The examiner

An exam's marker in AI form — what a persona holds, how yours gets chosen, and exactly what it changes on each surface.

Everything the platform writes for you is written for a particular exam. The examiner persona is how it knows which one. It is that exam's marker, described well enough for a model to imitate: what the exam looks like, what its marking rewards and punishes, the rubric an answer is scored against, and how its examiners speak.

There is no profession hard-coded anywhere. A persona is resolved from what you say you are studying, so the same machinery examines a dermatology fellowship, a solicitors' qualifying exam, a school physics paper and a job interview.

What a persona is#

One catalogue document, holding the things a prompt needs to be able to say:

• Who is studying — the learner it writes for ("a candidate sitting SQE1"), which sets the level, the vocabulary and what gets assumed rather than explained.
• What the subject is — the field, so notes are read as that field's material.
• Who is marking — the examiner's own identity, in the first person, as the marking prompts adopt it.
• The exam format — what the real paper or station actually is. Left empty when unknown; an invented format is worse than none.
• What marking rewards — the marking emphases, the two or three things this exam gives and takes marks for.
• How a good answer is shaped — the guidance behind the Must not miss core and a station's what-to-cover.
• How it speaks — the accent and register for anything read aloud.
• A rubric — the criteria a spoken answer is scored against, out of which the viva score sheet is built.

One examiner per subject. A subject with none set falls back to a neutral examiner — a rigorous but fair senior examiner for a professional qualification, with a five-criterion rubric that reads for any oral exam — never to another subject's. It is deliberately never a leak: the Exam, Oral and Lecture tabs ask you to set one first, because a paper is marked against a rubric and it should be the right one.

A persona's locale says which country's exam it is, not which language it speaks. Someone in Australia sitting a US exam wants those two to disagree — which language things are written in is a separate setting, on your account and on each subject.

How yours is chosen#

The strip at the top of a subject's dashboard is where an examiner is set, and it is the only place. Type what you are sitting — an exam, a qualification, a course, any field.

Two steps answer it. An exact match against a persona's known names and abbreviations is free and instant. Anything else goes to a model with the whole catalogue in front of it, which comes back with one of three things: a best guess (with the near misses under Did you mean one of these?), a set of options when the name is ambiguous — a college that runs several exams, say — or nothing at all. A model beats a similarity search here: it knows "A.M.C." is the Australian Medical Council, and a spelling-distance never will.

Picking one replaces whatever was there. You can change it whenever you like; see switching examiner for what that does and does not touch.

Asking for one we do not have#

If the catalogue does not have your exam, Write a draft for it does. It takes about half a minute, and the examiner it writes is yours to use immediately — assigned to the subject before any human has looked at it, and marked Draft · under review on the strip.

That immediacy is the point. Waiting on a queue used to mean the Exam, Oral and Lecture tabs simply refused until someone got to your request, which is a wall rather than an inconvenience.

Every draft is then read by hand. Approved, it joins the catalogue for everybody; declined, you keep it anyway and the catalogue just does not gain it. Either way the answer arrives in your notification bell. If somebody else asks for the same exam before it is reviewed, they join your draft rather than minting a rival one, so one decision answers everyone. Drafts are capped at five a day.

Cards#

Almost nothing. A flashcard prompt reads only who is studying and what the subject is — the examiner sets the level a card is pitched at, and no more than that. Cards are cut from your notes, and there is no marking in them.

Exams#

The most examiner-shaped surface after the viva, at both ends.

Setting a paper reads the marking emphases and the exam format, so the paper is the shape the real one is and asks what that exam actually asks.

Marking it reads all of that plus the examiner's own identity, and marks in the first person as that examiner. The comment on a lost mark, what it says you missed, and the model answer are all written from that standpoint. The Must not miss core on a board is built the same way — the examiner's identity, and its guidance on how a good answer is shaped.

Orals#

The whole thing, end to end. The examiner decides:

• the instructions the live examiner is given, so it questions the way that exam questions;
• the accent and register it speaks in;
• the rubric your answer is scored against, and therefore the criteria rows on the score sheet and what "the one thing that would have earned the rest" says;
• what a station covers, and the what-to-cover checklist you can read before you sit it.

None of that is sent from your device. The client names which examiner; the rubric and the instructions are read from the catalogue on the server, so nothing can be injected into them.

Lectures#

On a lecture the examiner does three things and no more.

It writes the script in that examiner's voice. The narration's style is the persona's speech style plus a fixed teaching clause — the voice is teaching, not marking — and the voice itself is pinned to one narrator (coral) for every examiner. So what changes between examiners is the accent and the register, not the speaker.

It pitches the script at that exam's learner. Both the planning pass and the script are written for "a candidate preparing for X", which is what sets the level, the vocabulary and what gets assumed instead of explained.

It stamps the lecture. A lecture records the examiner it was written for, and narration reads that examiner rather than the one currently in force — so a lecture written under one examiner still sounds like that examiner after you switch.

What does not reach a lecture: the examiner's identity, its marking emphases, its exam format and its checklist guidance. None of the marking machinery is in the lecture prompts at all — a lecture is taught from your own notes and nothing about it is scored. The examiner therefore matters less here than on Exam and Oral. Changing it changes how the next lecture sounds and who it is pitched at, and nothing else.

Switching examiner#

Changing a subject's examiner changes what is made from now on. It never re-marks, rewrites or re-narrates anything.

That is possible because every paper, lecture and station records the examiner it was made for when it was made. The rule it buys is worth stating plainly:

Making something uses the examiner in force. Everything done to it afterwards uses the examiner it was made for.

So a paper set under your old examiner is still marked by that examiner, and its lecture is still read in that voice. Where an artefact was made for a different examiner than the one now in force, the list says so and names both — and what to do about it is your call, because answering it means paying for a fresh generation. Anything made before the platform recorded this reads as unknown, never as changed.