Quant Library · The machine layer · Any

Conformal Scoreline Sets

A set of outcomes that holds the result 90% of the time, whatever the model gets wrong.

Split conformal prediction applied to a scoreline grid. The promise is distribution-free: it needs the future to look like the calibration window, and nothing else.

Read the specification (PDF) Run it on the API The the machine layer family · £1,199

Inputs

What it takes

Name	Type	Units	Where it comes from
calibration	list of (score matrix priced before the match, actual score)	-	a walk-forward run of any scoreline model
alpha optional	float	-	your risk level; 0.1 gives a 90% set
window optional	int	matches	how far back to calibrate; 380 is a season of a big league

Outputs

What it gives back

Name	Type	Units	Notes
q_hat	float	nats	the nonconformity quantile
set	list of scorelines	-
mass	float	probability	the model's own probability for the set, which is not the coverage
coverage	float	fraction	measured out of sample

Method

How it works

Score each calibration match by its nonconformity, −log P(result), under the price it was given before kick-off.
2 further steps in the licensed specification

Assumptions

Exchangeability: the coming match is drawn from the same distribution as the calibration window. A rule change or a mid-season shift breaks it, which is what the changepoint detector is for.

Limits

It gives coverage, not sharpness. A badly calibrated model gets a huge set, not a broken promise.
1 further limits in the licensed specification

Validation

What it was measured at

Walk-forward across every covered league, calibrated on a rolling window and checked on the matches after it.

Leagues	1
Calibration window	380 matches
90% promised	held 90.3%	average set of 13.7 scorelines out of 49
80% promised	held 80.8%	average set of 10.3 scorelines out of 49
67% promised	held 68.2%	average set of 7.1 scorelines out of 49

In the package

The full specification: inputs, method, assumptions and limits
The validation panel as measured on the day of purchase
A reference implementation extracted from the running source
Perpetual commercial use for one entity

Reference modules:

1 module(s), named in the licensed specification and shipped in the package

Read this before you buy. The platform's own walk-forward tests say the closing line forecasts football better than these models do. What they give you is a coherent price for every market and match, before the market opens, with the band and the working attached. Any model here that has been beaten by something simpler says so in its limits, and the validation panel above is generated from the platform's result files rather than typed in.