Editorial Intelligence vs FPL, part two: the decision before the result

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

My Fantasy Premier League team is more or less frozen.

And before a single Gameweek 1 point is scored, I want the decision on the record.

There is a reason for doing that beyond Fantasy Football.

I’m not really testing whether ChatGPT can “pick my team”. I’m using the team to watch how model reasoning coexists with my own judgement. ChatGPT can make a case. Claude can attack it. Experts and other managers can supply different signals. I still have to decide what I believe, what I distrust and when the evidence is strong enough to act.

That may sound like far too much work for FPL. That’s fine. The low stakes are exactly what make it useful: I can inspect the reasoning, disagreement and mistakes without pretending the outcome matters more than it does.

This article is therefore a snapshot before the proper season starts supplying the next kind of evidence. At this point most of the stories are still hypotheses: Haaland as the premium anchor, United’s opening fixtures as an opportunity, Arsenal as the defensive team to own, João Pedro as Chelsea value, Tzolis as the deliberate upside pick. Once competitive matches begin, some of those stories will strengthen and some will collapse. I want the starting version preserved so I can see how much the narrative changes after reality arrives.

In part one, I wrote about what happened after Arsenal beat Manchester City 3–0 in the Community Shield.

My instinct was to sell Haaland.

The process disagreed. One match was new evidence, but not enough to dismantle a squad we’d deliberately built around him.

So Haaland stayed, at reduced confidence, and one targeted change went through instead: Pascal Groß out, Christos Tzolis in.

---

What’s happened since

Between then and the deadline I added more signals: confirmed team news, expert FPL opinion, what other managers in the game are doing, and a second AI model.

The most useful intervention came from Claude.

I didn’t ask it to pick a better team. I asked it to attack our reasoning.

And it found something.

We’d argued that the Community Shield wasn’t strong enough evidence to sell Haaland. Fair enough on its own. But look at the captaincy call sitting right next to it: Bruno Fernandes got the armband partly because Haaland’s confidence had just been knocked by that same 3–0.

You can’t rule a piece of evidence too weak to justify one decision and then quietly lean on it to justify another, in the same breath.

That’s not a football opinion. It’s a hole in the argument, and it was ours.

Fixing the argument, not just the team

Finding a bad argument doesn’t automatically mean the underlying decision is wrong. It means the decision needs a reason that actually survives.

Keeping Haaland is a squad-structure call, built to hold for weeks. Captaincy is a one-week call, remade every Gameweek. Treating one piece of evidence as decisive for both, in opposite directions, was the mistake — not necessarily the pick itself.

So I took the Community Shield out of the captaincy question entirely and asked it again on its own terms: fixture, current form signal, and role in each team’s attack.

I’m still leaning Bruno.

That’s a genuinely different position from where I started this section, even though the team sheet doesn’t move. Before, Bruno had the armband partly because Haaland had been quietly marked down by a match we’d separately decided didn’t count. Now he has it because, asked cleanly, Manchester United’s opener still looks softer than Manchester City’s.

Editorial xG — Gameweek 1

Kinský

Gabriel — Maguire — Mitchell

Bruno Fernandes (C) — Mbeumo — Tzolis — Le Fée

Haaland (VC) — João Pedro — Igor Jesus

Captain: Bruno Fernandes. Vice-captain: Haaland. Chip: none.

Unless something genuinely new arrives before the deadline, that’s it. No more optimising for its own sake.

Why these players survived the cut

The useful thing about explaining the whole team is that it stops the Haaland argument swallowing everything else.

This is not a collection of the 15 players I think are individually “best”. It is a constrained portfolio: £100m, fixed positions, uncertain minutes, opening fixtures, captaincy, bench cover and the need to leave myself somewhere to go later.

So each player has a job.

Kinský — value and a route into Spurs’ defence. At £4.5m, the attraction is not spectacular upside; it is getting a likely starting goalkeeper without spending money I would rather use in attack. The wider signal around Spurs’ defence has also been positive enough that I am comfortable starting there rather than paying up for a premium goalkeeper.

Gabriel — the boring premium I am happy to pay for. Arsenal’s defensive level, his attacking threat from set pieces and his potential for defensive-contribution points make him less dependent on one route to returns. Calafiori became the fashionable cheaper alternative after scoring in the Community Shield, but ownership movement is evidence of what other managers believe, not proof that Gabriel has become a worse asset.

Maguire — price, fixture and set-piece threat. Manchester United’s defence has looked less convincing than I hoped, which is why he is not a long-term lock. But at £5m, with an attractive opening fixture and genuine threat from Bruno’s delivery, the case is still strong enough for Gameweek 1. This is a hold with an exit plan, not a declaration that United suddenly have an elite defence.

