Market calibration receipt
What this is. Every market MarketIntel lists is snapshotted once per UTC day. When the venue
confirms a result (Kalshi's settled result, Polymarket's UMA resolution), the market is scored against the price it
traded at 1, 3 and 7 days before it closed. Nothing is selected on outcome — every snapshotted market
that resolves is included, winners and losers alike. Every cell carries n and a Wilson 95 % interval; cells
under 30 outcomes are labelled thin and should not be quoted. This measures the markets, not
MarketIntel.
Collecting evidence — not yet quotable
Loading the latest report…
—
outcomes scored
—
markets watched
—
daily snapshots
—
latest frozen week
Priced → resolved YES
| Price band | n | Avg price | Resolved YES (95% CI) | Brier |
|---|
avg price resolved YES · equal bars = well calibrated
By category
All venues · price 1 day before close
| Category | n | Avg price | Resolved YES (95% CI) | Brier |
|---|
Weekly scoring
ISO week the market resolved · 1-day price · frozen snapshots never change after the fact
| Week | n | Avg price | Resolved YES | Brier |
|---|
No week has a scored outcome with a 1-day-before price yet — the daily snapshot ledger started recently. Rows appear as markets resolve.
Re-derive every figure
The CSV lists every scored outcome — market id, venue, category, the 1/3/7-day prices, the result and the resolution time. The Python script rebuilds this whole report from that CSV. If a number here can't be re-derived, tell us at [email protected].
Methodology
—
What this is not
A historical record of prices vs. outcomes on public venues, published so anyone can check it. It is not investment, trading or wagering advice, and it does not measure any MarketIntel forecast. Extreme prices on thin books are quotes, not probabilities — the price-band split exists because cheap-entry outliers dominate any single aggregate.