How it works

how to read the site, and every term it uses

How to read a prediction

Cholet Basket v Elan Chalon · LNB Pro A · 17:30 WAT

under 177.5 · MEETS CRITERIA · SportyBet 1.37 · model estimate 76%

  1. under 177.5: the prediction. It wins if both teams together score 177 or fewer points, overtime included.
  2. Meets criteria: the tier. The other tier is Lean, an exploratory call below the bar. The two are counted separately.
  3. SportyBet 1.37: the bookmaker and the odds the prediction was priced at when it was published.
  4. 76%: the model's estimated chance. Over many predictions, the 76% ones should win about 76% of the time; Results tracks whether they do.
  5. details: opens the prediction record: when it was published, the prices it saw, how it was graded, and any correction.

Predictions

Prediction
An over or under call on the combined points of both teams. We aim to make it about 60 minutes before the game starts; each record shows when it was actually published. It is recorded, not staked, and graded from the final score.
Meets model criteria
A prediction that cleared the model's bar: at least a 75% estimated chance and at least 3% expected value at the bookmaker's price. These are rare.
Lean
When nothing clears the bar, the model still records its single best call for the game (estimated chance at least 60%) so the record keeps growing. Leans are exploratory and are counted separately from predictions that met the criteria.
Ticket
Up to three games starting within 90 minutes of each other, combined into one bet with combined odds of at least 1.10 and below 3.00. It is sealed 10 minutes before the first of them starts, and each game is used in one ticket only. Its chance is the product of its legs' chances: three 85% legs make a 61% ticket. When no combination fits, the record says no ticket. A ticket wins only if every leg wins; a leg that lands exactly on its line is dropped.
Model estimate
The chance the model gave the prediction when it was published. It comes from the reference market: Pinnacle's ladder of lines fitted to a curve, then pulled a quarter of the way toward the bookmaker's own price, because the largest disagreements are partly fitting error. No statistics model enters it: on its own, ours did not beat Pinnacle. Results compare it with what actually happened.
Expected value (EV)
Estimated chance × odds − 1: the average return per unit if the estimate is right. +3% means 3 on every 100; −1% means the price was slightly against the call.
Fair total
The total the reference market implies, fitted from its full ladder of lines. The model prices any line from it.
Reference market (Pinnacle)
The model's chosen benchmark: Pinnacle's prices are widely used by analysts as a reference because it accepts large bets on low margins. We do not claim its price is always right; it is the anchor the model is tested against.
Ladder
A bookmaker's full set of alternate total-points lines for one game (for example 165.5, 167.5 … 185.5, each with its own price). "Main" means only the single main line was available.
Book line
The Nigerian bookmaker's main total for the game. It is not a prediction; the prediction says over or under.
Soft bookmaker
SportyBet, Football.com, Bet9ja: bookmakers with higher margins whose prices can lag the reference market.
Margin
The bookmaker's built-in edge. Odds of 1.90 each way add up to about 5% in the book's favour.
Scheduled
A game on the books more than 90 minutes away. The row says when its prediction is due.
Checking odds
The 90 minutes before start: bookmaker prices are read every 5 minutes, the reference ladder is read once, and the prediction is made about 60 minutes before start (or when starters are confirmed in the NBA, WNBA and EuroLeague).
No prediction
The model looked at the game and nothing reached the lean threshold, or no reference price was available. The reason is shown.
Won, lost, push
Graded from the final score, overtime included. A push lands exactly on the line.
Closing price (CLV)
How the published price compares with Pinnacle's last price before start. Positive means the market moved toward the prediction. Over hundreds of predictions it is an early sign of a real edge. Closing prices are collected overnight.
Calibration
Of all predictions the model rated, say, 75–90% likely, how many won. A calibrated model's 80% predictions win about 80% of the time. It needs a few hundred results to mean much.
Simulated return
What the qualified predictions would have returned at the stake the model recorded (quarter-Kelly on a ₦2,000,000 paper bankroll, capped at 2%). Leans carry no stake. Nothing is actually staked.

Live games

Live
The game is in progress. Score, quarter, time left and quarter scores come from the bookmaker's feed every 30 seconds; between reads the clock counts down in your browser.
Clock stopped
Time left did not move since the last read: a timeout, foul or review. The clock freezes until play resumes. A stop that starts between two reads shows at the next read, so the clock can run a few seconds ahead and then snap back.
Break, half-time
A quarter has ended (time left 0:00). Half-time follows the 2nd quarter; the other breaks follow the 1st and 3rd.
Overtime
Five-minute extra periods (OT1, OT2…) when the score is level after the 4th quarter. Totals include overtime.
Interrupted
The bookmaker reports the game as stopped or suspended. We show its own word for it.
Live scoreboard
The large panel with quarter scores, pace and expected totals. It opens under a live game that has a prediction; select any live game (click, or Enter on the keyboard) to open its scoreboard under it, and select it again to close it.
Chance the prediction wins
Two live estimates. Model: the points on the board plus the pre-game fair scoring rate for the time left, including the chance a close game goes to overtime. Bookmaker: read from the bookmaker's live over/under ladder, its own running forecast of the final. Both are recorded at every read so we can later check whether an 85% really wins 85% of the time.
Expected half-time and final total
For every live game: the points on the board plus the pre-game scoring rate for the time left (and, for the final, the chance of overtime). It starts from the fair total when the model priced the game, otherwise from the bookmaker's pre-game line, which is a rougher guide. After half-time the actual half-time total is shown instead.
Looking good, too close, in trouble
The prediction's live chance is 80% or more, between 20% and 80%, or 20% or less. The bookmaker's estimate is used when available, otherwise the model's.
Won already, lost already
Once the total passes the line, an over has won and an under has lost, whatever happens next.
Final
The final score is in. Predictions and site calls are graded from it as soon as the game ends, and the overnight run re-checks every grade.
Odds … ago, score read … ago
The header shows when bookmaker odds were last read (and, on wider screens, live scores). Each live scoreboard shows when its own score was read. Odds older than 2 hours show as stale.

Crowd Wisdom

Call
What a prediction site published for a game: a winner, a spread, or a total.
Winner
Which team wins the game (moneyline).
Spread
The Asian handicap in the site's own line. "PAOK −7.5" means PAOK must win by 8 or more; "Manresa +3.5" means Manresa can lose by up to 3.
Total
Over or under the combined points of both teams.
Strongest call
For each game, the market where the most sources agree. On a tie, total comes first, then spread, then winner.
Support
How many sources back the leading side. Single source means only one site covers that market, which is not agreement.
Weighted share
With two or more sources: each counts by its hit rate on that market, smoothed toward 50%. It is not a probability of winning.
Covering, not covering
While a game is live: whether a spread call is currently on the right side of its line. Totals show points still needed; winners show who is ahead.

How data is updated

  • Bookmaker odds: every hour, every 5 minutes in the 90 minutes before a game.
  • Live scores: every 30 seconds while a game is in play.
  • Grading: as soon as a game's final score is in, then re-checked overnight.
  • Closing prices and prediction-site records: overnight and every two hours.