"Why can't you just scan it?" How Tziyun B'Rega began

Every product has a first sentence. For Tziyun B’Rega it was a question in an email: why can’t you just scan a multiple-choice exam and have it graded in moments?

It sounded simple. This is the studio’s account of what happened next, checked against the project’s history where the history can check it. It is a story about looking around.

Key points

• It started with one email: why can’t you just scan a multiple-choice exam and have it graded in moments?
• AMC, an open-source exam tool, already reads ticks. We had our own reasons for not building on it, and they are not a verdict on AMC.
• The idea was to keep everything on one sheet: the student ticks right next to the question, the way a teacher checks by hand.
• Today the app reads a small box ticked by hand in pen. We have proven pen, not pencil, so we do not claim pencil.

The first search

We started by searching for what already exists. We found AMC, an open-source tool for making and marking multiple-choice exams, written mostly in a language called Perl. It is free software that other people built and share, and it is still maintained. We were glad it was there.

What caught our eye was the sample exam on AMC’s overview page (opens in a new tab). The answers are small empty boxes beside each choice, on the same page as the question, marked by hand with an X. They are not filled bubbles. That is what made us dig in.

The project history shows how seriously we took it. The repository began on 31 October 2025. The next day, 1 November, a commit added AMC to the project, and a first small test ran from the new program. On 13 November the AMC code was removed again. So AMC was part of the project for under two weeks, long enough to learn from it and no longer.

During that search we asked an AI assistant to read AMC’s code and show us where it checks the answers. That is how we learned the name of the whole field: OMR, optical mark recognition. Until then we had no word for what we were trying to do.

The lottery-ticket lesson

An engineer friend gave us the big picture in one comparison. Think of lottery tickets. The ticket, the scanner and the machine behind it are all built for each other. Everything is laid out in a fixed pattern, so the machine knows exactly what it is looking at.

That was our takeaway too: classic exam scanning works because the paper, the scanner and the software are designed together.

The bubble problem

Then the owner remembered something from childhood. Growing up in the USA, the owner took the standardized tests where you must fill each bubble in completely. It is not how a student naturally works. The owner remembers spending more time filling in bubbles than answering the question.

And it is more than the filling. In those tests the answers go on a separate page. Your eyes leave the questions, find the right row on another sheet, and come back. The student’s attention is split between two pieces of paper, when it should be on the exam.

So the idea became: keep everything in one place. The student answers right on the exam paper, next to the question, with the natural thing, a simple tick. That is also how a teacher checks an exam by hand. We did not invent a new way of examining. We kept the human way and added the automation, so the computer has to recognise a tick made by a person.

Why we did not build on AMC

AMC already reads ticks, and we want to be clear about that. The idea was not new to it. What we had to decide was how to build our own product, and these were our reasons. They are our reasons, not a verdict on AMC.

• It is written mostly in Perl. Perl is an old language. At the very least we wanted something modern.
• We did not need the whole program. Once we understood the small handful of values AMC’s marking depends on, we saw we could do that part ourselves.
• It comes as one whole program. To ship it we would have to ship the whole program, and for software that people download and install, that is a lot of disk space to waste.
• Memory. Our view is that modern languages, once they are turned into a finished program, use the computer’s memory more safely.

AMC was also built for a different job. Its own documentation says it is designed for use under Linux, though it can run on Mac and Windows with more effort. Ours is a Windows program that has to work offline.

Building a way to check ourselves

A tick drawn by a real person is unpredictable, so we could not just hope the new reader worked. In the project’s OMR repository we built a whole QA application, a tool whose job is to show us what the reader did on each page, so we could prove things one step at a time instead of guessing. The history shows this work in June 2026: the rewrite began on 3 June, and the first version of the QA viewer is dated 8 June.

One part of the road was long and hard: making the program reliably make sense of a real scanned page. A real scan is not a clean file on a screen. We are deliberately not going to describe how this works, because that part is ours to keep. We will say that a lot of the time went there, and that the QA tool is what kept it honest.

What the question taught us

Looking back, that simple question was a good one.

Tziyun B’Rega kept the human way of examining and made the machine adapt to the student. Today the app reads the small box a student ticks by hand in pen. We have proven pen. We have not yet proven pencil, so we do not claim it.

That one email did not make the product. The search left us with one rule: do not ask people to change how they work if the software can change instead.