How SPIN XI rates a cricketer
No rating here is copied from anywhere, and none is anybody’s opinion. This is how 42,165 player-seasons were scored, and the three corrections that were found by measuring rather than by eye.
A rating in SPIN XI is a number between 40 and 99 attached to one player in one season of one competition — Chris Gayle in the 2011 Indian T20 League, not Chris Gayle in general. It is computed from what he did in the matches that season, delivery by delivery, and from nothing else.
Distance from the middle, not rank
A player-season’s rating is how far it stands from the middle of the field, in standard deviations, against everyone who played the same format. It is deliberately not a percentile.
Percentile was the first attempt and it fails for a reason worth stating: rank cannot express distance. It says a player was better than eighty-five per cent of the field but never whether by a little or a lot. Under that method the great and the merely good landed a point apart, and 669 players sat at 90 or above — which makes 90 meaningless. Measuring distance instead, thirteen players reach 95 across the entire archive.
What the numbers are
- Batting — runs per innings, scaled by strike rate against the format’s par. Scoring 40 at a run a ball in a Twenty20 is not the same achievement as scoring 40 at a run a ball in a Test, and the par figure is what makes them comparable.
- Bowling — wickets per match, scaled by economy against the format’s par. Wickets alone rewards the bowler who buys them; economy alone rewards the bowler who never threatens.
- The three figures on a card are the record, not decoration. A batter’s
BAT / CONS / SRis his batting rating, his average and his strike rate. A spinner’sGUILE / WKT / ECONis his bowling rating, his wickets per match and his economy.
A quick bowler’s headline figure is THREAT rather than PACE. Ball-by-ball data records no delivery speeds at all, and calling the number PACE invited precisely the question it could not answer.
Three corrections, all found by measuring
Each of these was a real fault in a shipped version, and each was caught by checking outputs against cricket rather than by reading the code.
Competitions are graded against each other
The first version treated every non-international competition as equivalent, so a hundred in a weak league counted like a hundred in the Indian T20 League. John Davison — five Canadian seasons — came out the best player in the game at 95, above Tendulkar, Dravid, Ponting, Kallis and Sehwag, who all sat at 91.
Competitions are now graded from the cricket itself: a two-way model over players and competitions, solved by alternating between them and tied together by the people who appear in several. The output is not an opinion about which league is strongest — it is a measurement of what the same players did in both. It reports the SA20 and the Indian T20 League as the hardest places to bat, and the T20 Blast and the Bangladesh Premier League among the easiest.
Weak isolated cricket is less evidence, not just less value
Discounting a weak competition’s value could not fix the problem on its own, and this is the subtle part. A player who never leaves that cricket has nobody to be compared with, so the model believes his record because nothing contradicts it. Fourteen wickets in eight matches survived a forty per cent haircut and still came out in the high eighties.
Eight untested matches is closer to three matches’ worth of knowing, and three matches is not enough to be rated above Tendulkar on. Treating thin evidence as thin evidence moved Leverock from 94 to 84, Dhaniram from 87 to 79 and Davison from 95 to 88.
Short seasons regress to the mean
Measured in matches rather than in deliveries, because impact is a per-match rate: it is a short season that inflates it, not a short spell. A player with two brilliant games is not rated as though he had a brilliant season.
Career best, as against one summer
The game offers two ways to rate the same player. Season form uses the card for the season on it — a lean year is a lean card. Prime rates every player at his career best, whichever season that was.
A prime rating is not simply the highest number a player ever posted. One outstanding season out of ten is more likely to contain luck than one out of two, so a peak is pulled toward what the player typically was, and pulled less the longer the career it sits in. Opponents are rated the same way, so a prime draft is a prime field and the tournament stays a contest.
What a player was, as against what he did
Two different questions, kept in two different places. A player-season holds what he did that summer. The player record holds what he was across a career — his primary role, his career-best rating, and the side he is most associated with.
Roles are not inferred from figures where anybody has stated them. Each player’s role and bowling style are read from his own Wikipedia entry, joined by Cricsheet identifier — 2,077 bowlers have their pace or spin recorded that way rather than guessed. It matters because a medium-pacer and an off-spinner can have identical economy rates, and nothing in the ball-by-ball data says which is which.
Checking it
Every rule the archive is supposed to obey is checked against the whole archive rather than a sample: all 42,165 player-seasons, nineteen rules each, with any failure printed. The corrections above were found because outputs were compared against cricket, which is the only test that catches a rating being confidently wrong.
More on how this is built
- How the match engine works — SPIN XI plays every delivery between the man bowling and the man facing. The score is what comes out, and so is the result. This is how.
- The archive: where every number comes from — SPIN XI is built from 15,270 ball-by-ball match records published by Cricsheet. This is what is in the archive, how it is assembled, and what is deliberately not in it.
- The best seasons in the archive — The highest-rated player-seasons across 15,270 matches of Test, one-day and Twenty20 cricket — and why all-rounders dominate the top of the list.
- Which Twenty20 league is hardest to play in? — Not an opinion. A measurement of what the same players did in different leagues, from ball-by-ball records of every match.
- How to draft a side that actually wins — What ten thousand simulated seasons say about building an eleven in SPIN XI: where the rating misleads you, and what to take when the squad is thin.
- About SPIN XI — A free cricket drafting game built on the public ball-by-ball record. What it is, who made it, and what it is not.