You made 35,587 picks in the Player Power Rankings this year. Back in July I said I’d review how well the algorithm had worked, so I’ve gone back through every one of them. This is what I found, what I’ve changed for 2027, and a few things your votes turned up along the way.
- Your votes were good. The way I was counting them meant the ranking more or less stopped listening after February.
- I’ve changed the method so that recent votes count the most, and put all of 2026 back through it.
- David Clifford is first either way. Even my toaster would have got that one right.
Start with a pair. Who was the better player this summer?
of his pairs won, from June on
of his pairs
The voting said one thing all summer and the ranking said the other, which is the whole problem in one pair. I’ll come back to why it happened.
The idea behind the Player Power Rankings was to keep it simple: show two players, ask who’s been better lately, and repeat. There are no marks out of ten and no top 20 to fill in, and because you don’t choose the pair you’re shown, nobody can pile votes into their own team.
It’s worth saying early that this is a ranking of opinion. It tells you who the people voting rate at the moment, which isn’t always the same thing as who the best footballers are.
The list started with 98 players in January and had grown to 173 by August. For most of them the argument never stopped. For one of them it never started.
Every ranking in this piece comes from the new method unless I say it’s the old one.
Nobody but David Clifford
David Clifford was number one on every single Friday of the season. Nobody else got there even once, and he’s the only player who stayed in the top ten all year.
One dot for each pair he appeared in, in the order they were voted on.
To put nine in ten in context, nobody else on the list managed more than six.
The gap between David Clifford and whoever is second turns out to be bigger than the gap between second and 60th.
Final ranking. Further right is better.
Second, as it happens, is Con O’Callaghan, and in all the times the two of them came up together Con wasn’t picked once. Michael Langan got the nod ahead of David Clifford more often than anyone else did, and even that was only about one time in five.
Share of the pairs each player won when he was shown with David Clifford. Craig Lennon is left out because the two were only paired once.
At one stage he was picked in 67 pairs in a row. The next best run by anyone was 26.
It held at home too. When he was paired with his brother Paudie, David was picked 94 times in 100.
Where the ranking got stuck
Back to that first pair. Odhran Murdock earned his place in the spring: from March to May he won 62% of his pairs while Ryan O’Donoghue won 37% of his, and both the old method and the new one had Murdock ahead of him in May, rightly.
Then the voting turned and the old ranking didn’t turn with it. From June on O’Donoghue was winning 62% of his pairs and Murdock 30%, and Murdock was still three places ahead of him when voting closed.
The one I flagged at the time was Darragh Beirne. He was added during the season and was in the top ten within five weeks, while his Mayo team-mate O’Donoghue sat 22nd. I said in July that players I added later seemed to do too well. Having gone through it all, that wasn’t quite right. Beirne won 70% of his pairs and deserved to be up there. The real problem was that O’Donoghue, and players like him, couldn’t move.
Two things were behind that.
1. Most of the voting happened early
Nearly half of all the picks, 44%, were made in the first seven weeks of the season. The last five weeks, which included the All-Ireland final, brought in just 3%.
Tap a bar for the week.
Share of the season’s 35,587 picks. Weeks end on a Friday.
2. The old method got too sure of itself
The old method kept a running score for every player, and the more votes it had on someone the surer it became about where he belonged. That sounds sensible, and for the first couple of months it was, but it had two side effects I didn’t spot until the summer.
The first is that each new vote moved a player a little less than the one before.
Put that together with how much of the voting happened early and you get this.
Tap a bar for the week.
Every rating change, up or down, for the players on the list from the start, replaying the old method one vote at a time.
Three-quarters of all the movement in the old ranking had already happened by the end of February, and the last five weeks of the season accounted for about 1% of it.
The second side effect is that it stopped asking about him. The old method also decided which pairs to show, and it went looking for the players it knew least about, which by July meant whoever had most recently been added.
Measured against David Clifford, who came up least. Over those five weeks a player added late on came up about three times as often as one who had been there from the start.
So that’s how Beirne ended up above O’Donoghue. Beirne arrived when the old method knew nothing about him, so every vote moved him a long way and it kept asking about him. O’Donoghue had months of votes behind him, so the summer ones barely moved him and he rarely came up at all.
It thought it knew your opinion on O’Donoghue, and it did. It knew what you thought of him in February.
That would be fine if opinions never changed, or circumstances. They do. Seán O’Shea was second in the ranking for nine Fridays running. Plagued by injury, he drops to 21st, and the old ranking still had him 9th in August.
Nine different players sat second at some point, and twelve held a place in the top three.
What I’ve changed
The votes are the same ones. What’s changed is how they’re counted.
The old ranking worked a bit like a league table for the whole year, where once a player has hundreds of results behind him, one more barely shifts him.
The new ranking works more like a form guide. Every vote is still in there, but the most recent ones count the most, so who you picked in July matters more than who you picked in February.
