Foreline
Foreline/Products/Parlay Protection
Product 4 · Parlay ProtectionJoint prices that agree with your marginals
A bet-builder multiplies legs that are not independent. We price the joint outcome from a coherent score matrix built out of de-vigged marginals — 1X2 and totals agree with each other by construction, and the handicap, which the matrix is not fitted to, comes back out with a median error of 0.83 pp.
One matrix, so the legs cannot contradict each other
De-vigged marginals from the surface are assembled into a score matrix. Every market you quote is then a sum over the same cells — which is what makes a joint price consistent with the singles sitting next to it on your own site.
- 1X2 and totals agree by construction. They are read off the same matrix, so there is no combination of the two you can hold that is priced from two different worlds.
- The handicap is the held-out check. The matrix is not fitted to the handicap ladder, yet reproduces it with a median error of about 1 pp. That is the honest test of coherence: reconstruct a market the matrix was never shown.
- Joint probabilities for combinations — outcome × total and handicap × total: the exact families where a naive multiply-the-legs builder leaks the most, because result and goal volume are correlated in every league. Four families are calibrated inside the headline band; the rest are returned with their own, wider numbers rather than folded into an average.
- Every price carries its measured error. Each joint price is returned with the calibration band we measured for that family against realised outcomes, and with the worst family called out rather than the average one. The band describes how far the model has been seen to miss — it does not vary with whether a leg sat on a quoted rung or an interpolated one, and we do not claim that it does.
maxGap is the largest deviation between predicted and realised frequency across calibration buckets. We publish the maximum rather than the mean because the maximum is what a sharp customer will find and use.
Where the model is weakest
A pricing vendor that reports only its good buckets is selling you its residual risk. Here is ours, stated plainly.
- Why it matters to you: “home and over” is one of the most-built families in any bet-builder. If you take our number at face value on that family without the band, you inherit our error.
- What we do about it: the band is widened where the miss lives, the miss is reported per bucket, and the calibration is re-measured on every release rather than quoted from the launch report.
- What we do not do: quietly patch the family with a fudge factor and keep reporting an aggregate that hides it.
Audit first, feed second
The audit is the cheap way to find out whether this is worth an integration — it runs on prices you already have.
The marginals behind the matrix are the ordinary
GET /v1/surface response — if you already take
the surface, the audit needs nothing new from your side except the builder prices. Player props
(goals or shots for a named footballer) are on the roadmap, phase 3, and are not part of this
product today.
The other three products
Fair-Price Surface
The whole ladder, de-vigged and banded — the marginals this matrix is built from.
See the product →Player Radar
Scores accounts on CLV against our close — including on lines the reference never quoted.
See the product →Line Radar
Main-line changes as a fact-stream, plus warnings when your price drifts from fair.
See the product →Request a parlay audit
Send us the combination families you offer and the period you want checked. We reply with the export format and what the audit will tell you.
We usually reply within one business day. Pricing is quoted per scope — ask and we will send terms.