Findings / Grading the books
How accurate were the market sources?
Proportional devig (implied prob of side A = (1/a)/((1/a)+(1/b)) for decimal odds a,b), scored by Brier score against the realized outcome, on the shared-game subset every compared book quoted.
We graded the bookmakers on their own devigged accuracy: Pinnacle is barely sharpest on tennis match-winner (0.1975 vs Bet365's 0.1980), but on soccer over/under the books are a statistical dead heat -- no public book is meaningfully beatable by another on these markets.
Tennis -- match winner
Market: match winner — outcome: eventual winner — shared subset n=33,664
| Book | Brier | n |
|---|---|---|
| Pinnaclesharpest | 0.1975 | 33,664 |
| MarketAvg | 0.1977 | 33,664 |
| Bet365 | 0.198 | 33,664 |
Scroll horizontally for all columns.
In this published subset, Pinnacle has the lowest Brier score; the difference is 0.0005.
Soccer -- over/under 2.5 goals
Market: over/under 2.5 goals — outcome: target_over25 — shared subset n=15,112
| Book | Brier | n |
|---|---|---|
| Pinnaclesharpest | 0.2395 | 15,112 |
| MarketAvg | 0.2395 | 15,112 |
| Bet365 | 0.2396 | 15,112 |
Scroll horizontally for all columns.
Pinnacle and the market average are both 0.2395. Bet365 is 0.0001 higher. On this market the books are statistically indistinguishable.
How close they all are
Across both sports the spread between the sharpest and least-sharp source never exceeds 0.0005 Brier. The published sources are closely grouped on these markets -- the honest headline is the closeness itself, not a ranking. Being sharp is also not the same as being first: this exhibit measures accuracy at the timestamps captured, not who posts a line first.
Observation window
Tennis: data/domains/tennis/odds.parquet, 33,952 rows before the shared-subset filter -- committed odds corpora (decade-plus per sport), not a rolling cache.
Soccer: data/domains/soccer/odds.parquet, 16,322 rows before the shared-subset filter -- committed odds corpora (decade-plus per sport), not a rolling cache.
Confounds / notes
- accuracy is not the same as price leadership -- a book can be sharp yet never post first
- snapshot-timed closes, not a continuous feed
- proportional devig is one method among several -- it shifts absolute Brier levels, not the book ordering
- each book graded only on the shared-game subset all compared books quoted
- tennis Brier is winner-side only (the loser side is the complement by construction)
- soccer draws fold into the two-way over/under market as a non-goal outcome, not a separate line
- Results report published source-level Brier values in the stated observation windows.