I have just visited the official IOI2025 and read the Contest Rules here, when I suddenly saw this:

Each task is divided into some number of subtasks, each worth a portion of the total points. Each subtask is either a programming subtask (the solution is source code) or an output-only subtask (the solution is a set of output files).

For comparison, here is the 2024 version:

Each task is either a programming task (the solution is source code) or an output-only task (the solution is a set of output files). Each task is divided into some number of subtasks, each worth a portion of the total points.

It seems like we could see a lot of output-only tasks this year =))

I tried to find the 2024 IOI General Assembly meeting minutes but there seems to be none. International Committee is the only IOI committee that published its minutes, but their contents are totally unrelated to my concern. I searched the IOI Journal and it's useless also. I don't know if we have ever discussed this yet in Codeforces.

Therefore, I am currently wondering:

1. Did I miss anything?
2. Who initiated this idea? How did they came up with such a hybrid format? and Why did the IOI decided to change the format this year?
3. Has there any contest that had experimented with this format? How would this actually works this year? How would this affect the contestants' strategies and workflow?

Do you guys have any other ideas or questions about this new format?

Recently, I have seen a lot of blogs talking about the issues of cheaters. Therefore, I am currently thinking about using automatic system to catch them.

Currently, the most well-known automatic system for assisting of detecting plagiarism is MOSS(from Stanford). At first, I asked myself, why did not Codeforces use them? However, I look at the number of participants of each contests; it turns out that the count is approximately under 30000. So, we have to compare $$$4.5*10^8$$$ pairs of source code!

Assuming that the system can check $$$10^4$$$ pairs per second, we will need $$$45000$$$ seconds, which is just more than half a day, the same length as hacking procedure of Educational Rounds. But I believe that limit is much lower (I have not used it).

Is there any assistance like that could run that fast, if not MOSS? Is there any solutions that can drop the complexity of $$$O(n^2 * t)$$$? (assuming $$$t$$$ is the time for comparing a pair of code)