v2 calibrated edges →
2026 World Cup · prediction markets

The model's board, scored against the price.

A statistical model forecasts every World Cup market, pre-committed and timestamped against the live prediction markets, then graded the way desks do: Closing-Line Value (did the price drift toward the model?) and calibration, as games resolve. Model vs market; the gaps are the calls.
Title race
Model champion probability vs the market, top of the field.
● model    ● market  ·  the gap is the disagreement
Every forecast
Model vs market across all markets. Sort any column, filter by market, search a team. Edge = model − market. CLV = how far the price has since moved toward the model.
loading…
,

Closing-Line Value. Share of forecasts the market has since drifted toward, a skill signal independent of who eventually wins. Scope matters: this counts every board forecast that has a closing price. A narrower cut of the ledger — directional calls with an edge of at least 0.5pp, 222 of them — is 58.6% positive (writeups/_clv_results.json).

Early sample. CLV is conclusive only over a few hundred forecasts, and these are correlated across markets, so treat it as a signal forming, not a verdict.

Calibration
Biggest calls
Where the model most disagrees with the market, +/− edge in points, model v market. CLV ▲ if the price has since moved toward us, ▼ if it's drifted away.