misof authored easy and medium. Yes, the answer is medium is unique. Let's say there is bit $$$1$$$ at position $$$i$$$ if card $$$i$$$ is exactly there. This gives us a big binary number. This number must decrease with each operation, so it's optimal to decrease it by $$$1$$$ and there is just one way to do it.

I'm getting TC Newsletter, it has a section that lists upcoming SRMs.

ksun48 is congratulating tourist.

Thanks!

Both challenged :( and passed :)

For those who don't know the task: Final 1000 is a special-judge task where the contestant should return an array of doubles and the checker calculates certain values, which can have relative or absolute errors of at most 10^-7 to get accepted.

Here goes what I heard from the admin, misof. Against the challenge case made by Petr, my solution had errors around 10^-5 and hence it was a successful challenge. But at the same time writer's and tester's solution also causes errors around 10^-5. misof says the solutions are precise enough when run locally, so it might be an issue with handling doubles in the topcoder system.

The decision was announced before the closing ceremony. In short: the challenge scores are kept as is, while they rejudge the submitted solutions, tolerating much errors (seemingly 10^-4). Finally my 1000 passed.

Anyway, precision errors are really really hard to tackle with...

On the other hand, I feel I was cheated by the organizers (congrats on the correct solution to the problem, of course!).

From what I understand, you needed to return vector<double> in your solution. However, doubles turn out to be too inaccurate and these tiny rounding errors made the relative errors of the areas produced as large as 1e-5. Therefore, it was probably impossible to solve the problem within these limitations.

From my point of view, if statement says one need to fit within 1e-7, and no solution can fit within 1e-7, then, well, sh#t happens — no solution can solve the problem. Changing the allowable relative error to 1e-4 (?) feels:

• random (what if it's possible to cause even larger errors?)
• inappropriate (wtf, changing the statement of the problem after the end of competition? I understand you can't simply make the finals unrated here, but, well...)

It's so upsetting to me. Also, if the organizers chose 1e-7, they probably had some reason (isn't 1e-9 the default allowed error?). If they were afraid of these errors, they should have done everything to prove that their solutions will fit within 1e-7. They have not. Damn it, this is probably the most important round on this platform in the year, and you do this! You had more than 85 minutes to prepare this problem.

Disclaimer: I acknowledge that if the problem had been prepared correctly, the standings would have looked the same way they look right now. In some way, I "didn't deserve" to be higher, and I'm okay with this. Obviously, hos.lyric deserved her place, congrats! The organizers were crap. Thank you.

It's been so many times I've been deeply disappointed with a TC round. Maybe it's a good idea to take some longer break from it.

I'm sorry for such a long rant. I feel quite emotional about this and I just needed to show my point of view. Thanks.

Also, thanks to all other organizers, you did a great job and the everything else was amazing!

I agree, Topcoder has the best coverage, far ahead any other programming competition stream, ICPC organizers have a lot to learn.

It isn't solvable this way. Instead, you need a formula that involves product of sizes of connected components. I think there should be a saved stream on twitch with commentary and problem analysis.

Solving 26 problems from Atcoder dp contest: https://www.youtube.com/watch?v=FAQxdm0bTaw