A year out is ordinary in boxing. Checked on 75,168 corner rows.
Six months while the next bout is made, a year while it is negotiated. In boxing that is not an accident, it is the calendar. Which makes the old line about rust more pressing in this sport than in any other, not less.
The population is corner rows from 1970 on, where the fighter himself had at least 5 previous bouts behind his rating: 75,168 of them. A corner row is one side of one bout, so a bout makes two; it is not a bout count. Draws (1,687 rows) and the 4 rows where the two ratings match exactly are left out.
Split by the gap since the previous bout, and take the win rate as it comes.
Counts are corner rows, not bouts.
86.5% down to 32.5% -- 54.0 points. It looks like a headline.
It is not one. Change the population and the same figure reads anywhere from 32.5% to 62.8%: 30.3 points of movement across the 6 combinations of year floor and minimum experience. So it is discarded here. A number that holds under one choice of population is not a finding.
Look at who the layers-off are fighting. Below is the average of the fighter's own rating minus his opponent's.
+444.2 when the gap is short. -92.5 when it is long, and the sign has changed. A man coming back after two years is, at that moment, an average of 92.5 points below the man across from him. He is not losing because he rested; he is losing because he is the underdog. That explanation works first, and most of it holds.
So compare like with like. Take only the side the rating favoured going in, and split that side by whether it was coming off a lay-off.
Counts are corner rows, not bouts.
92.2% (30,757 rows) to 75.0% (716 rows). The 54.0-point fall becomes 17.2. Most of the raw number was the matchmaking. Something is still there.
Where does what is left show up? Split by rating difference into five groups, and look at the lay-off bands inside each.
Each cell is the win rate, with its corner rows in brackets.
All five groups fall from left to right. Even in the above +200 group -- the side nobody would call anything but the favourite -- it runs 94.1% (28,081 rows) down to 81.2% (623 rows).
"A good enough fighter wins anyway" does not hold, at least not in this data.
So far this is win rates. There is another view -- compare the probability this site published before the bout against what happened. If the prediction is right, the rating already knows about the rest.
+6.6 to -10.1. To check that this is not simply how much any two slices of this corpus differ, the same rows were split in two at random, on something with no connection to lay-offs: +3.6 (37,429 rows) against +3.4 (37,739 rows), 0.20 points apart. The lay-off bands are 16.7 points apart, about 83 times as far.
The gap from prediction, by how confident that prediction was.
Split instead by how confident the prediction was, and the sign turns over in all five groups.
Read that control again. Both halves came out at +3.5. The predictor is a little under the outcome everywhere, with nothing to do with lay-offs.
So -10.1 is not a number to compare against zero. Against the level the whole corpus sits at, it is -13.6, and the short band is +3.1. Subtract the model's own offset before reading any residual. Skip that and the short band's "better than predicted" looks larger than it is.
They are easy to confuse, so: a win rate and a gap-from-prediction are different quantities.
The first asks how often a man won, and to hold the opposition steady it had to be narrowed to the favoured side. The second already has both ratings inside it, so every row can be used. One flattens where the other keeps falling -- not a contradiction; they count different things.
The raw 32.5% cannot stand on its own. Across the 6 variants it runs 32.5% to 62.8%. What holds still is the favoured side (75.0%-78.2%) and the residual (-9.6 to -10.1).
The favoured side of the longest band is 716 corner rows and no more. Not thin enough to move the leading digit, not thick enough to take the decimal seriously.
And none of this earns the word rust. Age is not measured here, nor injury, nor a change of division. A fighter who sits out gets a year older while he does it, often because he was hurt, and sometimes he comes back at a different weight. What is above is the sum of those, not a breakdown of them.
One last thing, about using it. When you look at a long-absent boxer's rating on this site, read it as flattering. Over the range measured here, the prediction sat 10.1 points above the outcome in the longest band. Discounting the displayed number by about that much is the safer way to read it.
The same question, measured in another sport
Figures in this article are as of 24 September 2026.