Cricket
A plain-English, but complete, description of how every number on the cricket pages is calculated, taken from what the site actually computes today. Other sports.
Selection Room’s impact score is built on the published principles of the PCA Most Valuable Player rankings, the system designed by England’s professional cricketers (with analytics partner CricViz) to value players more fairly than batting and bowling averages. The Professional Cricketers’ Association publishes the principles behind it in Facts, figures, formulas: the PCA MVP. Selection Room is not affiliated with or endorsed by the PCA.
The table below takes each published PCA principle in turn and says exactly how, or whether, Selection Room applies it to a league’s own scorecards.
| PCA principle | How Selection Room applies it to league scorecards | Detail |
|---|---|---|
| More value on top-order runs | Applied, in the direction a league’s own data supports. A run is weighted by what runs are worth at the position it was scored from, using the division’s own runs and dismissals at each position — the same two numbers that price a wicket. In league cricket those show runs are far scarcer down the order, so the weighting lands the other way round from the professional game: a run at No. 10 is worth more than a run at No. 3, just as a No. 3’s wicket is worth more than a No. 10’s. Damped to a quarter of the gap, clamped to 0.75–1.35, and normalised so the division’s average run is worth exactly 1.00. | What a run is worth Selection signals |
| Scoring quickly is rewarded | Applied. Runs are multiplied by strike-rate-vs-par: the batter’s strike rate against the division’s own par strike rate, damped to a quarter of the gap and held between 0.85 and 1.15. | Batting |
| Wickets carry more value in high-scoring games | Applied. The same match-conditions measure is read in both directions: runs are worth more in a low-scoring match, and a dismissal is worth more in a high-scoring one. The wicket multiplier is exactly one divided by the run multiplier, so it inherits the same measurement and the same bounds (0.74 to 1.33). It scales the whole dismissal, so the bowler’s and fielder’s shares of it are unchanged. | Match conditions Pricing a wicket |
| Top-order wickets are more valuable | Applied. A wicket is worth what the dismissed batter’s batting position averages in that division this season, so dismissing a number 3 is worth more than dismissing a number 11. | Pricing a wicket |
| Low economy is rewarded | Applied. A spell is credited with the runs it prevented against the division’s own par economy, over the overs actually bowled: par × overs − runs conceded, floored at minus par per over. Runs prevented are priced 1:1 with runs scored, and the term sums to zero across the division, so it ranks bowlers against their own division without inflating the league. Economy remains on the board, and an economy of 3.5 or under is still one of the two ways a bowler qualifies for the “more overs?” signal. | Runs prevented Selection signals |
| Maidens are rewarded | Applied. Each maiden over is worth the division’s par runs per over — what an over conceding nothing actually saved — and the rest of the spell is priced on what those overs conceded, so the runs a maiden saved are counted once and never twice. Maidens also now have their own Md column on the contribution board. | Runs prevented |
| Catches, run-outs and stumpings count as fielding | Applied. A catch (including a caught behind) or a stumping pays the credited fielder or keeper 30% of the dismissal’s value and the bowler 70%. A run out pays the named fielder or fielders 100%, split evenly, and the bowler nothing. | Fielding |
| No separate wicket-keeping component, and no penalty for byes | Applied the same way. A keeper earns the fielder share (30%, bowler 70%) for a caught behind or a stumping, exactly as for any other catch. Byes are not charged to anyone. Keeping recorded on initials-only scorecards is undercounted. | Wicket-keeping |
| Captaincy bonus | Not used, deliberately. The score only prices runs actually scored and dismissals actually recorded. | Fairness principles |
| Winning bonus | Not used, deliberately. The team’s result is already counted in the league points and standings, so adding it to each player’s score would count it twice. | Fairness principles |
There is no all-rounder bonus: batting, wickets, fielding and runs prevented are each scored on their own terms and simply added together. A player who does two of them well is worth the sum of the two, not more.
