How the content is made
Where the statements come from, what each verification label means, how the exercises and the code are checked, and what has not been checked.
Archive statements
Every IMC, SEEMOUS and NKUA selection statement is transcribed from the official PDF. Nothing is ever written from memory. The image of the original always stays next to the transcription, so you can compare for yourself.
Each transcription carries one of three labels:
| Label | Meaning |
|---|---|
| Transcribed | Written from the image, with no second check. |
| Automatically checked | A second, symbol-by-symbol comparison with the image was made, not by a person. |
| Checked by a person | The editor compared it with the original. Only the editor sets this label. |
The build also compares the words of each transcription with the PDF text. That check does not see symbols, subscripts, limits or the direction of inequalities. Errors in the original are kept as they are.
Problems with a full write-up
Problems with hints, a solution and moves start as drafts and do not appear on the public site. They are published once the editor approves them. The site's solutions are our own. Where an official solution exists, it is linked.
Exercises and theory
The ladder exercises are our own, for practice. They are not contest problems. Each exercise has a small Python program (numpy, sympy, mpmath) that checks its claims numerically. The build runs it and stops if it fails. A numerical check catches errors in formulas and constants. It does not prove that the argument is correct.
The theory sheets have been checked in the same way and read by the editor.
C++ code
A template is labelled "tested" only if both of the following hold:
- it was compiled with
-Wall -Wextra -Wshadow -fsanitize=address,undefined; - it passed a stress test against a slow, obviously correct solution, on thousands of random cases.
The test result records the file's SHA-256. If the file changes, the label disappears until the test runs again.
Programming: what the site does not do
- It does not copy CSES statements or USACO Guide text. It only links to them.
- It does not run a judge. Submissions go to CSES or Codeforces.
- It never asks for a Codeforces password. It reads only public data from the official API, at most one call every two seconds, cached for one hour.
Review
When a problem comes back is computed by the FSRS algorithm (the ts-fsrs library), from how each attempt went. The computation runs in your browser.
Dates and rules
Contest dates and qualification rules are taken only from official pages, with a link and the date they were checked. Anything that could not be confirmed is stated explicitly as "not checked".
If you find a mistake, tell us.