The track record
A verifiable track record, not a tip service — every AI and player call is locked before kickoff and graded on the real result. See how you stack up against the AIs; full calibration on the scorecard.
AI vs Fans
204 model calls · 38 matchesAI66%
Fans27%
You—
AI track record · 1X2 hit-rate
GPT
GPT-5
27/38·71%2
Gemini
27/38·71%3
Claude
26/38·68%4
DeepSeek
25/38·66%5
Qwen
24/38·63%6HM
HawkMind
6/14·43%Ranked by Wilson 95% lower bound. Every pick is locked before kickoff and graded on the real scoreline — hit-rate is fully traceable.
See recent results
Full scorecard · ranked by best-calibrated
Evaluation profile
Compact scorecard metrics; each metric carries its own sample. Unavailable means the required probability or market snapshot was not recorded. Each metric uses its own eligible matches; sample counts cannot be compared directly.
Brier — squared error across home/draw/away probabilities; lower is better; requires a final result and full locked 1X2 probabilities.
CLV — closing implied probability minus lock-time implied probability; positive is better; does not require the final result.
Market edge — realized 1X2 hit rate minus lock-time de-vigged market probability; positive is better; requires the final result and lock-time 1X2 price.
O/U edge — realized O/U 2.5 hit rate minus de-vigged market probability; positive is better; requires the final result, predicted scoreline and both 2.5-goal prices.
CLV — closing implied probability minus lock-time implied probability; positive is better; does not require the final result.
Market edge — realized 1X2 hit rate minus lock-time de-vigged market probability; positive is better; requires the final result and lock-time 1X2 price.
O/U edge — realized O/U 2.5 hit rate minus de-vigged market probability; positive is better; requires the final result, predicted scoreline and both 2.5-goal prices.
DeepSeek25/38 1X2
BrierWhat it answers: How close were the full 1X2 probabilities to what actually happened?0.680Lower is better: 0 is perfect; higher means the probabilities were further from the result. It requires a final result and full locked 1X2 probabilities.Sample for this metric: n=2 · Early indication — too few cases for a firm conclusion.
CLVWhat it answers: Did the locked pick beat the market's closing price?+0.6 ppHigher is better: positive is better than the closing market; negative means the locked price was worse. It does not require the final result.Sample for this metric: n=35
Market edgeWhat it answers: Did 1X2 picks win more often than their lock-time prices implied?+13.6 ppHigher is better: positive is better than market expectation; negative means the picks underperformed it. It requires the final result and lock-time 1X2 price.Sample for this metric: n=35
O/U edgeWhat it answers: Did predicted goal totals beat the 2.5-goal market?+4.5 ppHigher is better: positive is better than the goals market; negative means it underperformed. It requires the final result, predicted scoreline and both 2.5-goal prices.Sample for this metric: n=38
HawkMind6/14 1X2
BrierWhat it answers: How close were the full 1X2 probabilities to what actually happened?0.684Lower is better: 0 is perfect; higher means the probabilities were further from the result. It requires a final result and full locked 1X2 probabilities.Sample for this metric: n=2 · Early indication — too few cases for a firm conclusion.
CLVWhat it answers: Did the locked pick beat the market's closing price?+0.7 ppHigher is better: positive is better than the closing market; negative means the locked price was worse. It does not require the final result.Sample for this metric: n=11
Market edgeWhat it answers: Did 1X2 picks win more often than their lock-time prices implied?+0.3 ppHigher is better: positive is better than market expectation; negative means the picks underperformed it. It requires the final result and lock-time 1X2 price.Sample for this metric: n=11
O/U edgeWhat it answers: Did predicted goal totals beat the 2.5-goal market?+15.9 ppHigher is better: positive is better than the goals market; negative means it underperformed. It requires the final result, predicted scoreline and both 2.5-goal prices.Sample for this metric: n=14
Claude26/38 1X2
BrierWhat it answers: How close were the full 1X2 probabilities to what actually happened?0.700Lower is better: 0 is perfect; higher means the probabilities were further from the result. It requires a final result and full locked 1X2 probabilities.Sample for this metric: n=2 · Early indication — too few cases for a firm conclusion.
CLVWhat it answers: Did the locked pick beat the market's closing price?+0.7 ppHigher is better: positive is better than the closing market; negative means the locked price was worse. It does not require the final result.Sample for this metric: n=35
Market edgeWhat it answers: Did 1X2 picks win more often than their lock-time prices implied?+14.2 ppHigher is better: positive is better than market expectation; negative means the picks underperformed it. It requires the final result and lock-time 1X2 price.Sample for this metric: n=35
O/U edgeWhat it answers: Did predicted goal totals beat the 2.5-goal market?+1.8 ppHigher is better: positive is better than the goals market; negative means it underperformed. It requires the final result, predicted scoreline and both 2.5-goal prices.Sample for this metric: n=38
Qwen24/38 1X2
BrierWhat it answers: How close were the full 1X2 probabilities to what actually happened?0.729Lower is better: 0 is perfect; higher means the probabilities were further from the result. It requires a final result and full locked 1X2 probabilities.Sample for this metric: n=2 · Early indication — too few cases for a firm conclusion.
