How to Read a Novus Breakdown: Your First Simulation

Step 1. Start with a synthetic market

Every number in Novus Odds comes from a simulation, not a live sportsbook. When you open the Odds Lab you configure a synthetic market — an odds range, a market margin, some pricing noise, a sample size and a random seed — and the engine generates thousands of independent events with known underlying probabilities.

That last point is what makes a lab different from real betting: the simulation knows the true probability of every event, because it drew the outcomes from that probability. Real markets never show you the truth. Keeping that distinction front of mind is the whole point of the exercise.

A good first configuration is deliberately boring: an odds range spanning roughly -200 to +250, a 4% market margin, a modest amount of pricing noise, 25,000 events and any seed you like. Boring settings make the structure visible. Once you can predict what a change will do before you press run, you have understood the panel.

Step 2. Read the four numbers that matter first

Read these together, not in isolation. A tempting edge on a tiny sample of selections tells you far less than a smaller edge measured across the whole tape.

Work an example. A price of +150 implies 100 / (150 + 100) = 40.0%. If the simulation's true probability for that event was 43.5%, the edge is +3.5 percentage points. Expected ROI is 0.435 × 1.5 − 0.565 × 1 = +0.088, or +8.8% per unit staked. That number is what the lab expects on average — not what any particular run will hand you.

Notice how small the inputs are. Three and a half points of probability is a rounding error to the naked eye, and it produced an 8.8% expected return. The reverse is equally true: a few points in the wrong direction turns a confident-looking model into a steady loser. Precision in the probability estimate is the whole game.

• Implied probability — the break-even rate baked into the price, before any view of the true chance.
• Simulated true probability — the hidden chance the lab used to settle outcomes.
• Edge — true minus implied. Positive means the model had an advantage on that price.
• Expected ROI — average profit per unit if the true probability is correct, before variance.

Step 3. Check the sample before you believe the number

The panel reports how many events the run generated and how many your selection rule actually took. Those are different numbers, and the second one governs how much you should trust everything else on the screen.

Standard error on an observed win rate falls with the square root of the sample. At 100 selections a measured 55% win rate carries a 95% interval of roughly ±10 points — wide enough to contain everything from a losing strategy to a strong one. At 10,000 selections the same interval is about ±1 point. Nothing about the underlying model changed; only your ability to see it did.

This is why the labs default to large samples and why the statistics pages report Wilson intervals rather than bare percentages. If the interval straddles the break-even rate, the honest reading is 'this run cannot tell', not 'this is slightly profitable'.

Step 4. Then read the discrepancy detector

The discrepancy panel collects every edge the run found — model-vs-market gaps, positive-EV flags, the no-vig fair gap, calibration drift and any synthetic arbitrage — and ranks them by severity. Signals marked look-ahead use the simulation's hidden true probability, so they could never be acted on before a real event. Structural signals like the vig gap are observable from the prices alone.

Reading those two groups separately is the lesson. The observable ones describe real market structure; the look-ahead ones only exist because this is a laboratory.

A useful habit: cover the look-ahead rows with your hand and ask what the run would have told you without them. Usually the answer is 'the market carried a 4% margin and the longshot band was priced worse than the favourite band' — real, structural, and much less exciting than the flagged edges above it. That gap between the two readings is the single most transferable thing the lab teaches.

Step 5. Change one thing and run again

The fastest way to build intuition is to hold the seed fixed and change a single input. Raise the market margin and watch expected ROI fall while the no-vig fair gap grows. Switch to an underdog-only filter and watch win rate drop but payouts climb. Each run is one history among many — change the seed to see how much of what you saw was variance.

Three experiments worth running in order: raise the margin from 2% to 8% and watch how much true-probability advantage it takes just to break even; hold everything fixed and step the seed through five values, noting the spread in observed ROI; then raise the sample size tenfold and watch that spread collapse. The first shows you cost, the second shows you noise, the third shows you why sample size is the price of certainty.

Step 6. Save the seed, and reproduce the run

Every result in Novus Odds is a pure function of its configuration and seed. The same inputs always produce the same tape, the same selections and the same breakdown — on your machine, on ours, today or next year. That is not a convenience feature; it is what separates a research tool from a slot machine.

So when a run surprises you, write the seed down. Then change exactly one thing and run it again. A surprise that survives re-seeding is worth investigating; a surprise that evaporates when you change the seed was variance wearing a costume.