Reading Book Accuracy
What it is
The Book Accuracy page summarizes how each sportsbook actually performs relative to its quoted prices, by sport and market. Reading it correctly tells you which books are sharp anchors (Pinnacle, Circa) and which books are profitable to attack.
When to use it
- When deciding which books to open or fund.
- Before sizing up on an edge: confirm the book offering it is actually soft on that market.
- When debating whether an edge is real or a trap.
How to read it
The columns explained
| Column | What it means in practice |
|---|---|
| Sample (n) | Number of graded lines. Below 1,000 = noise. 10,000+ = trustworthy. |
| Win-rate by Book | Win rate of betting the book's offered price. If the book quotes +110 favorites that win 47% of the time, that price is 1.4% off market. Bigger gaps in your favor = softer book. |
| Calibration Error | Mean absolute difference between book implied prob and realized prob across probability buckets. |
| ROI vs Book | The historical ROI of always taking the side the book mispriced. The actionable number. |
| Tier | S (sharpest, near zero error) through D (softest, large errors). |
| Trend | Last 90 days vs lifetime. Books recalibrate; watch for soft books getting sharper. |
How to interpret colors
- Green tiers (S/A): treat their quotes as sharp anchors. Edges at these books are usually arbs or stale lines, not value.
- Yellow tiers (B): occasional value, especially on niche markets where they have not invested in risk teams.
- Red tiers (C/D): your target zone. Consistent calibration errors in your favor.
What "calibrated" actually means
A perfectly calibrated book quotes 60% favorites that win exactly 60% of the time, 70% favorites that win exactly 70% of the time, etc. Plot implied prob on the x-axis, actual win rate on the y-axis. Sharp books sit on the diagonal. Soft books deviate.
Worked example
Book: Fliff Sport: NBA | Market: spreads | n: 31,400 Win-rate vs implied gap: -2.8% (favorites under-perform by 2.8 points) Calibration error: 3.1% ROI taking the opposite of Fliff's lean: +1.9% Tier: D Trend: stable
Translation: Fliff systematically overprices favorites on NBA spreads. Taking dogs against Fliff has been historically profitable. This is a real, exploitable, sample-sized edge. Tier D confirms it.
Book: Pinnacle Sport: NBA | Market: spreads | n: 180,000 Win-rate vs implied gap: +0.1% Calibration error: 0.4% ROI flat-betting Pinnacle: 0.0% Tier: S Trend: stable
Translation: you cannot extract systematic edge from Pinnacle by following the price. They are the price. Use them as the anchor, not the target.
Common mistakes
- Reading the table globally. Calibration is per-sport and per-market. A book sharp on NFL can be soft on golf.
- Trusting tiny samples. <1k bets is noise. Wait for the n to grow or look at the lifetime number.
- Forgetting recency. A book that was D-tier in 2023 may be B-tier now. Check the 90-day column.
- Confusing calibration error with vig. Vig is the cost of every bet. Calibration error is whether the prices are correct apart from vig.
- Acting on book tier without an actual edge. Soft book + no edge = still no bet. The tier tells you where edges tend to live, not that every line is a bet.