Skip to main content
Exam Evaluation marks written exams. Your platform uploads the question paper, the AI reads out the questions and their marks, your platform supplies the model answers, and then each student’s answer sheet is scored answer by answer. The API has a publishing step, and marks can be corrected before and after it. What it does not have is any notion of who is calling. Every call is authenticated as your API account, not as a teacher or a student, so your platform decides which of its users may upload, correct, publish, or read a result. Where this page says “a teacher”, read it as “a call your platform makes on a teacher’s behalf”.

​
What does the AI read from the question paper?

It breaks the paper into the same structure a teacher sees, so marks can be attached to the right place. A teacher can correct anything the AI got wrong, and can also add or delete a question, a sub-question, or a model answer.

​
What are model answers?

A model answer is the answer you expect for a question, together with the marks for each step of it. You upload the answer key as a PDF, images, or a link to a PDF, and the AI matches each answer to the question it belongs to. Step marking is what makes partial credit work. If a question is worth 5 marks and a student gets the method right but the final number wrong, the AI can award the method marks.

​
What do you get back for each answer?

Section totals and the paper total are worked out from the question marks, so they stay correct after a teacher edits a mark.

​
What can be changed, and by whom?

The API accepts all of these edits from your platform. It does not check who is asking, so your platform decides which of its users may make them.

Before answer sheets are marked

Edit the text and marks of any question or sub-question, edit a model answer and its step marking, add or delete a question or a model answer, set how many questions a section requires, and mark each item as approved.

After answer sheets are marked

Edit a student’s marks and the answer text, record a reason for the change, and mark the answer as approved.
Questions and model answers lock as soon as the first student answer sheet is sent for marking. After that the API refuses any edit, addition, or deletion, so the paper cannot move under students who have already been graded. Student answers and their marks have no such lock. They stay editable at every stage, including after results are published, and publishing again does not close them. If you need marks frozen once they are out, enforce that in your platform.

​
How are results published?

Publishing sets a flag on the exam, in two stages, and records the date each one was set. It is what opens the provisional result and final result endpoints.
1

Read and correct the scores

Your platform reads the AI’s scores and corrects what needs correcting.
2

Publish provisional results

The provisional result becomes readable: the marks, the model answers, and the feedback. The date of publishing is recorded.
3

Handle queries

A student who disagrees with a score raises a query on it. See below.
4

Publish final results

The final result becomes readable, with its own publishing date.
Publishing applies to the whole exam, not to one student, so you cannot release one student’s result on its own.
Publishing is not a privacy control. Two result endpoints refuse to answer before their stage is published, but a third one returns the same marks and feedback for a marked answer sheet with no publishing check at all. Nothing in the API tells a student’s request apart from a teacher’s. So publishing is a useful signal for your own screens to follow, and only your platform can actually keep a result from a student.

​
What happens when a student disagrees with a score?

The student raises a query against a piece of feedback and says what they think is wrong. That triggers an AI re-check of the answer. The re-check can revise the marks. It writes a justification for whatever it decides, whether or not the marks change, and the query is then marked as resolved. A teacher can also move a query to in review and keep it open, and either side can add follow-up messages to the same query. A query is one of three kinds:

​
Are there class dashboards and a student roster?

Yes, and they are not part of the documented API. Three dashboard views and a roster import are built and running, but they are deliberately left out of the API reference, so your developers cannot pick them up from the docs. Talk to CrazyGoldFish if you want them, and treat them as arranged with us rather than self-serve. Dashboards read the marks the AI has already awarded, and they do not wait for publishing. You do not need the roster to mark an exam. Pass your own student ID and name when you create each answer sheet instead.

​
What are the limits?

Exam Evaluation does not accept audio or video answers.
The readability check does not cover everything. It runs on PDFs and photographs that you upload. It is skipped for a Word file, and skipped entirely for anything you send as a link, which also means a linked PDF is not checked for size or page count. If you send work by link or as a Word file, check the quality before you send it.

​
Frequently asked questions

Yes. A student’s answer, its marks, and the approval flag stay editable, and a reason for the change can be recorded. There is no check on publishing state, so a published mark can change. Questions and model answers are the parts that lock. If a published mark must not move, your platform has to stop the call.
Uploaded PDFs and photographs are checked for readability before marking. If the photo is out of focus, badly lit, or at too sharp an angle, the upload is refused with a message asking for another photo. Retake it in better light, hold the camera square to the page, and upload again. Word files and files sent as a link are not checked.
Each exam, model answer, and answer sheet holds at most one PDF, and a PDF cannot be combined with images. If a student’s answer sheet is spread over several files, either merge them into one PDF of 50 pages or fewer, or upload them all as images.
Only until the first answer sheet is sent for marking. After that, editing a question, a sub-question, or a model answer returns an error, because students have already been graded against the old version.
The exam, model answer sheet, or answer sheet moves to a failed status, and the matching endpoint returns an error saying so. Upload a clearer file and send it for processing again.
That is your platform’s decision. The result responses include the model answer and its step marking next to the student’s own answer and feedback, so the data is there for you to show or withhold.

​
Next steps

How it works

The full flow from question paper to published result.

Use cases

How schools and platforms use exam evaluation.

Developer guide

Evaluate an exam with the API, step by step.

Answer sheet upload link

Give a student a link instead of building an upload page.