Where the numbers come from and what the language model does
This page exists so that every number in the app can be checked. Each one says where it comes from, which screen it appears on, and whether it is a published figure, an estimate of ours, or a number we invented for now.
The numbers in the app, one at a time
Invented figures for now
We invented these figures so that the screen could be looked at. They are not real data and nothing can be decided from them. The screen they appear on says so too.
Number of questions in the app
2032
Where it appears: Practice and the mock exam
To be replaced: T-109
Published figures
These figures come from CERMAT's publicly released materials. The date we last verified them is in the footer of every page.
Matematika — Length of the test in minutes
70
Where it appears: Home page — “What the exam looks like”
Where it comes from: CERMAT — the format of the unified admission exam
Matematika — Maximum number of points
50
Where it appears: Home page — “What the exam looks like”
Where it comes from: CERMAT — the format of the unified admission exam
Český jazyk a literatura — Length of the test in minutes
60
Where it appears: Home page — “What the exam looks like”
Where it comes from: CERMAT — the format of the unified admission exam
Český jazyk a literatura — Maximum number of points
50
Where it appears: Home page — “What the exam looks like”
Where it comes from: CERMAT — the format of the unified admission exam
Matematika — Average number of questions in the test
16
Where it appears: Practice — choosing a topic
Where it comes from: CERMAT — tests and answer keys 2022–2026, checked 11 Oct 2026
Český jazyk a literatura — Average number of questions in the test
30
Where it appears: Practice — choosing a topic
Where it comes from: CERMAT — tests and answer keys 2022–2026, checked 11 Oct 2026
Lyceum admission cut-offs, lowest to highest (percentage score out of 200)
18–158
Where it appears: Chance of a place
Where it comes from: CERMAT — national exam results (2026)
The smallest number of admitted pupils any one cut-off was computed from
1
Where it appears: Chance of a place
Where it comes from: CERMAT — national exam results (2026)
What the language model does and what it does not
What it does
- It picks from explanations people wrote. When an answer is wrong, the model picks the explanation from a finished set that best fits what happened in the answer.
- It decides which question comes next when the same topic keeps going wrong.
What it does not do
- It does not mark answers. Whether an answer is right is settled by a fixed calculation in code — the same input always gives the same result and it can be checked line by line.
- It does not compute marks, percentiles or chances of admission. All of that is arithmetic in code.
- It does not invent explanations of mistakes. People write the explanations; the model only picks among them. If it invented them, they would sound right in the places where they were wrong — and in exam preparation that is the worst mistake there is.
- Nothing about a pupil or a parent reaches it. We send no names, no email addresses and no individual child's answers to the model.
What the app will never have
- Percentiles computed from the results of our own users. A comparison only means something against CERMAT's national data; from our own users it would be a comparison against whoever paid for the app.
- A child's rank among other children. Nowhere, in any form.
- Streaks, badges or countdowns meant to create the feeling that something is slipping away. The users are 14 and 15 years old and are under exam stress.
- A probability of admission as a percentage. From the published cut-offs the only honest statement is how it would have gone in a particular year, and that the cut-off moves from year to year.