Mitchell — a cheap starter rather than an exciting theory. At £4.5m, I mostly want minutes and a usable fixture from this slot. I looked at more attacking alternatives such as Oscar Mingueza, but the evidence on role and starting security was not strong enough to force a late switch. Sometimes the least interesting player is doing useful structural work.

Bruno Fernandes — premium involvement plus the captaincy decision. United’s opening fixtures are attractive, his role keeps him close to penalties and set pieces, and he is central enough to the attack that I am willing to concentrate the captaincy risk here. The important thing is that the armband now rests on that case, not on a contradictory reading of Haaland’s Community Shield performance.

Mbeumo — the role signal I don’t want to give up yet. Pre-season use through the middle made an £8m midfielder look potentially more valuable than his listed position suggests, and United’s early fixtures strengthen the case. Semenyo became a serious alternative after City’s strong attacking pre-season, but for the first decision window I prefer Mbeumo rather than doubling down on City alongside Haaland.

Tzolis — the deliberate risk. He is the one actual change from the first draft. His Community Shield performance strengthened the case, and Arsenal opening at home to Coventry gives the upside an immediate fixture. But this is also where the experiment could expose recency bias: the reigning FPL champion made a strong case for the cheaper Groß as a set-piece-heavy enabler. I have kept Tzolis, but this is the pick I am most interested in reviewing honestly afterwards.

Le Fée — the mid-price enabler with a real fixture. He is not here because I expect him to outscore the premiums. He is here because the £6m slot lets the expensive structure work while still giving me a player I am prepared to start, rather than a dead budget placeholder. The opening fixture against Ipswich is good enough to justify that compromise.

Haaland — the structure itself. This has been the main argument of the week. £15.5m is enormous, City looked poor against Arsenal, and my instinct was to use the money elsewhere. The counterargument is that one match is weak evidence against several seasons of extraordinary early-season output, and that the rest of the player pool is cheap enough to build a credible team around him. The reigning FPL champion independently made essentially that case too. So he stays — not because the concern disappeared, but because it did not cross the threshold required to restructure the whole squad.

João Pedro — probably the easiest attacking hold. Chelsea’s pre-season attacking form and João Pedro’s repeated goals gave me a pattern rather than a single flashy result. At £7.5m, the balance between price, role and evidence looks strong enough that he has rarely been part of the debate.

Igor Jesus — the weakest starting assumption, consciously accepted. At £6m, his pre-season scoring and opening fixture made him worth holding, but his starting security is less convincing than João Pedro’s. This is not a blind spot: it is the softest position in the XI, and if the minutes evidence turns against him, it is one of the first places I would change.

And the bench

Dubravka, Bobby Thomas, John Egan and Regan Slater are mostly budget enablers. The aim is basic cover, not four extra players I am desperate to start every week. Thomas is the bench defender I am most comfortable with; Egan and Slater are there primarily because the money has been concentrated in the XI.

That is also why the Gameweek 1 chip call is none. This is not a Bench Boost squad. If I eventually want to use that chip properly, the bench will need to become a deliberate part of the team rather than a place to store the cheapest available players.

The distinction matters. A squad can contain individually attractive players and still be badly constructed. What I am trying to test here is whether each pick makes sense inside the system around it.

Where this leaves the experiment

ChatGPT didn’t make this decision.

Claude didn’t make this decision.

An FPL expert didn’t make this decision. Neither did the wider consensus of other managers building similar squads this week.

Between them they gave me different kinds of intelligence, including — and this is the part I think actually matters — reasons why my own reasoning might be wrong. That last one is the hardest to generate on your own, because you already believe your argument, and the whole point of a blind spot is that you can’t see it from the inside.

I still had to weigh it and decide. Nobody handed me a verdict; they handed me a better version of the problem.

That is really what I am using this season to learn: not whether AI is a better fantasy manager than I am, but where it improves my judgement, where it distorts it, where my own instinct should resist it, and how the two coexist when neither side has certainty.

The proper season should make that more interesting because the players and teams will start acquiring narratives of their own. The question won’t only be whether Palmer, Bruno, Haaland or anyone else has good numbers. It will be which story I am starting to tell about them, what evidence supports that story, what contradicts it, and whether it has become strong enough to justify changing the team.

That’s the same reason the whole thing gets written down rather than just thought through: editorial judgement that isn’t captured somewhere gets rediscovered instead of reused. Reasoning that’s on the record before the result is reasoning that can actually be checked and revised, not quietly reinvented next Gameweek.

Now comes the part I have no control over: the football.

I’ll come back after the games and set this record against what actually happened. But the question I actually want an answer to is not whether the team scored a lot of points.

It’s whether this was a good decision given what I knew at the time — even if the result says otherwise.

Editorial Intelligence vs FPL: 0–0.