Expected value: the average result if you could make the same bet a thousand times. Positive means that average is a profit. It is the only thing this site selects for, and this page is what that means and what it costs.
Every price implies a chance. That is all it is. Strip the costume off and you can compare it against what you actually think — and a bet is only worth taking when you believe the real chance is higher than the number in the implies column.
| decimal | implies | $100 returns | |
| −200 | 1.50 | 66.7% | $150 |
| −110 | 1.91 | 52.4% | $191 |
| +100 | 2.00 | 50.0% | $200 |
| +150 | 2.50 | 40.0% | $250 |
| +900 | 10.00 | 10.0% | $1,000 |
A coin flip is a 50% shot. At +100 that is exactly break-even — zero EV, forever. At −110 it is −4.55%: you lose four and a half cents on the dollar every time, and being right half the time does not save you. At +110 the same coin is +5.00%. Nothing changed about the coin. Only the price.
Not from picking winners. From finding a price that implies a chance you think is wrong. Here is the best-priced side anywhere on the board of Wednesday, September 2:
That is $100.23 back for every $100 risked, on average — +0.23% EV. It says nothing whatever about whether this particular bet wins.
Both sides of every market are priced to add up to more than 100%. The overflow is the house’s margin, and on this board it runs about 4.65%. You have to be better than the market by more than that before you have made a penny.
Which is why so little clears it. Of the 54 sides priced on Wednesday, September 2, taking the best number available at any of the books, 53 were still −EV.
| 01Today only | SAME_DAY_ONLYTrue | An edge measured four days out is mostly noise, and it can never land inside the closing-line window, so it could never be graded honestly. |
|---|---|---|
| 02Drop stale books | STALE_MINUTES25 min | A book that has not moved in this long is not quoting today’s market. Stale prices are the single biggest source of edges that are not there. |
| 03Enough books | MIN_BOOKS6 | Below this there is no consensus to measure against, only noise wearing a consensus’s clothes. |
| 04One number at a time | —consensus point | For run lines and totals, only books sitting on the most common number are compared. Measuring −1.5 against −2.5 is not a measurement. |
| 05Strip the vig | DEVIG_METHODpower | Both sides of every book’s quote are converted to a vig-free pair. The power method corrects for books loading extra margin onto longshots; the naive method would fill the card with +300 dogs. |
| 06Price band | MIN_PRICE / MAX_PRICE-300 to +185 | Taste, and admitted as taste. Outside this band are prices this site does not want to be in the business of recommending. |
| 07Leave one out | —5 others | The fair price averages every book except the one we would be betting. Include it and the outlier drags the consensus toward itself, manufacturing the very edge being measured. |
| 08Expected value | MIN_EDGE_PCT2.5% | The gap between the best price and the fair price, as a ratio. This is the +EV number. |
| 09Probability points | MIN_EDGE_PP1.0 pp | And the same gap in percentage points. Both have to clear. Ratios flatter longshots: a +200 dog reaches 5% EV on a 1.7-point edge while a −200 favourite needs nearly three. |
| 10Sanity ceiling | MAX_EDGE_PCT12.0% | Above this it is not an edge, it is a data problem — a stale line, a mismatched number, a book quoting a different game. Logged and thrown away rather than posted. |
| One side per game | — | Two sides of one event is correlated risk dressed up as diversification. |
|---|---|---|
| Per league | 3 | No league can take over the card, however good its slate looks. |
| Per day | 6 | A short card is a feature. It stops a bad day becoming fifteen losses. |
| Stake | 1 unit | Flat, on every play, always. Sizing by confidence would make the record a story about the sizing. |
It was +EV and it still did not go up. Being on the right side of the price is the beginning of the test, not the end of it.
0 playsfrom 54 priced sides on Wednesday, September 2
An empty card is not the system failing. It is the system working on a night when no book was far enough out of line to be worth taking. Anyone posting five plays every single day is not finding five edges every single day. They are filling a quota.
Our fair price being right. Every number on this page compares the market against our estimate. If the estimate is wrong, “+EV” is decoration. That is not a hypothetical: the strikeout projection on this site has been measured against 86 finished starts and it is 0.20 strikeouts less accurate than the line it was trying to beat. The game markets have not been measured yet.
So check us on the closing line. Whether a play beat the number the market settled on is a verdict on the bet rather than on the bounce, and it converges in tens of plays where profit needs thousands. It is the first figure on the track record page for that reason, and if it does not turn positive over a few hundred plays then nothing on this page is worth anything.
Nothing here is discretionary. No override, no gut call, no tidying the list afterwards. Every threshold above is read out of the configuration when this page is built, so changing a rule changes this page in the same commit.