The old method is measured from the votes. The new one is drawn to show the idea rather than measured.
Both lines use exactly the same votes. Only the top 60 is published, so a rank below that is shown as outside the 60.
For what it’s worth, I don’t think Ryan O’Donoghue was ever outside the best 60 footballers in the country. What happened in the spring is that the voting turned against Mayo players generally, and he was losing roughly three in four of his pairs, most of them against players in the top 30. The old method never put him up against anyone further down, so the new one had to guess how far he’d fallen, and it guessed too far.
The new line also jumps about more than it would in a normal season, because late in the year there were so few votes that a handful could swing a player. From January a player who’s dropping gets paired with the players around his new position, which should catch both of those. If it doesn’t, there’s more for me to do.
Two other things are different.
- Everyone comes up about as often as everyone else. Half the pairs are between players who are close together in the ranking, because those are the ones worth asking about.
- A new player has to earn it. He starts in the middle of the pack and needs 60 pairs before he gets a number, where the old method gave him one after 30.
What the new ranking looks like
I’ve put the whole of 2026 back through the new method, Friday by Friday, and that’s the ranking on the site now.
Final ranking, 7 August. The arrows show places gained or lost against the old one.
The top two don’t move and eight of the old top ten are still in it. Further down, the new ranking follows the players whose year changed direction.
Ranks are for 10 April and the final ranking. Four of the eight biggest drops in the second half were Meath players.
“How can you have him ahead of him?”
Sometimes I can’t, at least not with any confidence, and the ranking now says so. Every player gets a rank and a likely range, and the range is where the method would put him 8 times out of 10.
Bars run from 1st to 40th.
Overlapping bars don’t make it a toss-up, though. They mean the ranking is less than certain, and it can put a number on how certain.
This is the method’s own confidence, worked out from every pair either player was in. It is not their head-to-head votes.
You don’t all agree with each other either. Even when there were more than 40 places between two players, the lower-ranked one was still picked more than a third of the time.
Each pair is set against the ranking as it stood the Friday before. The mark is 50%, a coin toss.
And you don’t always agree with yourselves. When the same person was shown the same pair a second time, about one time in five they picked the other man.
Is my new ranking any better?
Here’s one test. Take the ranking as it stood each Friday and check it against the following week’s votes: when two players were paired, how often did the one ranked higher get picked?
Pairs where both players had a rank under both methods. Each Friday’s ranking uses only the votes cast before it.
From June on, the old ranking got that right 49% of the time, which is what you’d get from tossing a coin. The new one gets 55.5%.
Neither is ever going to get near 100, because the pairs are deliberately selected to be difficult. There isn’t much point asking you to choose between David Clifford and the 150th man. Granted, evidently there isn’t much point in asking about David Clifford full stop.
A few other things your votes showed
The team you picked
This is the best player in each line of the field in the final ranking, with the chance that each one really belongs in the fifteen. Kerry have five of them.
Thirteen counties in the 60
Kerry have the most players in the 60, while Louth have the most in the top 20.
Louth were the people’s team
Whenever a Louth player came up he was picked 57% of the time, which is the best record of any county.
Counties with enough pairs to judge. David Clifford’s pairs are left out, or Kerry would walk it.
You like a forward
A typical full-forward on the list gets picked over a typical full-back 58 times in 100, and seven of the top 15 are full-forwards while none are full-backs. The votes can’t tell me whether that’s a bias or whether the forwards on the list are just better.
A typical player in each line against a typical full-back, treating every line on the list as equally strong.
Six of the top 20 weren’t on the list at the start
They were added as the season went on, as were 21 of the 60. In all, 134 different players spent at least one Friday in the 60.
“Can’t split them”
You passed on 31% of the pairs you were shown. These are the players you were most willing to make a call on.
The pair you were surest about was Paudie Clifford and Seán O’Shea, passed on 14% of the time. The least certain pair was passed on more than half the time.
You vote in bed
A third of all picks were made between 9pm and midnight. Thursday and Monday were the busiest days and Saturday the quietest, and there were only four days all season when nobody voted at all.
The wildest week
The biggest one-week moves both came in the first few weeks, while the ranking was still finding its feet.
And two more
Paddy Small was picked in exactly half of his pairs, which makes him the most evenly split player on the list. And the pair you were asked most often was Con O’Callaghan and Joe O’Connor, which Con took 58 to 42.
Your turn
Here are five pairs of players who finished close together in the ranking. Pick one from each and I’ll show you how everyone else voted on that pair across the whole 2026 season. These don’t count as votes, which reopen in January.
The share who agreed with each pick comes from every vote on that pair across the 2026 season. “The same five” multiplies those together, so treat it as a rough figure.
Voting is closed until January and the 2026 ranking is final. When it opens again, the ranking should keep up with you a lot better than it did this year.
See the final 2026 rankingThanks to everyone who voted, and to AI for helping me tidy up the writing.