Editorial Intelligence vs FPL, part one: what counts as enough evidence to change your mind?

Part of Editorial Intelligence vs FPL, a running experiment in evidence, judgement, football and occasionally making terrible decisions.

I’ve decided to spend this Fantasy Premier League season running a slightly ridiculous experiment.

My team is called Editorial xG.

Rather than just picking players, forgetting about it for three weeks and then wondering why everyone else has twice as many points as me, I’m going to use FPL as a small test of Editorial Intelligence.

Not because fantasy football is somehow an editorial problem.

Because the underlying problem feels very familiar.

There is far more information available than anyone can sensibly use.

Player prices. Fixtures. Injuries. Pre-season minutes. Goals. Assists. Expected goals. Predicted line-ups. Managers talking at press conferences. Fantasy experts. Reddit. YouTube. Facebook posts telling me a 17-year-old Aston Villa striker is about to become essential.

The difficult part isn’t finding information.

It’s deciding what deserves to change your mind.

That’s the bit I’m interested in.

And there is a second reason for doing this that may sound even more excessive: I want a low-stakes way to watch how AI thinks alongside me.

Some people will reasonably look at the amount of analysis going into a fantasy football team and think this is a spectacular waste of time. That’s partly why it works. The stakes are low enough that I can deliberately slow decisions down, see what ChatGPT notices, bring in another model to disagree, compare that with my own instinct, and then watch what reality does to all of us a few days later.

I’m not trying to outsource the team to AI. I’m trying to understand what happens when model judgement and my own judgement coexist: when the model catches something I missed, when I resist it, when it sounds persuasive but weights the evidence badly, and when it stops me reacting too quickly.

The experiment

I’m trying to make my FPL decisions through a simple loop:

Information → signals → interpretation → decision → outcome → learning

And crucially, I’m recording some of those decisions before I know what happens.

Because otherwise hindsight makes everything look obvious.

If I captain someone and they score twice, it is very easy afterwards to explain why it was clearly the correct choice.

If they blank, it is equally easy to convince myself that I should have known better.

Neither is particularly useful.

The better question is:

Was it a sensible decision given the evidence available at the time?

That is much closer to how I think Editorial Intelligence works elsewhere.

You rarely have perfect information. You have competing signals. Different sources have different levels of reliability. You have to interpret what they mean, make a call, and then learn from what happens next.

FPL compresses that whole cycle into a week.

There is another thing I want to watch as the season starts properly: how quickly signals become stories.

Pre-season gives you hypotheses — Haaland is the safe premium anchor, Arsenal’s defence is strong, Manchester United have good opening fixtures, João Pedro could be value. Competitive matches start turning those hypotheses into narratives about players and teams. Some will strengthen. Some will collapse. Some will become compelling far too quickly after one result.

That matters because narrative is not just how you tell the story afterwards. It is one of the ways humans organise evidence while making decisions. I want to see whether AI helps me test those emerging stories, or merely makes them sound more convincing.

Building the first team

My first draft was deliberately fairly conservative.

Haaland as the expensive anchor.

Bruno Fernandes and Mbeumo because Manchester United start with Hull and Ipswich.

Gabriel because Arsenal have one of the strongest defences and open against Coventry.

João Pedro because Chelsea’s attack looked increasingly interesting.

A few cheaper players where I thought there was potential value.

And £1m left in the bank rather than spending every penny simply because it was available.

Then pre-season started producing signals.

João Pedro kept scoring.

Mbeumo started looking increasingly interesting because he was being used centrally.

Harry Maguire looked more secure for minutes, even while Manchester United’s defence looked shakier.

Brian Madjo, Aston Villa’s 17-year-old striker, appeared on my radar after scoring repeatedly in pre-season and then against PSG.

That was a useful little example in itself.

A Facebook post was the signal.

It wasn’t the evidence.

The next step was checking whether he was actually playing, who was ahead of him, whether Villa appeared to consider him a first-team player, and whether those performances were likely to translate into Premier League minutes.

He’s still on the watchlist.

That is different from immediately putting him in the team because he scored a few goals.

Then Arsenal beat Manchester City 3–0

This was the first time the experiment became genuinely interesting.

My immediate instinct was:

Get rid of Haaland.

He costs £15.5m.

City looked undercooked.

Arsenal were comfortably better.

If I’m spending that much of the budget on one player, do I really want him leading an attack that looks like it might need time to settle?

That was my gut reaction.

And I still think it was a reasonable interpretation of what I saw.

But it wasn’t the only interpretation.

The counterargument was that one match — even a competitive one — is not enough evidence to overturn several seasons of Haaland being an extraordinary early-season FPL asset.

He had also returned relatively late. City still start at home against Bournemouth. And selling a £15.5m player changes the entire structure of the squad.

So this became a much better decision problem than:

Haaland played badly. Sell Haaland.

It became:

Has the evidence crossed the threshold required to abandon the Haaland strategy?

My answer, narrowly, was no.

For now.

But my confidence dropped.

That distinction matters.

What did change?

Arsenal’s performance strengthened two signals.

The first was defensive.

Gabriel stays.

The second was attacking.

Christos Tzolis had already been interesting, but two assists against City, combined with Arsenal’s opening fixture against Coventry, made him more attractive.

So rather than reacting to City’s defeat by dismantling the entire squad, I made one targeted change:

Pascal Groß → Christos Tzolis

That used the £1m I had deliberately kept back.

Haaland remains.

But he is currently vice-captain rather than captain.

Bruno Fernandes has the armband against Hull.

Again, this could look brilliant next week.

Or ridiculous.

That isn’t really the point.

What matters is that the reasoning exists before the result.

This is where Editorial Intelligence gets interesting

The obvious fantasy-football version of this experiment is:

Can AI help me build a better FPL team?

Maybe.

But that isn’t the question I find particularly interesting.

AI can gather fixture information. It can compare players. It can surface injury news. It can find expert opinion. It can help me see patterns I might otherwise miss.

But eventually someone still has to decide:

Which evidence matters?

Which signal is noise?

When should consensus influence me?

When should I disagree with it?

How much evidence is enough to reverse an existing decision?

And:

How do I avoid confusing a good outcome with good judgement?

Fantasy football gives me a wonderfully low-stakes environment to experiment with all of that.

If I get it badly wrong, nobody loses their job.

I just lose my mini-league.

Editorial xG — current team

This is where I have landed ahead of Gameweek 1.

Goalkeepers

Kinský
Dubravka

Defenders

Gabriel
Maguire
Mitchell
Bobby Thomas
John Egan

Midfielders

Bruno Fernandes (C)
Mbeumo
Tzolis
Le Fée
Regan Slater

Forwards

Haaland (VC)
João Pedro
Igor Jesus

Current starting shape: 3-4-3.

I am not calling it final yet.

There is still time for injuries, team news and another piece of evidence to change the judgement before the deadline.

That is part of the experiment too.

What happens next

I’m going to keep Editorial Intelligence vs FPL running through the season.

Not every week.

Not as fantasy football tips.

Only when the game gives me an interesting problem involving evidence, uncertainty, consensus, bias, narrative or judgement.

The useful sequence is:

What I thought → what evidence I used → what story I was starting to believe → what I decided → what happened → what I learned

The first result is already useful because it exposed a disagreement between instinct and process.

My gut said sell Haaland.

The evidence did not quite clear that threshold.

Now the decision is recorded.

Next comes the uncomfortable bit: waiting to see what happens.

Editorial Intelligence vs FPL: 0–0.