HU ZOOM TOTAL KO POKERSTARS

Hi!
I played these games from late 2019 until 2026 when stars got closed for my country.

You are right that this is an unique format where our usual simulations of cash game/mtt run doesn't quite apply, however we can extrapolate and make some assesments of what a good and bad run looks like. Your's is what a bad run looks like (3% under AIEV) and you can look up what the opposite looks like by looking at the sharkscope leaderboards for this format.

When flipping a coin 11k times you are way more likely to be over/under ev than exactly at your ev.

Another factor to consider is that these games have shrunk in guarantees and field sizes by over 70% since 2020 and nowadays it's basically a reg battle with very few fish, and your stdDev will be higher than 150. In 2019-2020 it was glorious, so many fishes and shitty passive regs you could just print. Nowadays it's becoming a multi table HU sng just played at deep stacks, there are some gifted players who will emerge with a higher winrate than others, for some the swings are gonna be mental.

I had 11 month stretch where I ran 11k bb under ev in this format and it was the worst shit ever, in reality it feels (and is) WAY worse cause you are losing your $EV, chip EV and future EV (think losing a flip and being out vs winning a flip and being in the tournament with a chiplead and position to get more KOs). Still had deepruns every day, probably finished top 7 about 100 times with nothing to show for it.

Statistics doesn't look anomalous to me, sample size is not that singificant. Focus on improving your strategy, perhaps you need to go all in more? 😋

DM me if you want more secrets revealed

I didn’t play during 2019–2020; I entered this format in 2023. However, I still consider the field to be very soft. In 2023–2024 my winrate was around $30 per tournament, while running about 1.5% under AIEV, so it’s hard to say variance was on my side 😀 In 2025 my winrate dropped to around $6 per tournament with roughly 3% under AIEV.

Overall, you’re using the coin-flip analogy and arguing that being above or below EV is more likely than landing exactly on EV. When I suggested calculating this more precisely using AI, the point was that the probability of such an outcome is around 1 in 4–5 billion. For comparison, the world’s population is only about 8–9 billion. That makes this outcome practically impossible in real life. Wouldn’t that be considered anomalous statistics?

I’ve been running over 11k below EV for 13 months straight across 210, 000 hands, on top of the anomalous deviation in races.

Regarding my strategy: I track statistics where the opponent’s stack is converted into dollar value and use a calculator built around that. I understand that in this format EVBB is less important than knowing where to push. Even though I often push at the expense of EVBB, my overall EVBB is still strong, which suggests my postflop play is solid as well. Overall, races are an extremely important metric, and any deviation from expected outcomes - especially one with a probability of 1 in 4–5 billion - has a massive impact on $EV.

It's too small of a sample, with 10 year mtt experience you should know that. And 6$ per tournament in 2026 is still more than most mtt regs will make.

It can happen and it feels worse than words can explain

I'm curious to know what AI suggested you the odds of this, but running a couple simple sims in primedope both for tournaments and HU cash games we can see that a 13BB/100 player after 200k hands can be in the red.... or up heaps and feel like a rockstar

Bro, i didn’t focus on a large deviation in EVbb at all. Please reread the post.

What I specifically described is an anomalous deviation in the Races metric. I explained how I identified this using AI-based analysis. I am not claiming that the AI calculations are 100% correct, but the result looks unusual, which is exactly why I shared it to hear opinions from people who are knowledgeable in statistics and probability theory.

Possible reasons in my opinion:

1.Problem with HM3 data collection
2.Incorrect calculations in Races (I manually checked 20 hands - all ok)
3.The AI miscalculated the probability
4. Just unlucky
5. Problem with the RNG

I used chatGPT to try and recreate your experiment, it calculated that odds of being 300 flips down after 10k is roughly 1:950 million when standard deviation is 50

It then offered me to translate this into poker winrates and I gave it a bunch more information to work with, and it's answers were generally reasonable in my opinion. Make it calculate your downswing for a game with a stdDev of 150 and your downwing is not unusual at all.

You have multiple factors that increase the variance at play : heads up, tournaments, total KO structure

It's unlucky and psychologically brutal, better not to think about it 😂

I understand what you mean about variance in MTTs and Total KO tournaments - indeed, taking all factors into account (stack sizes, bounty structure, multiway pots), the effective standard deviation is much higher.

However, all these factors do not affect the Races metric at all. Here we are only looking at 1v1 all-ins, where the cards are known and the exact equity of each hand is also known. In this case, each all-in is a pure binomial event: you either win or lose the hand. The standard deviation here is strictly determined by the binomial formula:

\sigma = \sqrt{\frac{p(1-p)}{N}}

where p is the equity and N is the number of all-ins. There are no other sources of variance — σ cannot change.

So when I calculate −6.3σ for my 11,154 1v1 all-ins, this is mathematically correct for this metric. The “large standard deviation” you are referring to applies to full tournament results, where many other factors influence the variance. But for pure 1v1 all-ins with known equity, that reasoning does not apply.

P.S. I used ChatGPT a bit in my reply to express my point more accurately and factually, and to explain why my calculations are more likely to be correct.

Okay so if your run over 200k sample is 1:4.5Billion unlucky ...
and my run over 200k sample is 1:4.5Billion unlucky ...

Nice to meet you, we might be the only ones on this planet 😃

Hopefully someone else chimes in but id guess less than 100 ppl have 200k hands in this format

So, I started digging deeper into my HM3 database and found an error in the calculations. HM3 counts split pots as wins, which is exactly why those anomalous values appeared.

So it’s not that scary after all. There is some underperformance, but it’s well within standard variance:
2.7σ ≈ 1 in 150–200.

Overall, this metric is actually very good for identifying downswings. And ChatGPT’s calculations using the binomial distribution are absolutely correct.
The only important thing is to apply the following filter to races metric:
NetBigBlindsWonOrLost ≥ 1.00

This excludes spots where HM3 makes counting errors.

So the topic is closed. Mikus, thanks for the feedback - thanks to you I dug much deeper into the database. Good luck!