Were we right?
Every match — group and knockout — our pre-match odds set against Kalshi's pre-match price. Two scorelines: who forecast better (de-vigged Brier & log-loss), and how a bankroll backing our disagreements into Kalshi's real, vig-inclusive prices would have done.
Lower Brier & log-loss = sharper. Market price is de-vigged to a fair probability first. Green marks the better of the two.
Compounds a $1,000 bankroll, sizing each bet by 0.5× the Kelly edge. Sensitive to model overconfidence & thin liquidity — treat as a ceiling.
The ledger
0–0
0–1
1–5
1–1
0–2
2–1
0–0
3–1
1–3
3–3
0–1
2–1
1–1
1–1
1–2
3–0
2–0
2–13–2
2–0
3–02–1
2–0
1–13–21–00–11–43–20–02–02–11–23–10–235 of 104 played games scored (10 group, 25 knockout) · 32 had no logged pre-match prediction (games before we started snapshotting) · 37 had no Kalshi market. Knockout ties are graded on Kalshi's 2-way “advances” market (extra time & penalties included); group games on the 3-way moneyline. Bets are placed at the single largest GAFFER edge per game when it clears 2%, entered at Kalshi's pre-match ask. Accuracy compares de-vigged probabilities; P&L uses the raw, vig-inclusive ask — the honest cost of actually trading.