Gameweek 5: Our Three Most Confident Calls Went 1 From 3

FootyPulse's model entered Gameweek 5 with three matches sharing its highest confidence rating of the round — 68% apiece for Manchester City, Nottingham Forest and Leeds United. Only one of those three calls was correct. Meanwhile, the model's closest calls of the week — three narrow 38% forecasts — landed two from three.

It is a small sample, and it would be wrong to draw a broad conclusion from it. It is still a useful, honest reminder of what a 68% probability means: roughly a two-in-three chance, not a certainty. This week, two of the three were wrong.

Across the full round, FootyPulse called 5 of 10 outcomes correctly — a solid rebound from Gameweek 4's 2 from 10.

The three 68% calls: 1 from 3

• Manchester City (68%) vs Sunderland — correct. City won 5-3 in a chaotic match. The home-win call held, even though the eight-goal scoreline was far beyond the model's 3-0 forecast.
• Nottingham Forest (68%) vs Coventry City — wrong. Coventry arrived on a four-match losing run and left with a 1-0 win, ending that sequence at Forest's ground.
• Leeds United (68%) vs Crystal Palace — wrong. Leeds were forecast to win after scoring four against Newcastle the previous week; Palace held out for a 0-0 draw.

How confidence lined up with results across the whole round

Fixture	Pre-match call	Confidence	Actual result	Correct?
Man City vs Sunderland	Home, 3-0	High (68%)	5-3 (H)	✅
Nottingham Forest vs Coventry	Home, 2-0	High (68%)	0-1 (A)	❌
Leeds vs Crystal Palace	Home, 3-1	High (68%)	0-0 (D)	❌
Everton vs Ipswich Town	Home, 2-1	Mid (58%)	1-0 (H)	✅
Bournemouth vs Liverpool	Away, 1-2	Mid (54%)	0-1 (A)	✅
Tottenham vs Aston Villa	Home, 1-0	Mid (48%)	2-3 (A)	❌
Brighton vs Arsenal	Away, 1-2	Mid (48%)	3-0 (H)	❌
Brentford vs Chelsea	Home, 3-2	Closest (38%)	3-0 (H)	✅
Newcastle vs Hull City	Home, 2-1	Closest (38%)	2-1 (H)	✅
Fulham vs Man Utd	Away, 1-2	Closest (38%)	1-1 (D)	❌

Grouped by confidence tier, the 68% tier went 1 from 3 (33%), the 48–58% tier went 2 from 4 (50%), and the 38% tier went 2 from 3 (67%) — the reverse of the order you would expect. With only three or four fixtures in each group, this is not evidence that higher-confidence calls are less reliable. It is a reminder that a single gameweek is a small sample and that 68% always leaves a real chance of being wrong.

The result nobody saw coming

Brighton's 3-0 win over Arsenal was the shock of the round — and it was not only FootyPulse that missed it. Arsenal arrived with a perfect away record and two away clean sheets this season; Chris Sutton, the BBC guest and BBC's AI all forecast an Arsenal win. Nobody in the comparison panel saw Brighton coming.

For context: how we compared with the BBC panel

Gameweek 5 was not built as an AI-vs-BBC verdict, but the comparison is useful because the numbers were close. FootyPulse tied BBC's AI and the BBC guest at 5 from 10, with Chris Sutton on 4 from 10. The full, updated head-to-head is tracked on the AI vs Pundits page.

One call worth highlighting came at Brentford vs Chelsea. FootyPulse forecast a home win while two of the three BBC calls went with Chelsea. Brentford won 3-0.

Season record after five gameweeks

Gameweek 5's 5 from 10 takes FootyPulse's season record to 18 correct outcomes from 47 scorable fixtures — 38%, up from 35% after Gameweek 4. The full, published breakdown by gameweek is on the accuracy page; last week's difficult round is covered in the Gameweek 4 review.

What's next

Half-time re-prediction data is now accumulating across five gameweeks. That gives us a growing record to examine in a future article; the methodology page explains how the live re-predictions work. Gameweek 6's preview lands Friday, and we will return to how the model's strongest calls perform next time.

FAQ

How did FootyPulse's AI do in Gameweek 5? FootyPulse correctly called 5 of 10 Gameweek 5 outcomes, a rebound from Gameweek 4's 2 from 10. Its three highest-confidence calls of the round, all rated 68%, went 1 from 3, while its three closest calls went 2 from 3.

What was FootyPulse's biggest miss in Gameweek 5? Brighton's 3-0 win over Arsenal was the standout surprise. FootyPulse, Chris Sutton, the BBC guest and BBC's AI all forecast an Arsenal win.

Does a 68% prediction mean the AI is certain? No. It means the model rates that outcome at roughly a two-in-three chance based on its inputs. Gameweek 5 illustrated the remaining uncertainty: two of the three 68%-rated calls were wrong.

How does FootyPulse compare with the BBC's Premier League predictions? In Gameweek 5, FootyPulse tied BBC's AI and the BBC guest at 5 from 10, ahead of Chris Sutton on 4 from 10. The full season comparison is tracked on the AI vs Pundits page.