How Novus Odds works

Shared experiment engine

Novus Odds labs share a seeded Monte Carlo kernel, odds mathematics, bankroll/drawdown reducers, and analytics visualizations. Large simulations run in memory; only aggregated results and sampled curves are stored locally.

Every lab is educational and synthetic. No live odds APIs or real team/player names are used.

Odds Lab model

The Odds Lab generates synthetic two-outcome betting markets. Each simulated event receives a quoted American price, an implied probability, an estimated fair probability after removing a simplified proportional overround, and a simulated underlying outcome probability.

Selection filters (favorites, underdogs, thresholds, edge) and optional A/B compare arms decide which events are staked. Repeating the same configuration with the same seed reproduces the same experiment.

Other labs

• Strategy Lab applies stake and selection rules to synthetic tapes and reports bankroll paths, ruin events, and compare arms.
• Sports Lab simulates fictional matchups from ratings or multi-runner fields, then derives fair and priced odds.
• Props Lab samples fictional player distributions and sweeps lines for EV curves.
• Casino Lab models house-edge games (roulette, blackjack, baccarat, craps, slots) with transparent bet-type mathematics.
• Datasets builds fictional multi-season universes and exports aggregate JSON samples.

Odds generation

Market prices are sampled in log-odds space rather than directly across American odds. This provides a more mathematically consistent probability scale and avoids allowing the numerical size of positive American odds to dominate the generated distribution.

The generator uses a triangular sampling pattern within the selected range so that prices near the centre of the experiment range occur more often than the extreme boundaries.

Market overround

The margin control represents a simplified proportional two-outcome overround. The quoted implied probability is divided by one plus the configured margin to establish the simulation's baseline fair probability.

This is an experimental abstraction. Real sportsbooks may apply unequal margin, market-specific pricing, correlated risk adjustments, promotions, limits, and other factors that are not represented by this simplified model.

Pricing uncertainty

Pricing uncertainty introduces a normally distributed deviation in logit space between the baseline market estimate and the simulated true probability. Applying uncertainty in logit space ensures generated probabilities remain between zero and one.

Increasing pricing uncertainty creates a wider spread between quoted market probabilities and simulated underlying probabilities. It should be interpreted as an experimentation control rather than a measurement of any real sportsbook's accuracy.

Experiment metrics

• Observed win rate is the percentage of simulated events that resolved as wins.
• Expected ROI is calculated from each event's simulated true probability and quoted odds before the random outcome is resolved.
• Observed ROI is calculated from the actual simulated wins and losses in the experiment.
• The 95% interval uses the Wilson score method to describe uncertainty around the aggregate observed win rate.
• Longest winning and losing streaks measure consecutive experiment outcomes and demonstrate how variance can appear even in large samples.
• Odds buckets aggregate results by American price range for comparison.

Local data storage

The current release does not require an account or remote experiment database. Completed experiment summaries and chart data are saved in the browser's local storage so they remain available on the Results page on the same browser and device.

Clearing site storage or using the Results page's Clear History action removes this locally stored experiment history.