What moves the price
Each bar is one model input's push on the win chance, in percentage points from 50%. Blue favours Team A, orange Team B.
Side by side
Title chances
Every remaining fixture is played out thousands of times using the best XI the model can pick from each squad on the day. Ratings for results and form update inside each simulated season. The top four go into a 1 v 2, 3 v 4 finals series (the BBL, IPL, SA20 and PSL format); ladder ties are split at random.
Squads and availability
List who is in each squad and the model picks the XI for every match: the eleven available players with the highest combined batting and bowling strength, with at least five bowling options, a keeper and no more overseas players than the limit above, batting in their usual positions. Tick Must play to force a selection. Adding a player here takes him out of any other squad. Give a player absence dates if he will leave for Tests or another league; set the chance below 100 if it is not certain, and each simulated season decides once whether it happens.
This team's fixtures
Team ladder by rating
Elo tracks results only. Lineup strength is the latest XI's combined batting and bowling rating, in runs per match against an average side. Select a row for its history.
Grounds
Run environment is runs per innings above or below a typical ground, after allowing for who was batting and bowling. Spin edge is how many runs per innings cheaper spin is than pace there, relative to the bowlers' own ratings. Select a column heading to sort.
Running accuracy
Average log loss of every scored prediction so far. Lower is better; 0.693 is a coin flip. The odds line covers matches that have closing odds in the database.
Running profit on value bets
One unit on each side the model priced above the closing odds. Units won or lost so far.
By competition and month
Scored predictions
The price shown is the last one logged before the match started. First price is the earliest one logged, usually made with an assumed XI.
Logged and waiting for a result
Input weights
Weight on each input per one standard deviation of difference between the sides. Longer bars matter more.
Accuracy by competition
Log loss on matches predicted before they were played. Lower is better; 0.6931 is a coin flip.
BBL: model against closing odds by season
BBL calibration
When the model gave Team A a chance in each band, how often Team A won.
Optional after-the-toss price
The standard price ignores the toss. Choosing who bats first on the Match tab switches to a second set of weights that adds who is chasing and how chasing sides have fared in that competition. This table compares the two on matches where the toss is known. The gain is small and not statistically proven.
Model against closing odds, by competition
Log loss on the same matches for the bookmakers' closing prices, the model, and an equal blend of the two. The return column backs every side the model rated as value, at flat stakes.
BBL betting simulation
Flat one-unit stakes on every side priced above the minimum edge. The 90% range shows how wide the luck band is.