College basketball upsets
and seed records.

A seed line is a prediction. This page is what happened to those predictions across 41 tournaments of the 64-team era, counted out of the archive the game itself plays from — 1,311 opening-round games, every one of them a real result.

On this page

What a seed line is actually worth

In the opening round of a 64-team bracket the seeds are paired so they add up to 17: 1 against 16, 2 against 15, and so on down to 8 against 9. Those matchups let us compare results across seed lines. Here is every seed line's record in that round from 1985 to 2026.

How these records were counted

These are calculations from the Cut the Nets historical game records, not simulated odds. Each completed round-of-64 game contributes one win and one loss; preliminary play-in games and later rounds are excluded. The cancelled 2020 tournament contributes no games. The 2021 Oregon–VCU matchup was a no-contest, so seeds 7 and 10 each have 163 decisions while the other seed lines have 164. Percentages are rounded to one decimal place.

Historical context: the official tournament record book documents the no-contest. Our archive totals include results through the completed 2026 tournament and retain on-court results even when later vacated.

Three things fall out of that table immediately.

The top of the bracket is nearly safe, but not safe. A 1 seed has lost its opener 2 times in 164 tries. That is a 98.8% win rate — overwhelming, and still not certain, which is the entire reason people fill in brackets.

The 8–9 game is a coin flip, and the 9 is winning it. The 9 seed is 87–77, a 53.0% win rate, marginally better than the 8 seed it faces. The committee puts a team on the 8 line rather than the 9 line and the games do not notice.

The 12 seed is not magic, but it is real. The famous 12-over-5 upset happens at a 35.4% clip — 58 wins in 164 games. That is much more often than the 13, 14, 15 and 16 lines, and it is why one specific number has its own folklore.

The two that should not have happened

A 16 seed has beaten a 1 seed exactly 2 times in 164 attempts.

• 2023 — Fairleigh Dickinson beat Purdue, 63–58.
• 2018 — UMBC beat Virginia, 74–54.

Both came in the last decade, after thirty-odd years of the line going winless. A 1.2% win rate over four decades is the kind of number that looks like zero right up until it isn't.

The biggest seed gaps in the archive

Ranked by the gap between the winner's seed and the loser's. A gap of 15 is the largest the bracket structurally allows.

Why any of this matters to a simulation

An upset is only interesting if it was unlikely and still possible. A model that never lets a 16 beat a 1 is wrong, because it has happened twice. A model that lets it happen one time in five is also wrong, and much worse, because it makes the bracket meaningless.

Cut the Nets plays every game out possession by possession from the ratings of the players actually on those rosters, so a simulated upset reflects the model’s ratings, decisions and randomness. A heavy favourite usually wins. Sometimes it doesn't, and when it doesn't there is a box score explaining exactly how. If you want the mechanics, how the simulation works goes through them.

The point of a seed is that it is usually right. The point of the tournament is that "usually" is not "always".

One caution about the numbers above: they are the historical record, not the game's odds. The game does not look up what a 12 seed has done historically and copy it. Any win rate you see inside Cut the Nets comes out of the simulation itself.