Auspex prices NHL moneylines and NBA player props with models built on every game of the last several seasons. Each pick is posted before the game with a timestamp and the book, and stays up win or lose, graded at the posted price.
House rules. Picks lock before puck drop or tip-off and get a timestamp. They're graded at the posted price and book, juice included, and the closing price is logged so you can see whether each one beat the market. Nothing gets edited or deleted. Backtests only use games the model hadn't seen.
Picks lock before the first puck drop. Highlighted games are bets; the rest show the model's lean.
The closing line, taken about 15 minutes before puck drop, is the sharpest price the market produces. Beating it consistently is the clearest sign of a real edge, even in weeks when results run cold.
| Date | Pick | Stake | Posted | Close | CLV | Result | Units |
|---|
Every team rated on the danger of the chances it creates and allows, at even strength and on special teams.
Confirmed goalies, projected lines, injuries, rest and travel adjust each game.
A pick needs the model's price to beat the no-vig market by at least 5 points.
Each game's props lock about 50 minutes before tip, after the injury report. Every pick is 1 unit at the best price we found, and the book is named.
| Date | Pick | Price | Book | CLV | Actual | Result | Units |
|---|
The 2025-26 test, on games the model never saw.
| Where you bet | Bets | ROI | 95% range |
|---|
Most ideas didn't beat the market. They're listed here anyway.
| Test | Result | Verdict |
|---|
| Stat | Lines | Books | Model alone | Final |
|---|
The numbers behind the picks, including the parts that don't flatter them. Every backtest below uses only games the model never saw while it was being built.
When the NBA props model says an over hits 40% of the time, it should hit about 40% of the time. Each dot is a group of lines; on the diagonal means perfectly calibrated. The books' own no-vig prices are shown for comparison. Unseen 2025-26 games.
Return on investment grouped by how big the model's edge was when the bet was made. It isn't a perfect staircase, and a real backtest rarely is.
Edge in percentage points over the no-vig market, 2022-26
Expected value at the best available price, unseen 2025-26 games
Log loss measures how good the probabilities are (lower is better). Beating the market's log loss is hard, and most seasons the model doesn't; the edge comes from picking the games where it disagrees most.
| Season | Games | Log loss: model | Log loss: market | Bets | ROI |
|---|
Everything on this site comes from these files. Download them and check my work.
In ancient Rome an auspex watched birds in flight and read what they signalled before any big decision. The word literally means bird-watcher. My version reads every shot and every play instead, prices the game or the prop, and posts the call before the line moves.
Every pick is free on this page while the live record builds. After a month of live NBA results, members will get instant alerts when picks lock and the best book for each bet.
See today's picksPicks are information, not financial advice. Past results, including backtests, don't guarantee future results. Bet only what you can afford to lose, and only where sports betting is legal for you. 18+ (19+ in most of Canada, 21+ in most US states).
If betting stops being fun, help is free and confidential: 1-800-GAMBLER in the US, ConnexOntario 1-866-531-2600 in Ontario, or your provincial helpline.
Graded at the posted price and book. NBA props that are voided because the player doesn't play count as no bet. 1 unit = 1% of bankroll.