(this is more of a vent post but if i'll read any suggestions in the comments)

For the last 6 months, I felt like I've had some horrendous luck/consistency with my contests:

• EPIC Institute of Technology Round Summer 2025 (Codeforces Round 1036, Div. 1 + Div. 2): Misread 2124E - Make it Zero, lost a ton of time. (-63)

• Codeforces Round 1040 (Div. 1): Decent contest, could be more optimized but overall it's not bad. (+40)

• Atto Round 1 (Codeforces Round 1041, Div. 1 + Div. 2): Absolute disaster, somehow managed to save myself by still clearing 5 problems. (-71)

• Codeforces Round 1046 (Div. 1): Good contest, no catastrophic errors. (+78)

• Codeforces Global Round 29 (Div. 1 + Div. 2): 2147C - Rabbits costed me a lot of time and 4 penalties. (-47)

• Squarepoint Challenge (Codeforces Round 1055, Div. 1 + Div. 2): Got lucky with 2152F - Triple Attack, 2152E - Monotone Subsequence was a bit slower than what I wanted. (+60)

• Codeforces Round 1058 (Div. 1): Misread a condition in 2159C - Twin Polynomials ($$$a_i \le n$$$ instead of $$$a_i \le 10^{9}$$$), dropped around 110 places from that. (+4)

• Pinely Round 5 (Div. 1 + Div. 2): Lost an unreasonably absurd amount of time on implementing 2161E - Left is Always Right. (+5)

• Codeforces Global Round 30 (Div. 1 + Div. 2): Massively overthought 2164E - Journey, thinking took about 90 minutes. (-28)

• Codeforces Round 1064 (Div. 1): Was insanely close to solving 2165D - Path Split. (+12)

• Codeforces Round 1066 (Div. 1 + Div. 2): Lost a bit of time and 6 penalties to 2157C - Meximum Array 2. (-9)

• Codeforces Round 1069 (Div. 1): Overthought 2174C1 - Beautiful Patterns (Easy Version). Overflow bug on 2174C1 - Beautiful Patterns (Easy Version), accidentally deleted the partially finished 2174C2 - Beautiful Patterns (Hard Version) code while fixing it. Was 5 minutes too slow to submit 2174C2 - Beautiful Patterns (Hard Version). (-17)

• Codeforces Global Round 31 (Div. 1 + Div. 2): Absolute disaster again, 2180E - No Effect XOR solved around 5 minutes after contest ended. (-63)

• Good Bye 2025: Good contest, though I should have went for 2178G - deCH OR Dations instead of 2178I - Numbers or Fireworks. (+94)

• Hello 2026: Lost 70 minutes implementing 2183D2 - Tree Coloring (Hard Version), overthought 2183F - Jumping Man. (-36)

• Codeforces Round 1073 (Div. 1): Overthought 2190D - Prufer Vertex, finished implementation 1 minute after contest ended. (+15)

Any extra advice on how to be more consistent in contests, besides from:

• Doing virtual contests (last virtual contest I did was a while ago and I haven't had much time since).

• Practicing problems of similar difficulty and tags to the problem that caused the bad performance (didn't make a big difference).

• Being more relaxed during contests (didn't really make a difference either).

• Get more sleep (I haven't noticed much of a difference).

• Stop caring about rating (Would be nice but I doubt I can do that, at least not until i hit a specific rating).

GIven this problem:

Given $$$n$$$ integer pairs $$$a_i$$$, $$$b_i$$$, pick $$$k$$$ of them so that $$$\frac{\sum_{}^{}a_i}{\sum_{}^{}b_i}$$$ is maximized, output the maximum value.

Constraints:

$$$1 \le k \le n \le 200000$$$

$$$1 \le a_i, b_i \le 10^{6}$$$

I have this scuffed greedy algorithm which does the following:

1. Assume the result is $$$\ge x$$$, then let $$$u_i = a_i - xb_i$$$

2. Greedily pick the largest $$$k$$$ values of $$$u_i$$$.

3. Update $$$x$$$ to the new optimal value, if the new value is not better then stop.

After a lot of random tests, the worst case only made it perform $$$7$$$ iterations.

Are there any tests that makes it TLE, or at least run a lot more than $$$7$$$ iterations? If not, is there a proof? What is the runtime in the worst case?

#include <bits/stdc++.h>
using namespace std;

long long a = 1, b = 1000000;

void simpl(long long& u, long long& v) {
    long long G = __gcd(u,v);
    u /= G; v /= G;
}

bool doIteration(vector<pair<long long, long long>>& AB, int k) {
    //a-bx a-b(u/v) av-bu
    int n = AB.size(), i;
    vector<pair<long long, int>> U(n);
    i = -1; while (++i < n) {
        U[i].first = AB[i].first*b-AB[i].second*a;
        U[i].second = i;
    }
    sort(U.begin(), U.end());
    long long S = 0, d = 0;
    i = n-k-1; while (++i < n) {
        int u = U[i].second;
        S += AB[u].first; d += AB[u].second;
    }
    simpl(S,d);
    if (S == a && d == b) return false;
    a = S; b = d;
    return true;
}

bool comp(pair<long long, long long> u, pair<long long, long long> v) {
    //u.first/u.second < v.first/v.second
    return u.first*v.second < u.second*v.first;
}

int main() {
    ios_base::sync_with_stdio(false); cin.tie(0); cout.tie(0);
    int n,k,i; cin >> n >> k;
    vector<pair<long long, long long>> AB(n);
    i = -1; while (++i < n) cin >> AB[i].first >> AB[i].second;
    sort(AB.begin(), AB.end(), comp);
    while (true) {
        if (!doIteration(AB, k)) break;
    }
    cout << a << " " << b << "\n";
    
    return 0;
}

So I'm changing my pfp every day for december (because it's cool or smth)

Obviously that'll mean 31 pfps and i have decided that i will let other people pick them.

So for every day, I will pick the highest upvoted reply that I haven't picked yet and set it as my profile picture, obviously with a few rules:

1. Please no nsfw or offensive :prayer:
2. Ideally you should send an image.
3. If there are no more pfps to pick from then it will be a pick of mine.
4. No duplicates unless it's interesting (and no you can't just change a single pixel and call it a new image)
5. If it is heavily downvoted then it won't be picked. (somewhere below 10 downvotes?)
6. Maximum of 2 pfps per person.

Also i might add more rules if someone tries to loophole (or if i forgot one because stupid) (please behave properly).