Why a Perfect NCAA Bracket is (Almost) Impossible — and how 1 Trillion tries change the math

The core math

If each game were a fair coin flip, the number of possible tournament outcomes is 2⁶³ ≈ 9.22×10¹⁸. Even if you could generate a million distinct brackets per second, you’d need centuries to enumerate all outcomes and 74 million terabytes of storage.

Reality isn’t 50/50. Teams have unequal strengths and matchups matter. This lowers the effective search space, but perfection remains very difficult.

1TB has estimated that, across the entire NCAA mens tournament history, there have been between 1-2 billion brackets created- lets call it 2B. This means that 1TB has made 500X more brackets than have ever been created, but this still only covers a tiny portion of the possible outcomes.

Why I think it could happen

With informed priors and targeted sampling, the bracket tree becomes tractable enough to cover meaningfully:

• Stronger inputs: Metrics for team strength, injuries, matchup styles, travel, rest, and tempo can yield calibrated game probabilities—sometimes near 99% and often in the 55–75% range.
• Effective shrinkage of the space: Applying those probabilities across 63 games compresses the truly plausible outcomes from >9 quintillion to a far smaller—though still enormous—set that heavy compute can explore.
• Scale + search: With massive scale and proprietary models, onetrillionbrackets has, retroactively, constructed a perfect bracket for the 2025 NCAA tournament using data that was only available before the tournament began. However, attempts at the previous 9 tournaments did not yield the same success. Based on that evidence, I estimate a ~10% chance of achieving a perfect bracket this year (though this estimate is very rough).

What can lead to success?

A bracket is considered “chalk” when only favorites win. For a sense of a chalk-leaning bracket, think of many of Barack Obama’s public brackets—favorites forward, limited early chaos. If this year trends chalky, potentially due to increased influence of NIL money, I believe I can land a perfect bracket.

But that is not the expectation. For example, as recently as 2023, when the last four teams were a 9-seed, two 5-seeds, and a 4-seed, you would have needed quadrillions of brackets to reliably cover those paths. If a single 16-seed wins their first game, as in 2023 and 2019, then 99% of brackets are immediately eliminated; I would instantly only have 10 billion brackets remaining. In contrast, last year required “only” hundreds of billions.

With a total number that sits between those regimes, I have a shot.

Back-of-the-envelope intuition

Suppose your per-game accuracy against the true outcome were 64.6% on average. The chance of hitting all 63 games is 0.646⁶³ ≈ 1×10⁻¹². With one trillion independent shots, your expected perfect hits would be about 10¹² × 1×10⁻¹² ≈ 1. But—this is the catch—brackets aren’t independent, per-game accuracy isn’t constant, and downstream correlations reduce your effective independence.