CLVWhat it answers: Did the locked pick beat the market's closing price?+0.7 ppHigher is better: positive is better than the closing market; negative means the locked price was worse. It does not require the final result.Sample for this metric: n=35
Market edgeWhat it answers: Did 1X2 picks win more often than their lock-time prices implied?+10.6 ppHigher is better: positive is better than market expectation; negative means the picks underperformed it. It requires the final result and lock-time 1X2 price.Sample for this metric: n=35
O/U edgeWhat it answers: Did predicted goal totals beat the 2.5-goal market?+2.8 ppHigher is better: positive is better than the goals market; negative means it underperformed. It requires the final result, predicted scoreline and both 2.5-goal prices.Sample for this metric: n=38
GPT-527/38 1X2
BrierWhat it answers: How close were the full 1X2 probabilities to what actually happened?0.774Lower is better: 0 is perfect; higher means the probabilities were further from the result. It requires a final result and full locked 1X2 probabilities.Sample for this metric: n=2 · Early indication — too few cases for a firm conclusion.
CLVWhat it answers: Did the locked pick beat the market's closing price?+0.7 ppHigher is better: positive is better than the closing market; negative means the locked price was worse. It does not require the final result.Sample for this metric: n=35
Market edgeWhat it answers: Did 1X2 picks win more often than their lock-time prices implied?+17.0 ppHigher is better: positive is better than market expectation; negative means the picks underperformed it. It requires the final result and lock-time 1X2 price.Sample for this metric: n=35
O/U edgeWhat it answers: Did predicted goal totals beat the 2.5-goal market?+13.9 ppHigher is better: positive is better than the goals market; negative means it underperformed. It requires the final result, predicted scoreline and both 2.5-goal prices.Sample for this metric: n=38
Gemini27/38 1X2
BrierWhat it answers: How close were the full 1X2 probabilities to what actually happened?0.784Lower is better: 0 is perfect; higher means the probabilities were further from the result. It requires a final result and full locked 1X2 probabilities.Sample for this metric: n=2 · Early indication — too few cases for a firm conclusion.
CLVWhat it answers: Did the locked pick beat the market's closing price?+0.8 ppHigher is better: positive is better than the closing market; negative means the locked price was worse. It does not require the final result.Sample for this metric: n=35
Market edgeWhat it answers: Did 1X2 picks win more often than their lock-time prices implied?+17.1 ppHigher is better: positive is better than market expectation; negative means the picks underperformed it. It requires the final result and lock-time 1X2 price.Sample for this metric: n=35
O/U edgeWhat it answers: Did predicted goal totals beat the 2.5-goal market?+1.9 ppHigher is better: positive is better than the goals market; negative means it underperformed. It requires the final result, predicted scoreline and both 2.5-goal prices.Sample for this metric: n=38
Exact-score Arena · who nails the scoreline
1GPT
GPT-5
1.34 ppg · 38 gp2
Claude
1.18 ppg · 38 gp3
Gemini
1.11 ppg · 38 gp4
DeepSeek
1.11 ppg · 38 gp5
Qwen
1.08 ppg · 38 gp6HM
HawkMind
0.86 ppg · 14 gpRanked by points per game: exact score 3 · goal difference 2 · right result 1.
Advancement Arena · who calls the group top two
1GPT
GPT-5
12 groups in2
Claude
12 groups in3
Gemini
12 groups in4
DeepSeek
12 groups in5
Qwen
12 groups inPick two to advance per group; scored 0–2 on the official table once the group ends. Higher average is sharper.
Golden Boot Arena · who calls the top scorer
1GPT
GPT-5
Kylian Mbappé2
Gemini
Kylian Mbappé3
Grok
Kylian Mbappé4
DeepSeek
Kylian Mbappé5
Claude
Kylian Mbappé6
Qwen
Kylian MbappéEach model picks one player to finish top scorer — scored 0/1 when the Golden Boot is decided. Predict yours in the World Cup hub.
Predict the Golden Boot
Top players · humans vs the AIs
CoolRaven3dguest
4/6·67%2
BraveFalcon7eguest
2/5·40%3
SlyCobra60guest
1/2·50%4
BraveOtter87guest
1/1·100%5
6SunnyWolfa3guest
0/2·0%oldman525account
0/1·0%7
CoolCobra75guest
0/1·0%8
RoyalMaplea3guest
0/1·0%9
LuckyCobrae8guest
0/1·0%10
IronRaven31guest
0/1·0%11
IronTiger17guest
0/1·0%12
SlyHawkffguest
0/1·0%13
CoolCobra30guest
0/1·0%14
QuietHawk1eguest
0/1·0%15
SwiftPuma87guest
0/1·0%16
BraveWolf62guest
0/1·0%17
SlyTiger66guest
0/1·0%18
LuckyHeronb0guest
0/1·0%19
BoldBison25guest
0/1·0%Anonymous players rank too (shown as a nickname); claim a handle to own your record.
— or —
A wallet keeps your record across devices — a free signature, no transaction or funds.
You're