A degree should mean the student can explain the work.

08Questions

Questions we expect

Does this add another thing to review?

No — that is the point of it. The pass through every session goes away. What arrives instead is two or three recordings per section where the explanation didn’t match the work, each with the student’s own words quoted beside the finding. If it ever becomes a second queue on top of the first, it has failed.

Is there a camera?

No. Camera-based flagging is the part that produced both failures above, room scans have been ruled unconstitutional elsewhere, and video is exactly the review burden faculty are asking to be rid of. Noeta uses voice and the open microphone only.

Then how do you know they weren’t reading off a phone?

For a determined student, we don’t — and we tested that rather than assumed it. Photographing an on-screen question, handing it to a chatbot and reading the reply back scored 93 out of 100 with nothing raised. That is why the questions are now spoken and never displayed: there is nothing to photograph. Relaying is still possible — a second device can record audio — but it costs a round trip on every question, and that wait sits in the recording between noeta finishing and the student starting. Noeta reports the wait; it does not claim to have caught anyone. No remote check can honestly promise more, and the research on camera proctoring says the same.

Isn’t this hard on students with speaking anxiety?

It is the objection worth taking seriously, and the design answers it with time rather than with an exemption. Delivery is explicitly not graded — noeta assesses reasoning and refuses to score cadence, fluency or confidence. There is no visible timer and no cost to thinking: after about forty-five seconds of quiet noeta says some version of “take your time,” never “hurry up,” and a student can ask for the question again as often as they need — re-asks are recorded on the session, but nothing scores them. It is also a machine rather than a room of peers, and repeated low-stakes speaking is the established way apprehension comes down. What anxiety does not do is turn the check into a typing exercise. Typed answers are an accommodation for a student who cannot speak, granted by the instructor through the same route as any other accommodation on a campus — and today that setting applies to a whole section, so a per-student grant would need roster integration the demo has not built.

Is student voice data used to train anything?

Not in the demo, and not in a pilot without written, revocable, opt-in consent that carries no grade consequence. A recorded voice is a biometric identifier — it cannot be de-identified the way a name in a spreadsheet can — so any training work would use derived text and instructor labels with identifiers stripped, never raw audio, and would go through UWF’s IRB if it counts as research.

What if a student’s microphone fails?

Noeta asks the question again. If two answers in a row come through silent it stops and tells the student the microphone isn’t being picked up, rather than recording blanks. A session that captured nothing is reported to the instructor as a session to re-run — never as a finding about the student.

Is this fair to non-native English speakers?

It scores substance, never accent, fluency or delivery, and the check can run in the student’s strongest language where course policy allows. The comparison worth making is against the status quo: AI detectors falsely flag non-native English writing at 61%.

Who decides whether something is misconduct?

The instructor, in every case — and that is a position on integrity, not a way of avoiding one. A misconduct finding is an academic judgment that belongs to a faculty member and a process. Noeta never assigns a grade, never issues a verdict, and posts no penalty automatically. Where it has no evidence it says so; “insufficient evidence” is a first-class result rather than a euphemism for guilt.