What is different from the professional game. League cricket has no ball-by-ball data, so where a professional model can compare each ball with what was expected, Selection Room builds its expectations from division-level baselines (runs per wicket, par strike rate, par economy, and what each batting position averages) and from the match’s own scorecard (both innings’ runs and wickets). One consequence is worth stating plainly: the PCA weights top-order runs up, and on this league’s own numbers the same principle — price a run by how hard runs are to come by where it was scored — weights lower-order runs up instead. The site follows its data rather than the professional game’s coefficients.
Every game a player appears on the scorecard for a team earns one impact figure: batting + wickets + fielding + runs prevented, each priced as described below, and all four in the same currency: runs. Impact per game is the total divided by the number of those games, so a game in which the player did not bat, did not bowl and took no catch still counts as a game.
Nothing here is a chosen number. Every price comes from the division’s own season — its runs per wicket, its par strike rate, its par economy, and what each batting position scores — recomputed as results land. Where a figure has to be bounded so one freak match or a 40-innings sample cannot swing a season, the bound is stated with the term, and the two weightings that redistribute value (position weight, runs prevented) are built so the division’s total is unchanged.
The board shows it twice: Season (every game for this team this season) and Last N (the form window, below). The board ranks by Last N. Only games for the team whose page you are on count; a player’s games for the club’s other teams appear on those teams’ own pages.
Batting value for an innings is runs × match scarcity × strike-rate-vs-par. A did-not-bat innings contributes nothing, and nothing against the player.
Match scarcity has its own section too, because the bowling side reads the same measurement backwards: match conditions.
The PCA weights top-order runs more heavily. Selection Room applies the principle behind it — that a run is not worth the same everywhere — but takes the weighting from the league’s own data rather than importing a professional game’s numbers, and in this league the data points the other way.
For each division and season the site already measures, for every batting position, the runs scored there and the dismissals there. Their ratio is the position baseline: what that slot typically returns. The weight on a run is then:
This is the same logic that prices a wicket, read from the other side of the ball. Dismissing a No. 3 is worth more because that slot scores more; scoring a run at No. 10 is worth more because that slot scores less. The exact weights the page is using for its own division are listed in the impact score’s own i popover on the board.
It does not change the “bat them higher?” signal, which still compares a lower-order batter’s average with the unweighted baselines of the positions they actually batted.
One measurement covers both sides of the ball: match scarcity, the division’s runs per wicket divided by that match’s own runs per wicket (both innings’ runs over both innings’ wickets), held between 0.75 and 1.35.
Using the reciprocal rather than a second measurement means the two sides can never disagree about a match, and the bound on one is the bound on the other. A match in which no wicket fell floors scarcity at 0.75, which is the strongest wicket multiplier: both sides read that pitch the same way. The multiplier scales the whole dismissal before it is split, so the 70/30 bowler-fielder shares and the run-out’s 100% are untouched.
Runs prevented (below) are deliberately not adjusted by it: a bowler’s own economy is part of what makes a match high- or low-scoring, so adjusting it by the match’s scoring rate would price the same thing twice.
A wicket is priced at the dismissed batter’s position baseline: the batting average (runs divided by dismissals) of that batting position across every innings in that division this season. That value is then shared by how the batter was out:
That whole value is then scaled by the match’s own conditions, so the same dismissal is worth more in a high-scoring game than in a low-scoring one. The shares above are applied to the scaled value, and still sum to exactly 100% of it.
Two fallbacks keep every wicket priced. A batting position with no baseline yet is priced at the division’s flat runs per wicket instead. And when a bowling figure records more wickets than can be matched to a named dismissal (mostly scorecards recorded by initials only), each unmatched wicket is priced at 100% of the division’s runs per wicket, with no fielding share, because there is no dismissal text to split.
Wickets are not the whole of a bowler’s score: the overs themselves are priced separately, in runs prevented.
A bowler who concedes nothing has done something a scorecard records and an average hides. Selection Room prices it in the same currency as everything else — runs — against the division’s own par economy: every run conceded by every bowler in that division this season, divided by every over bowled in it. In 2026 that is a little under 4 runs an over in every division.
