How Our Live EV Tracker Works — Methodology

Every expected-value figure on this site comes from one model, applied identically to every game. This page documents it completely, including the parts that are uncertain, so you can decide how much weight to give the output. Try it on the jackpot tracker or any game's expected-value page.

The core formula

For one ticket:

EV = Σ over divisions of (prize × probability) − ticket price

Both inputs need care.

Probabilities: derived, never copied

Division probabilities are computed from each game's published matrix using exact integer combinatorics — BigInt fractions, reduced, converted to floating point only for display. No probability on this site is copied from an operator's marketing page.

Where an operator publishes its own odds, we show them alongside ours as a cross-check, and an automated test fails the build if any derived figure disagrees with a published one by more than 2%. The method is set out in how lottery odds are actually calculated and in full on the methodology page.

Prizes: only what is published

This is the assumption that most distinguishes our numbers from other sites'.

• Fixed prizes (US Powerball's $1,000,000 for Match 5, UK Thunderball's whole table) are used directly, because the operator publishes them as fixed amounts.
• Pari-mutuel divisions — those paying an unknown share of a pool — are excluded from the sum entirely unless you supply an amount yourself.

We do not model, estimate or infer pool-share payouts. That means for games like Australian Saturday Lotto, where most divisions are pool shares, our EV output is explicitly a lower bound and is labelled as such on screen. A number that says "at least this much" is honest; a number built on an invented average is not, and inventing them is how competitor sites produce EV figures that look precise and are unfalsifiable.

The jackpot: three adjustments

1. Cash value. The advertised jackpot on annuity games is not a present sum. Enter the cash value if you want a present-value answer; enter the advertised figure if you want the annuity view. We do not silently convert, because the discount rate is the operator's and it varies. Background: annuity vs lump sum.

2. Tax. Applied as a single effective rate you control, multiplying every prize by (1 − rate). Real tax is progressive, jurisdiction-specific and different for lump sums and annuities, so a single rate is a simplification — a deliberate one, since the alternative is pretending to model 43 tax codes. The sourced per-country rules are in lottery tax by country.

3. Co-winner sharing. The most important adjustment and the one most models omit.

If S tickets are sold and the jackpot probability is p, we treat the number of other winning tickets as Poisson with mean λ = S × p. Conditional on your ticket winning, your expected share of the jackpot is:

(1 − e^(−λ)) ÷ λ

This is exact under the model's assumptions, and it is applied to the jackpot only — lower divisions with fixed prizes are unaffected by sharing, which is precisely why roll-downs can be exploitable.

At λ = 0 the factor is 1 (no sharing). At λ = 2.05 — roughly a billion-dollar Powerball — it is 0.425, meaning you expect to keep about 42% of the prize you win.

The break-even jackpot

Setting EV to zero and solving for the jackpot gives:

J* = (price − other divisions' EV) ÷ (p₁ × share factor × (1 − tax))

This is exact for a fixed sales estimate. Its weakness is structural and we flag it on every page that shows it: sales rise sharply with jackpot size, so the share factor falls as J rises, pushing true break-even higher than the formula reports. Treat the output as a floor, not a target. See the break-even jackpot calculator.

The assumptions, ranked by how much they could be wrong

1. Tickets sold. Much the weakest input. Operators rarely publish sales per draw promptly, and sharing is highly sensitive to it. A factor-of-two error in sales moves the share factor materially. Our tracker uses operator-entered estimates and timestamps them.
2. Independence of ticket combinations. The Poisson model assumes other players' combinations are spread uniformly. They are not — birthdays and patterns cluster heavily (how prize pools are split). If your numbers are popular, real sharing is worse than the model says; if unpopular, better. The number-sharing risk checker flags the common patterns.
3. Cash value ratio. Varies with interest rates; we ask rather than assume.
4. Effective tax rate. A single rate standing in for a progressive system.
5. Excluded pool divisions. Biases our EV downward, deliberately.

What the tracker does not do

• It does not scrape jackpots. Figures are entered from operator sites and shown with the date they were entered. A stale figure is visibly stale rather than silently wrong.
• It does not recommend playing anything. Ranking games by EV identifies which is least bad. At normal jackpots, all of them are negative.
• It does not model your risk. Positive expected value says nothing about variance or ruin — see Kelly and why +EV jackpots aren't worth chasing.

If you find an output you believe is wrong, the inputs are all on screen and the game's operator source is linked from its page. Check us.