Test against the past, in order
Walk-forward results use earlier information to evaluate later games. History is not a promise about the next slate.
Independent sports intelligence · 8 desks
Game forecasts, transparent paper records and a DraftKings lineup desk — built to show what the model knows, where the data came from and what still needs proving.
The board
Moneylines and totals from the Kalshi top-20 pool kicking off today in Eastern Time. Every pick cleared the swarm review, then a fresh confirmation round of passes the first round never scored — only the tickets both rounds agree on are shown. Ask −200 or better; a total within a point of the model number.
Games, top 20, and DraftKings lineups. Resets at midnight ET.
Checking today…
Loading today's moneylines…
Loading today's over/unders…
Eight desks
Follow the games and markets you care about. Each desk keeps its own slate, sources, model context and results.
The record, in the open
Per-sport paper books and confirmed exchange settlements, refreshed from the live desk feed. Open picks are not settlements; a blank implies no result.
Combined net is the sum of separate sport books. Paper performance is simulated and is not a customer return. Availability, fees, slippage and sample size matter.
Paper ML / ATS / totals by sport. Exchange chart: confirmed settlements only.
Paper and exchange behavior only changes after the configured review window. The status below is read from the desk API.
Loading paper and confirmed fill records…
Method
A forecast is only useful when you can see the sample, the market baseline and the limit.
Walk-forward results use earlier information to evaluate later games. History is not a promise about the next slate.
Paper tracking and exchange fills stay separate. A combined figure sums isolated books; it is not shared capital.
Freshness, sample size and availability change what a desk can support. Missing lines stay missing — never invented.
Fine print
Clear terms make it easier to read a number for what it is — and not more.
No. Forecasts are uncertain estimates. A model can be wrong, and past performance does not guarantee future results.
Evaluation follows time order: information available before an event is used to score forecasts on later events, rather than fitting and testing on the same future.
No. Paper books are simulated tracking. Exchange records are reported separately and only from confirmed settlement data.
A source may be unavailable, the event may not be covered, or the sample may be too small. An empty field beats invented market data.