ICPC NAC 2026 — A North American Championship Diary

Day -inf#

Where to start.

At last year’s ICPC NAC I performed terribly and placed near the very bottom of North America. After that I didn’t want to touch ICPC at all. I interned all summer, and right up until the ICPC Regional I had barely written any problems — my algorithm skills had decayed to the point where I was failing OAs.

Rationally I knew that a dozen or so problems would bring the feel back, and that I had no business failing an OA. The problem was that during that stretch I didn’t want to touch it at all.

By ICPC NAQ it turned out no unusually strong contestant had suddenly appeared at our school, so I teamed up with last year’s teammate Jeff plus a new teammate, Yehor. I didn’t prepare for the Regional either, the problems were fairly bad, and after writing the warm-up problem I lost interest and left the rest to my teammates. Looking back, I was being a bit of a jerk. Luckily they’re good people and never held it against me. Grateful for that.

And so my teammates carried me into the North American Championship, ICPC NAC.

Day -inf / 2#

Over winter break I realized I had to make some trade-offs. Going for the World Finals makes it very hard to juggle everything else. After talking to a lot of friends and thinking it over calmly, I decided to take one more shot at WF. I didn’t want to leave myself with regrets.

So on 2025-12-29 I restarted rehab training. Just like during my secondary-school contest years, I started logging my daily problem count and rehab progress, and kept it up all the way to ICPC NAC. Slowly I rediscovered that grinding algorithm problems and learning new algorithms still gives me the strongest positive feedback. I really do love the jolt a beautiful solution gives you.

Over the break I also went through WHITE ALBUM2 in one pass. It moved me a lot. WHITE ALBUM2 is a masterpiece; Maruto is a god.

Once the semester started I immediately got my teammates training together, roughly once a week. They were all enthusiastic, Jeff especially — genuinely passionate. The first time we VP’d a past ICPC NAC, we realized we actually had a shot at WF, and just like that we had confidence.

During training I also gradually realized how talented my teammates are. They may not know that many algorithms, but they are very sharp. Right after I finished explaining the closed form for Catalan numbers, one of them turned around and derived the combinatorial interpretation himself. Genuinely startling.

It was also during this period that I truly understood that ICPC is fundamentally a team contest. Since starting college I seem to have always been thinking in terms of solo play: solve it if I can, and if I can’t, check out. I had never really trusted my teammates, never really expected them to share the load, always defaulting to being the backstop myself.

This time I really felt it: to reach WF you have to trust your teammates, you have to build a team strategy — and it looked like we genuinely had a chance. So I trained with the team on weekends while continuing my own individual rehab. I knew I was nowhere near my peak, but my form had come back a fair bit, and my mindset was far better than in high school, which is the thing that affects my performance most.

Day -10 to -1#

Spring break landed right before NAC, giving me one continuous block of free time to train intensively. And the very algorithm that ended up carrying me into WF was one I reviewed during that stretch. Looking back, that feels like fate of a sort.

During this period I also worked up the courage to revisit the 2022 provincial selection — something I hadn’t dared touch in the years since I retired — and finished the problems I couldn’t solve on the day. It turned out that many of my core ideas back then had actually been right, and I’d failed to finish for various other reasons. A genuine shame. I was a little down for a few days, but I had finally faced and gotten past my own demon, which felt pretty good.

Also during rehab training, I suddenly started dreaming frequently about my high-school contest days. That hadn’t happened once in two or three years. Occasionally, lying awake late at night, I couldn’t help wondering: if I hadn’t made those mistakes, if my mindset had been a bit better then, would my life now be very different? If I really had made it to Tsinghua or Peking, who would I be now? I know life has no “what if”, and I still couldn’t stop thinking about it. In the days before NAC I thought about this almost every night.

Over spring break I also interviewed with a company back in China on a whim and unexpectedly landed a fairly large package without much trouble (though I have no plans to take it). Still, it added a bit of confidence.

The night before we left, Jeff and I were in the lab messing with templates together.

And then we were off.

Days 0 to 3#

Activities scheduled from morning to night. Felt like boot camp.

I also saw a lot of friends from last year. The quant recruiters somehow still remembered me a year later. Lots of socializing, lots of great conversations.

There were some practice contests in between. Purely on ranking, we had a real shot at WF — assuming no disaster in the official contest.

During the OpenAI challenge I was working on a simulated-annealing parameter-tuning problem. Another problem was actually convex-hull-trick DP; my teammate wasn’t familiar with it, wrote out the DP recurrence, and sat there thinking about it without showing me in time. When he finally did I solved it instantly, but by then it was too late. A bit of a shame.

After that round his confidence took a hit and he kept saying he wouldn’t touch the keyboard on contest day. All I could do was reassure him: it’s fine, we’ll adjust the strategy properly on the day, and anything you think of or get stuck on you can hand straight to me.

After dinner we watched a bit of Tourist’s one-versus-four exhibition, then went to bed early.

Day 4#

Contest day.

Eight and a half hours of sleep. The last time I slept that well before a contest was probably high school — in college I routinely go into a contest or exam on four or five hours.

