CupThe Guardtower(Nov 1)Austin Charleswon withDragapult
ChallengeMidwest Cards Columbus(Oct 27)Brian Daowon withAlakazam Dudunsparce
ChallengeTabletop Cafe(Oct 23)Armando Martinezwon withDragapult
CupFive6Gaming(Oct 22)Mitchell Courtneywon withMega Lucario
ChallengeThe Guardtower(Oct 20)Tyler Pricewon withMega Lucario
ChallengeLittleroot Games(Oct 19)Anthony Elliswon withRocket's Honchkrow
ChallengeThe Warp Gate(Oct 18)Tyler Pricewon withOgerpon Meganium Hydrapple
CupThe Warp Gate(Oct 18)Anthony Elliswon withRocket's Honchkrow
ChallengeBeyond The Board(Oct 17)Anthony Elliswon withRocket's Honchkrow
ChallengeThe Collector's Quarters(Oct 17)Potatowon withMega Darkrai
CupRound Table Games(Oct 17)Tyler Pricewon withSlowking
ChallengeColumbus Card Shop(Oct 16)Ethan Deiswon withDragapult
ChallengeFive6Gaming(Oct 15)Gabe Nephewwon withDhelmise
ChallengeRound Table Games(Oct 8)Troy Campbellwon withSlowking
ChallengeCard Father Games(Oct 7)Tyler Pricewon withDragapult
Night March
Night March
← Back to articles

Using LLMs to ID into top cut

not sponsored

Anthony Ellis
November 3, 2026

Yes, you CAN calculate resistance. Manually.

now hear me out. some of you might immediately click off this article or go "my AR15 says you lose" when I mention using the AI. Some might feel like it's cheating. I really don't know.

But we all know the feeling. Opening up the Swiss calculator on your phone and having a 15 minute discussion going into round 5 about who is gonna make it. And honestly that's part of the experience.

But honestly I like thinking less about that and more about "yeah shoot I really messed up in round 4, maybe I do this in round 5." Dying in tourneys because I let my nerves or excitement get the best of me is a... common story. Still working on it. But if there's anything that gets me thinking about "please I don't wanna play anymore and go to wendys" it's praying for the top cut ID chances.

TOM does sometimes feel like a mystery. though I think it's been documented enough online for an AI to understand how it works in how it does pairings. Or at least I hope so. I'm on lunch break right now writing this so I really don't know. Case in point, you can do the following prompt:

Prompts

I am in a Pokemon cup. It's 5 rounds. I just won round 4. My name is NAME. Going into round 5, should I ID? I've attached all the pairings as pictures to this message so you can calculate resistance. I'll give you more data at the end of round 4 k. Give me an action plan.

Or maybe

I am in a Pokemon cup. It's 5 rounds. I just lost round 4 and now my record is 2-2. My name is NAME. Is it possible I still can get into cut if I win my next round? give me the scenarios. I've attached all the pairings as pictures to this message for you to calculate resistance.

In the above of course, it requires you take a picture of the pairings sheet every round. Doesn't have to be perfect or even the entire sheet (it can figure out partial screenshots too). Yes it can read screenshots. Yes it can do so fast.

Effect

You might get output based like the below. Sometimes it takes a bit more prompts. I am not a prompt engineer.

But here you can see some important things

  • It denotes strategy on IDing based on who else IDs
  • It denotes what matches in round 4 to watch
  • It tells you to do or not do based on your resistance.

Particularly, I actually like point number 2. Even though it could feel like "oh yeah this is totally gaming the system" this is adding a new level of depth at tournaments we might not always get to.

Who to cheer on.

Your resistance is always in the hands of your opponents. Your opponents always have been people you have to...uhhh, beat. Ruin their lives. Make them cry and 0-2 drop. Send them to the streets. But now there's a secondary reason to watch them play and hope for their success. And specifically, who and how their success will affect yours.

Also, for trying to sneek into cut, especially with some odd numbers and downpairings, you might know easily "will I have a chance as 3-2? Who needs to win to get me a spot?"

What model to use?

Not chatgpt. can't upload enough images unless you're paying the money Claude is cool but probably same image upload issue. or it's just tooooo slow. speed is everyhing

I've been using gemini due to a funny free student discount. But it's also really good at being fast but "smart enough" for this use case (now. months ago it was turrrrrbo lame). I think the free tier also can do it too, due to less image upload constraints.

Warnings

use this less as the whole truth and an exact precise knowledge of exactly what to do. Instead, it's a better estimator than what we have now. And something to give you some peace of mind Don't think it's probably going to get right the resistance between 50.4% and 50.3%. Ultimately your gamble is your choice.

reminder every round you need to snap a pic of pairings for this to work.

but trust me anyways i do a lot of gambling right now every time i pick up birds i am a gambling PHD

Comments (0)

  • Be the first to comment.

Add a comment