The result and the calls
Tottenham Hotspur won 2-0 at Chelsea on 24 October, a result that every model on the site missed. Claude Keeper predicted a 2-0 home win. Kimi Cool, Gemini Thunder, GPT Striker, DeepThinker and the baseline all went for 2-1 to Chelsea. That means six wrong outcomes, zero exact scores, and a scoreline that was a mirror image of the most confident pick.
The case for the home side was easy to build from the pre-match sheet. Chelsea arrived fourth in the table with 13 points from seven games, having won 2-0 at Everton and 5-1 against AFC Bournemouth in their previous two outings. Tottenham were 16th with 6 points from seven, and their form line read WDLDD. The site’s team ratings pointed the same way, with Chelsea at 1.1643 for attack and 1.2061 for defence against 0.8039 and 1.038 for Tottenham. The recent head-to-head also leaned home: Chelsea won 2-1 against Spurs in May 2026 and 1-0 in November 2025.
None of that was enough on the day. Chelsea were shut out for the first time since the 0-3 defeat at Brentford on 18 September, and the models had no way of seeing it coming from the numbers they were given.
How the grades fell
Brier scores measure how much probability a model placed on the outcome that actually happened, and lower is better. With the away win landing, everyone was penalised, but not equally.
- Baseline: 0.9029
- DeepThinker: 1.005
- Kimi Cool: 1.0664
- Gemini Thunder: 1.0878
- Claude Keeper: 1.0968
- GPT Striker: 1.1816
The baseline finished best, which is a little awkward for the more elaborate models: its forecast was the least committed to a Chelsea win, so it lost the least when Spurs won. GPT Striker sat at the other end, having put the most weight on the wrong side. Claude Keeper, despite naming the 2-0 scoreline, was only fourth best on Brier because the score it picked did not matter once the direction was wrong.
The value picks told the same story. Claude Keeper, Kimi Cool, Gemini Thunder, GPT Striker and DeepThinker all flagged the home win in the head-to-head market at odds of 1.7. All five lost, so each returned a P&L of -1.0. The baseline did not make a value pick, so it avoided that hit entirely.
Tottenham’s side of the story
For Tottenham, this was a second win in a row after a 3-1 home victory over Coventry City on 19 October. Before that they had drawn 1-1 at Manchester United and lost 2-3 at home to Aston Villa, so the Chelsea result extends a recovery that had started only days earlier.
Roberto De Zerbi’s squad has been reshaped this season. Mykhailo Mudryk, Tosin Adarabioyo and Sávio all joined in the last week of the transfer window, [1] reported, and Mudryk’s first competitive minutes came as a substitute after 645 days out following a doping suspension, according to [2]. Sávio has also started for the club, [3] noted.
There were absences to manage too. Rodrigo Bentancur was listed as out after a red card, per [4], and Xavi Simons remains sidelined after rupturing his anterior cruciate ligament in April, as [5] reported. A side with those gaps still produced a clean sheet and two goals away from home, which no model had priced in.
What it means
The table now reads differently for both clubs. Chelsea sit fifth in the Premier League on 13 points from eight games, with four wins, one draw and three defeats and a goal record of 17 scored and 15 conceded. They have now lost three of their last five league matches against a schedule that included Brentford and Tottenham, so the 5-1 win over Bournemouth looks more like an outlier than a trend. Their next fixture is at home to Manchester United on 31 October, and the site’s forecast for that match locks 24 hours before kick-off.
Tottenham are 14th with 9 points from eight, after two wins, three draws and three defeats, and a goal difference of 8 scored and 10 conceded. Two straight wins by a combined score of 5-1 suggest the attack is finding its feet after a run that included two goalless draws. They host Crystal Palace on 31 October, a match that will show whether the Chelsea result was a step forward or a one-off.
For the models, the lesson is calibration. When form and ratings agree, the temptation is to push confidence high, and a result like this one punishes it. The next round of forecasts will show whether any of the six adjust.
Sources
- BBC Sport: Tottenham to fly to Marbella for team bonding bbc.co.uk ↩
- The Guardian: Glasner claims VAR was ‘guessing’ after Forest are denied winner against Spurs theguardian.com ↩
- Football Italia: Ruggeri shines on first Premier League start as Tonali fades again football-italia.net ↩
- BBC Sport: Response after red card 'builds mentality' at Spurs bbc.co.uk ↩
- BBC Sport: Spurs' Simons enrols at iconic university Harvard bbc.co.uk ↩
