What is in this guide

1 · The method

The principle every screen in the workspace is built on.

A course on the platform starts from the official syllabus: its lessons, its order, its terms. That is the base and it does not change — the student is examined on it, and a term translated to taste costs them a mark. On top of that base, two things are added:

2 · Building the course

From a syllabus to a course that runs.

1. Divide the syllabus into units

   One subject per unit, with its lessons in the syllabus's order. Unit 1 stays open and free in full — that is the platform's way in every course, not a temporary offer.

2. Put in the syllabus lessons

   One lesson for each, with the same name and the same order. Each takes the four steps in the next section.

3. Add the extra lessons

   After reading the whole syllabus and knowing where the gaps are. An extra lesson goes into its place in the sequence and carries its own mark.

4. Set the order and the conditions

   From the lesson order page: whether lessons open in sequence or all at once, and what opens each one. It has a student preview, so you see what they will see.

   The default condition is now one stated thing: a lesson counts as finished when the student scores 75% or more on its “Practice: Advanced level”, and the next lesson opens on that — not on opening every file or completing every activity. You can change it per lesson on the same page: choose other activities to count, raise or lower the percentage, or leave a lesson open with no condition at all.

5. Place the activities on the steps, and decide free or paid

   The “Lessons and access” page shows a lesson's activities in one table: for each one its step (The explanation · Lesson files · Assessments · Practice, or “automatic”, which puts it where it is expected), and its access, free or paid. Both are yours: open a whole lesson as a sample, or leave a single activity open; the student sees the change at once. What stays with the administrator is the course price, the enrolment methods and publishing it on the site.

3 · A lesson's four steps

The student sees the same four steps in every lesson; you decide what goes in each one.

4 · The question bank

All of a course's questions in one place, split by lesson and level.

The bank is where every piece of practice and every exam comes from. Questions are split by lesson and by level, and each one records its type and the thinking it asks for — so practice does not come out as one long recall test.

5 · Exams

An exam at the end of every unit, at more than one level.

A unit exam is composed from the exam bank on the same logic as practice: a question count, a time, a number of attempts, and a spread across thinking levels. You can leave out the questions that need human marking when you want the result to be immediate.

6 · Marking and review

Written answers are marked automatically; you are the one who confirms.

1. The automatic mark

   The answer is measured against the model answer and the mark scheme you wrote in the bank. Nothing is judged out of thin air: the scheme is the reference.

2. The mark waits for you

   It reaches you in the review queue as a proposed mark with its reasoning. The student does not see it before you confirm it.

3. If you change it, they are told why

   A changed mark reaches the student as a notification with the reason. That is what makes a student trust the number instead of arguing with it.

7 · The AI assistant

It answers students from your course's lessons, and you decide what it may read.

The assistant searches in three steps: the lesson the student is in, then the rest of the course, then it asks the student whether they want an answer from outside the course. The first two read from the course's knowledge base — which is yours to control.

The knowledge base page

In the workspace, under the course: «The assistant's knowledge base». It holds:

• Uploads in md or PDF, each tagged with the lesson it belongs to, so the assistant can return it in the right place.
• A switch per source, without deleting it. A source that is off is not read.
• Deletion of what no longer holds, and a view of what was imported and when.

AI on the platform: two jobs, not a general «intelligence»

The platform does not run a model in order to be able to say it runs one. There are exactly two jobs, each with its own model chosen in the platform's settings:

And how do you know whether to trust a suggestion?

There is a page called «how accurate the AI's suggestions are»: it compares each model's suggestions with the marks instructors gave themselves, over the answers that have both. Answers the suggestion marked on its own are left out of the comparison — they would only agree with themselves. So «how far do I trust this» rests on a number rather than an impression.

What is recorded

• A usage log with one line per request: the model, how big the request was, how long it took, and whether it worked.
• What is never written there: a single word a student wrote, or any key.
• The model runs on the academy's own server, so neither students' answers nor your course's content leaves for an outside service.

8 · Questions and sessions

What the assistant does not replace.

9 · Following the class

Who is working, who has stopped, and where they stopped.

10 · Course settings

The last part, and usually done once.

• Course settings: what the student sees and what is locked, the leaderboard and the badges.
• Translation: the course's text in the other language, for students who switch in their preferences.
• Prices and enrolment: the price of the course and of a session, with Unit 1 staying open and free.