A spell is split into two parts that cannot overlap:
The two add up to par × overs − runs conceded, the spell’s whole runs-prevented figure. That identity is the guarantee against double counting: a maiden is rewarded as its own named term, and the total is still just what the spell saved. Ten overs for 20 in a division whose par is 4.00 is worth 20 runs-equivalent whether those overs included three maidens or none; the maidens are simply how it was earned.
The bound. The reward side needs no clamp: an over cannot concede less than nothing, so it can never save more than par. The cost side can run away (an over can go for 30), so it is floored at the same size — an over is worth at most +par and at worst −par. The bound is the division’s own measured par, not a chosen number.
Why this cannot flatter anyone. Because par is the division’s own average, the term adds up to zero across the division: every run credited to a bowler who went under par is a run charged to one who went over. It re-ranks bowlers within their division; it cannot inflate the league’s scores. And it is priced 1:1 with batting runs, on the plain principle that a run prevented is worth a run scored.
Overs are counted in balls (7.3 overs is 7 overs and 3 balls, not 7.3), a card claiming more maidens than overs is capped at its overs, and a division with no bowling recorded yet prices no run prevention at all rather than borrowing another division’s par.
Fielding value is the fielder’s share of the dismissals above: 30% of the dismissal’s value for a catch or a stumping, and 100% for a run out, split evenly when two fielders are named. The shares are unchanged by any of the 2026-09-17 additions; what moves is the dismissal’s own value, with the dismissed batter’s position and the match’s conditions. Credit is only given where the scorecard names the fielder and the name matches a player on the club’s own roster exactly, so scorecards recorded by initials only undercount fielding. The Field column on the board is the plain count of those credited dismissals; the score uses their value.
Keeping is scored the way the PCA publishes it: catches, stumpings and run-outs count as fielding, with no separate wicket-keeping component. A keeper earns the fielder share (30%, bowler 70%) for a caught behind or a stumping, the same as any other catch. The PCA publishes no per-dismissal values or bowler and fielder splits, so Selection Room’s 70/30 split is its own choice.
Work a scorecard does not record, such as byes saved and taking every ball, is not counted. Specialist keepers therefore naturally rank lower on a combined board, which is why a keeper carries a wicket-keeper tag and is never flagged for output below opportunity. A player gets the tag when they kept wicket in at least 40% of their games for that team this season; a game counts as kept when the scorecard credits them with a caught behind or a stumping, so a game in which the keeper took no catch and made no stumping cannot be seen. Caught behind is identified from the scorecard’s keeper marker, a same-innings stumping naming the keeper, or the team’s usual keeper. Scorecards recorded by initials only carry no keeper marker and undercount keeping.
Every price above is specific to one division in one season, never one number for the whole league. Three sets of figures are recomputed from that division’s own innings every time new results are imported:
Where a division has no figures yet, the league-wide runs per wicket for that season is used.
The form-window slider sets N, the number of each player’s most recent games for this team that count as current form. It starts at the team’s whole season to date. Every “Last N” figure on the board, the in-form, dipping and output flags, and the promotion ranking use the window; the Season column always stays alongside as the comparison. Narrow windows react fastest but are the noisiest. A separate minimum-games slider (default 3) hides players with fewer games from the board, the signals and the promotion tiles.
Both groups cover the player’s whole league season for this team, not the form window. Dismissals are classified from the scorecard text; the columns are B (bowled, with hit wicket folded in), LBW, CB (caught behind), ST (stumped) and Ct (every other catch, including caught and bowled). Run outs carry no bowling signal and are not shown.
A read only ever considers the strictly largest of three clusters: bowled plus lbw, caught behind, and stumped. A tie for largest, or a largest cluster that misses its own threshold, gives no read at all.
In plain English: this list is not written by AI or chosen by a person. It is picked out automatically from the scorecards, using the fixed rules below. The same rules apply to every player at every club. It shows under “Selection notes” on each team page, as “Players worth talking about”.
Signals only consider players with at least 4 real opportunities and at least the minimum games set on the page.
In plain English: these picks are not made by AI or by a person. They are picked out automatically from the scorecards, using the fixed rules below: the standout batter, all-rounder and bowler from the team below, but only if they are among the best in their whole division.
