What is graded. Every +EV price the board surfaced is written to the ledger the moment it is shown, before the game is played — so the population settled here is exactly the population users saw. Only finished games appear; nothing is added or removed after the fact, and losers cannot be dropped.

One market, one bet. The same wager offered at several books — CIN +1.5 at eight sportsbooks — is one bet with one outcome, and it counts once, settled at the best price the board showed (the copy the board itself ranked first; taking the best number is the product). Two expressionsof one market collapse the same way: Over 9.5 and Over 10 on the same game are one opinion about one total, and the earlier claim is the one that counts. Otherwise a single game could swing the record several bets' worth on what is really one call. Every per-book price is still recorded and graded: the by-book split below and the CSV export stay at the per-price level, so this headline can be recomputed from the raw rows. Taking the best price raises the return, so the same bets settled at the firstprice recorded for each are shown beside the headline ROI (“at first price seen”).

Never against ourselves.Both sides of one market cannot be better than fair at the same time — the two prices' implied probabilities sum to about 100% — so a “value” reading against a side already claimed on a game means the reference moved, not that a second edge appeared. Once a market is claimed, nothing else on it is surfaced for that game: not the other side, and not another line. This deliberately gives up middles; building middles is a different job from finding one price worth taking. Rows recorded before this rule existed remain in the ledger — nothing is removed after the fact.

Beat the close.Each settled bet is also valued against the reference's closing price for the same line — its last pre-game quote. Beating the close is the faster test of the method: it converges in weeks, where wins and losses need a thousand bets. Bets whose line moved, or with no close inside 30 minutes of the start, are not counted.

Pricing. Each bet risks one unit and settles at the exact American price the board displayed — a +140 winner returns 1.40u. Never a flat −110 assumption: the entire claim is that we found a better number, and normalising prices would erase the thing being measured.

Fair, and the reference.“Fair” is the de-vigged price at Pinnacle (two-way, or three-way — home, away, draw — for a soccer moneyline), a low-margin, high-limit book whose price is the best cheap estimate of true probability. Where Pinnacle hasn't priced an event, the lowest-hold of a small consensus set stands in, and those bets are marked and graded separately: a consensus-referenced edge is weaker evidence.

Denominators. Pushes are excluded from both hit rate and ROI, and so are voids — bets on games that were postponed or cancelled, where the book returns the stake. Hit rate is hits ÷ (hits + misses); ROI is net units ÷ settled bets. There is no single break-even hit rate — every bet carries its own price, which is why return, not hit rate, is the number to judge.

Lines.Spread lines are shown from the picked side's perspective — “TOR +1.5” is Toronto's own line. A different line is a different bet and is never compared or graded across lines.

Expected value is an expectation, not a promise; variance dominates small samples. Past performance does not guarantee future results. This is market research, not financial advice.