Standings

The Model Arena

Season Table · Scored on the Forecasts We Issued

Standings

Every settled Forecast Points contract is a bout: each model is scored by its edge vs the consensus, realised against what actually happened, and every pair plays an ELO match (winner = the sharper Brier). Ranked by ELO — a running skill rating built from real issued forecasts.

Loading the arena…

P&L = Σ (p − q)·(y − q) — a model's probability p vs the consensus q, realised against outcome y. ELO starts at 1500 and moves on head-to-head Brier. Populates via points --mint + ledger settle-due.

Lead-Time Board

1 mile = 1 day

How many days ahead each player was already right — and held it. We walk each model's run-by-run odds back from valid time to the earliest committed & held call (past climatology), weighted by how surprising the outcome was — the axis where post-processing proves it buys usable lead. Climatology sits at 0.

Loading the Lead-Time Board…

HI = surprise-weighted days of lead · Lead = mean Usable Lead · Commit = share of outcomes it took a stand on · Recant = share it later broke. Fills via points --mint + ledger settle-due, read with ledger horizon.

Rating over time

Each settled contract advances every model's rating — watch who pulls ahead.

Head to head

Row model's record vs each column (W–L–T on head-to-head Brier).

Breakdowns

Leader per group (by mean Brier). Category/Lead fill as the market book widens.