A club’s teams are ranked by division. The side below a team is the club’s next team down; a team that shares its division with another of the club’s teams has no promotion panel. Candidates come from the side below: at least 4 games there (or the page’s minimum games, if higher) and at least 4 real opportunities there. A player with 2 of their last 3 games already at this level or higher is treated as promoted and shown separately, with their record since moving up once they have 3 games at the new level.
The quality bar. Being the best in the side below is not enough on its own. A player is put forward only if their season impact per game in the side below is in the top quarter of their role across that side’s whole division this season. The quarter is drawn from every club’s players in the division with at least 4 games there, split into batters, all-rounders and bowlers by the same role rules the board uses (see Roles). The line is the score of the player at the one-quarter mark: with 20 batters it is the 5th-best, and anyone level with that score is in. Each tile states the bar it applied. If no one in a role clears it, the tile says so, gives the bar, and names the nearest player and their figure. It is never filled with a player below the bar.
Why the comparison stays inside the division. Every score is priced in its own division’s terms: runs per wicket, par strike rate, par economy and the value of a run at each batting position are all measured from that division’s own games. A score of 25 in Division 5 and a score of 25 in Premier B are priced against different yardsticks, so the bar only compares like with like.
Tried already. A player who has played at least 3 games for this team this season, at an impact per game below the median of this team’s regulars (4 or more games) in the same role, is not put forward again. They are listed under the tiles as “Tried at this level this season”, with their numbers at this level, next to the players already promoted and holding their spot.
Ranking, and the written notes. Players who clear the bar and have not been tried already are ranked by form-window impact. The written notes (which are written by AI) follow the same rules: they are only given the players the tiles would put forward, and a note that suggests promoting anyone else is rejected before it is published.
There is one tile each for batters, all-rounders and bowlers, showing the single top candidate. Each candidate carries up to 3 reasoning lines, in this order, from their own numbers: in form; averaging at least 30% above the position baselines (not for bowlers); a wicket-taking method or a known method of dismissal once its read qualifies; the team winning at least 15 percentage points more often with them in the side below (at least 4 decisive games); and games already played for this team.
Roles drive the board’s filters, the promotion tiles and the badges. They never change the score.
In plain English: the short write-up and the suggestions “For the selection meeting” are written by AI, from this team’s numbers. Before anything written by AI appears, every number in it is checked against the scorecards, and anything that fails the checks is not published. The “Players worth talking about” list and the step-up picks next to them are not written by AI: they are picked out from the scorecards by fixed rules (see Players worth talking about and Ready for a step up?).
Who sees what (18 September 2026). During the season, the “Selection notes” section (the written notes and the players worth talking about), the flags on the board (rising, fading, underused and the rest) and the “Ready for a step up?” picks are for approved members of the club, league officials and site administrators. Everyone else sees every statistic on the page, with a locked preview in place of those sections. All of it opens to everyone when the season ends on 30 September.
Each team gets a few short written cards (“The write-up”) and a short list of suggestions (“For the selection meeting”), written by AI from the team’s computed season statistics. The notes are generated each Sunday night after the week’s results are imported, with a retry pass on Monday morning, and only for the current season; earlier seasons’ notes are kept and shown, never regenerated.
Before anything is published it must pass automatic checks: 3 to 5 cards and 3 to 5 suggestions; every plain number must appear in the statistics the AI was given (within 0.15); no literal recent-form figures (those appear as live values that follow the form-window slider); every name must be a player on the team’s roster or a known team or division; no language about dropping players; exactly one highlighted takeaway per card; and every suggestion must name a player or cite a number, with at least 3 naming a player. A failed attempt is retried once with the reasons; if it still fails, a more capable model gets the same two attempts under the same checks; if that fails too, nothing is published for that team that week.
The statistics are always open to everyone. While a season is in progress, the AI selection notes are shown to logged-in, approved members of the team’s club, to league officials for that league, and to site administrators. When the season closes (at the end of 30 September), that season’s notes, frozen as last published, open to everyone.