accuracy page, and every figure there traces to a script we can re-run.

The same tool for both sides. There is no student edition and no institutional edition. A student can check their own work before submitting, and an examiner can check the same document afterwards, and both get the identical verdict and the identical error rate. A tool that told each side something different would be worth nothing to either.

Students stay anonymous. We never ask for a name, a university, a supervisor or a matriculation number. Payment and document storage live in separate systems joined only by a derived code, so what you checked is never linked to who paid. Using this site is not evidence of anything, and we made sure it cannot become evidence of anything.

We say what a detector cannot do. It measures style, not authorship. It cannot know who wrote a sentence. Where a language or a document length is untested, we say so instead of implying coverage. When a measurement contradicts something we wanted to be true, the measurement wins — that has already happened more than once, and it is written down.

What we will not do

We do not sell a way to beat a detector. Our writing tool removes filler phrases, hedges and tired transitions — the things that make any prose worse and that a supervisor would strike out anyway. It is the same advice a good writing centre gives, and we are content for a professor to read exactly this page and know what we offer.

If a piece of work is largely machine-written, rewording it does not fix the problem, and we say so instead of selling a rewrite that would.

Who we are

ThesisDraft is operated from the European Union by Morphica Technologies. It is small and new. We would rather tell you that than borrow credibility we have not earned — which is the same reason our verdicts come with an error rate instead of a confident percentage.