Pinnacle's true line is the price you get after you divide its margin back out of the posted odds, leaving two probabilities that add to exactly 1. The posted price is never fair. A two-way market quoted at -110 / -110 implies 52.38% on each side, which sums to 104.76%, and that extra 4.76% is the vig. Strip it and you get 50% / 50%, the market-implied fair estimate that Pinnacle's price points to once the margin is gone. Every edge calculation downstream, every +EV signal, every graded closing-line-value number, starts from that stripped probability and not from the price on the screen.
This matters because Pinnacle is the reference book that the rest of the market prices against. It runs a low-margin, high-limit model and welcomes sharp action instead of limiting it, so its line absorbs information faster than books that profile and restrict winners. When CLV.gg builds a fair value to test a soft book against, it anchors to the sharp books and exchanges that price most accurately rather than to the soft books that profile winners. The fair number is only as good as the vig-removal math behind it, and that math has more edge cases than the standard "divide by the overround" explanation admits.
What does it mean to strip the vig from a Pinnacle price?
Stripping the vig means converting each side's odds to an implied probability, then rescaling those probabilities so they sum to 1. Start with decimal odds. The implied probability of a decimal price d is 1 / d. For an American price, convert first: a negative price -110 is 110 / (110 + 100) = 0.5238, and a positive price +120 is 100 / (120 + 100) = 0.4545. Add the two raw implied probabilities and you get the booktotal, also called the overround. For a -110 / -110 market that is 1.0476. The overround minus 1 is the vig, here 0.0476, or 4.76%.
The rescaling step is where bettors diverge. The booktotal tells you how much margin is present, but not how it is distributed across the two outcomes. The two common methods, multiplicative and additive, disagree most exactly where it matters most: on lopsided prices where one side is a heavy favorite.
Multiplicative vs additive vig removal: which is correct?
Multiplicative removal is correct as the working default, and additive removal should be reserved for the rare case where you have evidence the book loads margin as a flat probability shift. Multiplicative removal divides each raw implied probability by the booktotal, so both sides shrink by the same proportion. Additive removal subtracts an equal slice of the total vig from each side. On a symmetric -110 / -110 market they give the same answer, 50% / 50%, which is why the distinction is invisible to people who only ever look at coin-flip lines.
The two methods split on asymmetric prices. Take a Pinnacle moneyline of -300 / +260. The favorite's raw implied probability is 300 / 400 = 0.75, the dog's is 100 / 360 = 0.2778, and the booktotal is 1.0278.
| Method | Favorite fair prob | Dog fair prob |
|---|---|---|
| Raw (with vig) | 0.7500 | 0.2778 |
| Multiplicative | 0.7297 | 0.2703 |
| Additive | 0.7361 | 0.2639 |
The additive method takes the same 0.0139 off each side, which pushes the favorite higher and the dog lower than the proportional method. That is the wrong shape. Books do not charge a flat probability tax. They load margin roughly in proportion to each outcome's price, so the proportional method tracks reality better and, just as important, never produces a negative fair probability on a longshot the way additive removal can on a deep favorite. Use multiplicative removal unless you are reproducing a specific vendor's published number that you know used additive.
How do you write the vig-removal math in code?
Here is the multiplicative two-way version, the one you will reach for in nearly every two-way market, with American-odds input.
def american_to_decimal(odds: int) -> float:
return 1 + (odds / 100 if odds > 0 else 100 / -odds)
def no_vig_two_way(odds_a: int, odds_b: int) -> tuple[float, float]:
"""Return (fair_prob_a, fair_prob_b) via proportional vig removal."""
p_a = 1 / american_to_decimal(odds_a)
p_b = 1 / american_to_decimal(odds_b)
booktotal = p_a + p_b
return p_a / booktotal, p_b / booktotal
# Pinnacle NBA moneyline -300 / +260
fair_fav, fair_dog = no_vig_two_way(-300, 260)
print(round(fair_fav, 4), round(fair_dog, 4)) # 0.7297 0.2703
The fair probabilities are the input to everything else. The edge on the dog appears when a soft book's implied probability is lower than Pinnacle's fair probability for that side, because a lower implied probability means a higher payout than the fair odds justify. Pinnacle's fair dog probability here is 27.03%, which is fair odds of about +270. A soft book offering the dog at +240 implies 29.41%, higher than fair, so it pays less than the dog deserves and there is no edge. A soft book offering +320 implies 23.81%, below the 27.03% fair mark, so it pays more than fair and that gap is your expected value. That single comparison, soft implied versus sharp fair, is the spine of +EV detection.
What changes for three-way markets and asian handicaps?
Three-way markets break the two-way shortcut because you now have three raw probabilities to normalize, and the proportional method generalizes directly: divide each side by the sum of all three. A soccer 1X2 priced at 2.10 / 3.40 / 3.60 decimal has raw implied probabilities of 0.4762, 0.2941, and 0.2778, a booktotal of 1.0481, and fair probabilities of 0.4543, 0.2806, and 0.2650. The draw is the side most bettors mis-handle, because the vig on a three-way is spread across an outcome with no clean favorite-dog intuition, and the proportional method is the only one that stays coherent there.
Asian handicaps and push-eligible totals add a different wrinkle: the line itself can refund. A total of 8.0 in baseball pushes on exactly 8 runs, so the two-way no-vig probability is conditional on no-push, and the real fair value has to weight in the push mass. For quarter-lines like 2.25, the stake splits across 2.0 and 2.5, and each half-line needs its own vig removal before you blend them. The clean rule is to strip the vig at the level the book actually prices, not at the rounded line you see in a grid. CLV.gg's pricing pipeline handles the half-line split and the push-eligible conditioning per market rather than treating every total as a clean two-way, because a 10-cent error on the vig compounds into a wrong edge call.
Why does Pinnacle's hold shrink from open to close?
Pinnacle posts its widest margin at the open, when limits are low and the line is least certain, then tightens the hold as limits rise and sharp money refines the number. A market often opens with a noticeably wider hold than it carries at the close, and the line itself can move as the book learns, sometimes by a point or more. This is not Pinnacle being generous. It is the mechanical result of a book that profits from accuracy and volume rather than from trapping square money, so it pays the book to sharpen the price and widen the gates as the event approaches.
That cycle is the reason CLV.gg grades against the close, not the open. Closing line value measures whether the price you got beat the market's final, sharpest estimate. Grading against the open would reward you for beating a soft, low-limit number that nobody could bet into size, which tells you nothing about whether your read was correct. The close is the number the market converged on with real money behind it, and beating it consistently is the cleanest available proxy for a real edge. When the CLV-grading pipeline closes a detected signal, it pulls Pinnacle's final pre-event price, strips the vig with the proportional method described above, and scores the bet's entry against that fair probability. The open is noise. The close is the verdict.