In-house odds: our own models, not a copied market
Every price on Nocturn Sport comes from models we built and tested ourselves, and operators can use the same prices.
Updated October 5, 2026
In-house odds means that a sportsbook calculates its own probabilities instead of copying the prices of other bookmakers or buying a price feed. Nocturn Sport does exactly that, for its own site and for the operators it supplies. Here is how the models work and why it matters for a business.
Why build our own models
- Independence. A copied price disappears when the source does. Our odds depend on our models and our results data only.
- Consistency. All markets on an event come from one model output, so the 1X2, the totals and the handicaps never contradict each other.
- Explainability. We can tell an operator why a price is what it is, which helps risk teams decide when to override.
- Control. The margin, limits and closing time are applied on top of our probabilities and set per operator.
The three model families
| Sports | Model | Markets produced |
|---|---|---|
| Football, ice hockey, baseball | Attack and defence strengths give expected goals; Poisson (with a Dixon-Coles low-score correction in football) or negative binomial for baseball runs | 1X2, totals, both teams to score, handicaps |
| Basketball, American football, rugby | Ratings in points plus home advantage give an expected margin and total, each normally distributed | Winner with overtime, main spread, main total; 1X2 with the draw in rugby |
| Tennis, MMA, League of Legends | Elo ratings give the chance to win one game or map, then the whole series is computed | Series winner, map handicap ±1.5, total maps, correct score |
Measured calibration
We test our models on matches they have not seen. Our football model, measured on past matches, was trained on 10,783 games and then evaluated on 1,973 later games; our basketball model was trained on 8,665 and evaluated on 1,712. In both tests, the models achieved a lower log-loss than a naive base-rate forecast. We do not publish other accuracy claims.
From probability to price
The bookmaker margin is added proportionally across a market: price = 1 / (probability × (1 + margin)), with the margin chosen by each operator. Prices are kept within the operator's minimum and maximum odds, and markets close one minute before the start by default, so late team news cannot be exploited against stale prices.
Risk tools on top of the model
No model replaces judgement. The back office lets operators suspend events, override any price, settle results and monitor liability, with limits on stake, payout and selections per ticket. See the sportsbook software overview and the odds feed structure for details.
Our odds can be integrated through SoftAggregator or directly by API. Prices are on request: the contact page lists every channel, including [email protected] and @mentionso on Telegram. For the player-facing explanation, read how we price odds.
Frequently asked questions
Are your prices taken from other bookmakers?
Never. Each price starts as a probability produced by our own models from historical results, and no external price source is involved.
How accurate are the models?
On held-out test sets of 1,973 football games and 1,712 basketball games, both models scored a lower log-loss than a simple base-rate benchmark. That is the only accuracy figure we publish.
Can operators adjust the prices?
Yes. Operators set their margin and odds range and can override any price or suspend any event from the back office.
Which sports are covered by the in-house models?
Coverage grows sport by sport, from football, basketball and ice hockey to American football, baseball, rugby, tennis, MMA and League of Legends.