Every 12-Over-5 Number You Have Ever Heard Was Measured on a Bracket That Stops Existing in 2027.

Every March someone tells you how often a 12-seed beats a 5-seed. It is the most-quoted number in American sport, it gets repeated to the decimal point, and it is about to become quietly wrong.

Not because anybody lied. Because the thing it describes is being replaced.

What Expansion Does to a 12-Seed

There are 31 Division I conferences, so 31 automatic bids. That means today’s 68-team field carries 37 at-large bids; the 2027 field of 76 carries 45. Eight more.

Those eight are the best teams currently being left out — bubble teams, not champions of small leagues. They do not land on the 1-line. They land in the middle, and everyone below them shifts down. The weakest automatic qualifiers get pushed toward the opening round, and the 11-, 12- and 13-seed lines fill up with better basketball teams than they hold now.

Simulating 20,000 brackets, that is exactly what happens. The average 5-seed does not move at all. The average 12-seed gets measurably stronger — and the gap between them narrows by about a third.

Which moves the upset rate from roughly 33% to 38%. About a five-point rise, or one extra 12-over-5 upset every five tournaments.

Now the Uncomfortable Part

The quoted rate rests on less evidence than its precision suggests. There are four 12-versus-5 games per tournament. Across the 64-team era — 1985 to 2025, with 2020 cancelled — that is 40 tournaments and 160 games. A 33% rate measured on 160 games carries a 95% interval of roughly ±7 points.

So the predicted shift is smaller than the uncertainty in the number it is shifting.

Ask how long it would take to confirm a five-point change by watching. Four games a year. To detect a shift that size with ordinary statistical confidence takes about 1,200 games — roughly 300 tournaments.

Why This Is Worse Than Being Wrong

A wrong number that can be checked gets corrected. Somebody runs the data, the estimate moves, everyone updates.

This one cannot be checked on any human schedule. The tournament produces four of these games a year, and no amount of caring changes that. The rate will be quoted every March, it will feel confirmed every time a 12-seed wins, and the confirmation will be meaningless — because at four games a year, a 33% world and a 38% world look identical for centuries.

That is the trap: the evidence arrives too slowly to correct the belief. Not a hard problem, an impossible one, and impossible in a way that is invisible from the inside.

What To Take Home

The useful habit is to ask two questions of any long-running statistic, not one.

The first is the familiar one: how was it measured? The second is the one almost nobody asks: has the thing it measures changed since? A rate calculated honestly on a system that no longer exists is not evidence about the system that replaced it, however many years went into it. The CFP went from four teams to twelve. The Preakness moves off two weeks’ rest next year. March Madness goes to 76. Every one of those quietly retired a shelf of numbers, and the numbers went on being quoted.

And when you cannot check — when the data arrives four games at a time — the honest move is not to pick a number and defend it. It is to say plainly how much you could possibly know, and then hold the belief that loosely.

In March somebody will tell you a 12-seed wins a third of the time. The right answer is: it used to, probably, give or take seven points, in a tournament that is being discontinued.