There are over 9 quintillion possible combinations in a 63-game tournament. If you generated a billion brackets per second, you'd need 300 million years to cover them all.
The best anyone's done is 49 games correct in a row before busting, back in 2019.
The approachWhy a trillion might be enough
The trick is that not all brackets are equally likely. A 1-seed losing to a 16-seed happens about 1% of the time, while the 8 vs. 9 matchup is more like 50/50. If you weight your brackets toward realistic outcomes, you can cover a lot more of the probable space.
Intuitive math: If your model averages 64.6% accuracy per game, then 0.646⁶³ ≈ one-in-a-trillion. Generate a trillion brackets with 64.6% accuracy, and you'd expect roughly one perfect hit.
I use KenPom ratings, first-round betting odds, and a few other basic factors like strength on the road and injuries to generate brackets that lean toward plausible outcomes. To understand what these probabilities imply, see my statistics page.
Proof it works (sometimes)I already hit one
Running this same model (SAMSIM) against past tournaments, I was able to retroactively generate a perfect bracket for the 2025 NCAA Tournament from under 400 billion attempts. That year was unusually chalk, where all four 1-seeds made the Final Four, which helped significantly.
Other years weren't so kind. In 2021, I couldn't crack 44 consecutive correct games even with a trillion tries; a neuroseurgon from Columbus did better than that in 2019.
I need a boring tournament
My best shot is a "chalk" year, where favorites mostly win. Like a bracket made by Barack Obama, I need the seeds to be indicative of who wins. If NIL money keeps concentrating talent at top programs, we might see more of those.
But March is March. In 2023, the Final Four had a 9-seed, two 5-seeds, and a 4-seed. You'd need quadrillions of brackets to reliably cover paths like that. If a 16-seed wins their opener (happened in 2018 and 2023), 99% of my brackets are instantly dead.
Based on testing against 10 past tournaments, I hit perfection just once. So call it 10% odds in any given year, and when looking at my performance those same years, 50% odds that at least one of my brackets ends up being the best ever submitted publicly.
Given the small sample size of just 10 years of testing, it's hard to say what is likely to happen. So I will invoke a meme.