As for why I usually sleep so little, it definitely isn’t because I stay up messing around too long.

The strategy I set before the contest: early on, everyone thinks independently, and the moment you believe you have it, get on the machine and write, tracking the scoreboard as closely as possible. Mid-to-late contest, switch to one person on the machine while the other two work a problem together. For any solution you’re unsure about, explain it to someone else first to validate the idea. For problems I solved, anything I thought a teammate could implement got handed off so I could move on to new problems. And if a teammate hit an algorithmic wall — say, a DP recurrence they didn’t know how to optimize — they’d hand it straight to me.

At the start my two teammates each wrote a problem. Mine was 0/1 fractional programming (binary search on the answer) plus the Manhattan-to-Chebyshev distance transform. I asked my teammates and found they didn’t know that trick, so I wrote it myself. My hands were rusty but it still went in within about 20 minutes, putting us at 14th — already inside the WF qualifying zone. A decent start.

Then we got stuck for a while. The two of them worked one problem together while I worked another. An hour later they got theirs in, moving us to 9th. Meanwhile I still hadn’t cracked mine. A teammate started scanning other problems but had no immediate ideas, so I asked him to explain the one he was on. Listening to it, I realized brute force was enough — it just needed a computational geometry template. We had the template, but I wasn’t comfortable with it, so I handed it to Jeff, trusting his implementation. The other teammate and I worked a constructive problem together, solved it jointly, handed it to him to write, and I moved on. When Jeff got his in we were back to 12th, and then Yehor started on the constructive problem.

When Yehor’s submission passed, we looked at the scoreboard for the remaining feasible problems: a probability DP, a construction, and a circular DP. My head felt a bit foggy and I didn’t trust myself on construction. Jeff is good at math, so I gave him the probability DP and took the circular DP.

Step one was how to break the cycle. A moment’s thought showed that you can just cut at the maximum and handle the clockwise and counterclockwise directions separately. I wrote out the $n^2$ DP recurrence and first guessed it was a divide-and-conquer/monotonic-decision optimization — I happened to have read about a new toy the month before, a simplified LARSCH algorithm — so I started trying to prove the quadrangle inequality. Halfway through explaining it to a teammate I suddenly realized that only the two ends of the interval can contribute; simplifying the recurrence turned it into plain convex-hull-trick DP. I got ready to write immediately, then realized the transition index $k$ isn’t monotonic, so a queue-maintained hull won’t work — it needs a Li Chao tree. I had skimmed Li Chao trees the week before without writing any code, and the last time I actually coded one was at least five years ago. I calmed down, re-derived it on scratch paper, found it wasn’t hard, and decided I could write it.

At this point a teammate’s probability DP got a rejected submission. The scoreboard was already frozen; we were 12th going into the freeze, which is not at all safe, and the pressure was building. But they still chose to trust me: they printed their code and gave me the machine. I wrote the Li Chao tree DP in 20 minutes flat, not yet debugged. Then they said they’d found their bug, so I handed the machine back and their probability DP passed — seven problems, and it started to feel safe (after the contest it turned out to be not at all safe, which is a little terrifying).

They gave me the machine back. Debugging for a bit, I found I’d messed up the cycle-to-chain step — the Li Chao tree itself was fine. First submission: WA. Twenty minutes left. My heartbeat skipped and I got a little panicked. I immediately went back through the code line by line and within a minute found it: while debugging I’d set the Li Chao tree size to 10 and never changed it back. Fixed it, resubmitted — Accepted!

My teammates and I cheered on the spot. WF was almost certainly ours.

We spent the last dozen minutes on the remaining constructive problem, and actually solved it in the final minutes — with no time left to write it. Didn’t matter. WF was locked.

None of the three of us ate lunch during the contest. Afterwards I was too excited to eat much at all and had to force some food down.

My teammates were thrilled too.

Then came photos, conversations, the ceremony, and waiting for the scoreboard to unfreeze. Final result: 9th in North America, officially qualified for WF.

It didn’t feel quite real.

All these years of competition, and here it finally looked like a happy ending.

My parents got up at four or five in the morning to wait for my result. It felt like the world had finally been gentle with me.

And another of my own dreams came true.

This time I really was the one the team could count on, and I finally came through for myself.

Afterword#

The night we qualified for WF I stayed up all night celebrating. My teammate, meanwhile, was doing upsolving. Terrifying.

Thinking about it, Jeff really did give it everything this time. This was his last chance — he graduates from his master’s soon — which is why he worked so hard and took it so seriously.

I’m deeply grateful to my teammates for trusting me, and grateful to myself for not giving up.

In that final stretch, analyzed rationally, the constructive problem probably had a better chance of passing than mine did. But my teammates chose to trust me unconditionally anyway and just said “let him cook”. And I didn’t let them down.

I’m very lucky, and very grateful, to have found teammates this good.

So — see you at ICPC WF 2026 in Dubai!

Zhihu link: https://zhuanlan.zhihu.com/p/2021209593309660945 ↗