Anti-roadmap
Things we will never build.
Every product has things it has not got round to yet. This is the other list: features people have asked for that we have decided against, and why. If one of these ever ships, something has gone badly wrong.
Notice: not in this lab, now or later.
SafeExamOS
Applies to every version
Webcam feeds
The camera is never switched on, streamed, or saved.
A camera makes every student a subject of surveillance, and every bedroom, face, and nervous habit becomes data. Lab exams already have an invigilator present. If a student needs to be watched, a person can do that with judgement and context, and without producing a recording that outlives the exam.
Screen recording
Screen contents are never captured or recorded.
When the network permits only the exam’s sites, the screen holds nothing worth recording. A recording would capture everything a student typed, tried, and deleted. It would then need to be stored, protected, and eventually reviewed. Stopping the problem at the network is the cleaner approach.
Keystroke logging
Student keystrokes are not recorded.
Keystrokes amount to a transcript of someone’s thinking, including passwords and half-formed ideas. Whatever they reveal about rule-following, a network log reveals more directly and without capturing everything a person typed in the process.
Microphones and room audio
The microphone is never on.
The reasoning is the same as for the camera. A lab exam takes place in a room with other people present. Audio would capture them too, and they have not agreed to any recording.
AI “suspicion scores”
No automated model scores or rates student behaviour.
Automated flagging systems have a documented record of misreading darker skin, disability, anxiety, and ordinary fidgeting, and their flags tend to be treated as findings. SafeExamOS reports facts: this seat requested this hostname at this time, and the request was blocked. A person interprets what that means.
Face recognition and biometrics
No face matching, ID photos, or biometric templates.
Identity verification works the way it always has in a lab: a seat, a roster, and an invigilator. Biometric data is among the hardest categories to protect, and it cannot be changed once it is compromised.
Eye tracking and “attention” metrics
No tracking of where students look.
Students look away from the screen when they think; measuring this adds stress and reveals nothing about the exam itself.
Software on students’ own devices
Students are not asked to install anything on personal laptops.
SafeExamOS runs on lab machines that the institution owns and manages. Requiring students to grant exam software deep access to a personal laptop is a different arrangement, and one this system is not designed to make.
Selling or mining exam data
Request logs belong entirely to the institution.
Request logs remain with the institution that ran the exam. They are not aggregated across customers, used to train models, or sold in any derived form.
Submitting on a student’s behalf
Students submit their own work on the actual site.
Collecting answers centrally would be convenient, but it would make SafeExamOS a party to every answer and suggest that the system acts on behalf of the student. SafeExamOS secures the environment; the student does the exam.
Our own judge or grading
SafeExamOS is not a contest or grading platform.
CodeChef, DOMjudge, Moodle, and similar platforms already handle this well. SafeExamOS makes a lab safe for using them. Remaining outside the grading process keeps this system outside disputes about marks.
Replacing the invigilator
An invigilator should remain present in the exam room.
A secured lab eliminates most opportunities to cheat; it does not eliminate the need for a person who can answer questions, notice a student who is unwell, or manage a power failure. That person is better deployed helping students than watching screens.
Threatening copy
Blocked pages show plain text; the timer stays calm.
A blocked page states that the site is not permitted and points back to the list. The timer does not turn red. Students under time pressure benefit from calm, plain language, and most blocked requests are typos.
How do you catch cheating?
Mostly by making it hard. If the only sites that load are the exam’s, there is not much to cheat with. What is left is a log of facts: which seat asked for which site, and when.
What about remote exams?
We do not do them. If an exam has to happen at home, proctoring tools exist for that. We would rather do lab exams well than remote exams badly. How they compare
Who decides what goes on this list?
We do, and we publish it so you can hold us to it. If you think something belongs here, tell us.