The question guide

Test scorecards record innings, not each day’s score, so “day one” is refused rather than treated as the first innings — and so is “nightwatchman”, which means the last overs of a day. Who took the catch and what kind of ball it was are not recorded, though how a batter fares against spin or pace is, at both ends. Limited-overs chases are available at the moment a batter arrived; Tests have no published target in this dataset, and rain-revised targets are excluded rather than reconstructed. Phases apply to limited-overs cricket only — a Test innings has no powerplay.

Knowing the score when someone walked to the crease means replaying the innings ball by ball, which reaches back to 2001 for Tests and 2002 for ODIs. A named batter is answered in either format with that caveat, and the IPL and T20Is are answered outright — but an all-time Test or ODI leaderboard of it is refused, because the batters such a question is about made most of their runs earlier. Bounded windows inside the delivery era are answered in full.

A verdict scores every measure as a percentile against qualified peers, so it needs enough of them to rank. “Best IPL batter coming in inside the first ten overs” has 34 and answers; “coming in at exactly three down” has 18 and is refused, with the count stated — below twenty the order is mostly who cleared the bar rather than who was better. The same question in Facts mode gives the averages and strike rates, which are real at that sample size. A verdict over where a batter came in is also refused for all of Test or ODI cricket, on the same coverage limit as the leaderboards, and cannot be combined with an innings phase: a phase splits an innings into spells and an arrival describes the whole innings.

The figures are available — a bowler’s economy, wickets and average over exactly the balls he sent down above ten an over, or below six. What is refused is a verdict on them, and the reason is the cricket rather than the sample: a bowler who takes a wicket or bowls a tight over pushes the required rate up and keeps bowling, so the best bowlers spend the most balls under the highest rates. Scoring that against peers would reward the situation rather than the bowling. It is also thresholds only — “above ten an over”, not a band between two numbers — because at ball-by-ball grain a one-run band is a fraction of a spell, and it is refused for all of Test and ODI cricket on the same delivery-coverage limit as the leaderboards above.

A batter’s chase figures are measured at the moment he walked in — the target, the runs still needed and the asking rate as he took strike. What is not yet available is the rate ball by ball from his end: “how he batted while they needed ten an over” would count only the balls he faced above that rate, wherever in the innings they fell. The bowler’s end of exactly that calculation is available. So a question naming both an asking rate and a phase of the innings is asked about rather than answered, because “when the required rate was above ten in the death overs” can honestly mean his death overs in the innings he arrived needing ten, or his whole innings in those matches, and neither is the ball-by-ball reading most people have in mind.

“Fastest to 3,000 runs” is answered, on a clock the reader can name — innings, balls faced or bowled, matches, or days from a debut. What is refused is a race over a narrower career: against one opponent, at one ground, inside a window. The running total accumulates the whole career on the scorecard, and a race over a filtered set is a different sum that has to be built from the coarser grains before it can be counted honestly. The totals themselves are available, and so are the passages inside one innings that the deliveries record: the fastest hundred, the slowest route through the nineties, how long a batter took from fifty to three figures.

The figures are available and exact — how a batter fares against spin, how a bowler bowls to left-handers — but a verdict on them is refused. How much of the innings a batter faces from spinners was decided by opposing captains and by which attacks he met, and about two balls in three go to right-handers whoever is bowling, so scoring either slice against peers would rank the fixture list as much as the player. The same questions in Facts mode give the runs, balls, average and strike rate, with no verdict. Who a *player* is, rather than who he was facing, does carry a verdict: best spinner and best left-hander both work, because a style is a property of the player and not of the oppositions he ran into.

A bowler’s style and a batter’s hand are biographical fields, so a player no source classifies is in neither the spin group nor the pace group, and a batter with no hand on record is in neither the left-handers nor the right-handers — the figures shown are exactly right about the cricket they counted, and a comparison quietly leaves that player out. The residue is almost entirely associate cricket: 11.5% of the balls non-full-member sides bowl (121,379 of 1,053,104) and 10.3% of their batting innings carry no attribute. The full members are essentially whole — 895 of 3,236,301 balls, from ten part-timers the record itself lists with a null style, and 115 of 193,034 innings. Every style page the fetch can reach has been read, and the 189 matches still on that list are all associate-only, so no further run changes a full-member answer. The attributes come from a player record frozen at the end of 2023; each answer counts the balls or the players its own window affected and says so.

Finals, semi-finals, knockouts and the IPL’s play-offs are all filterable, back to 1975 — the rounds the sources never labelled were identified by hand and checked edition by edition. The group stage is deliberately not one of them: what that round is called has changed with every edition (First Round, League Stage, Super 8, Group A) and several editions name it nothing, so a question about it is asked about rather than answered, with the two exact readings either side of it offered in place of it. A round asked for without a competition is every competition in the format — the ODI knockouts are 131 matches and most are not the World Cup’s — so those answers say how many matches they counted. Qualifying tournaments, regional finals, the Asia Cup and franchise leagues are not the three ICC events, and “in the IPL” is a format rather than a competition.

Player-of-the-Match awards come from the source’s own record, which begins where the ball-by-ball data does — 2002 in ODIs, 2001 in Tests, and the first match of T20Is and the IPL. A Test or ODI award list is therefore an era list rather than an all-time one, and it says so. Inside that span the record is still short: a match the source left blank counts as neither a win nor a loss, and every answer says how many of the matches it counted carried no award. An award is never part of a verdict — deciding what one match’s judgement is worth beside a batting average would be my opinion rather than a calculation — and there is no phase or batting position to narrow one to, because a match is a whole match whatever happened in it.

Catches and stumpings are counted, but no source records who captained or kept wicket in a given match, so “his average as captain” or “only the innings he kept” cannot be filtered. Taking one role *out* is refused for the same reason: a keeper’s catches are stored as catches, so “most catches excluding wicketkeepers” or “by outfielders” cannot be compiled — the same question without it ranks MS Dhoni first — and it is refused before the answer is paid for, with the two questions either side of it offered instead.

A named ground is a filter, by name, nickname or city: “at the Wankhede”, “at the Gabba”, “at Chepauk”, “in Mumbai” — and it stacks with everything else, so “Indian batters in Australia at the MCG” works. What is left is what no string rule reaches. “County Ground” is seven grounds and “Lord’s” is two, so a shared name is either settled by how much cricket each holds or put back to you as a question, and a city with several grounds (“in Mumbai” is three) counts all of them and says so. A ground renamed keeps both names — the Feroz Shah Kotla and the Arun Jaitley Stadium are one place — and one match in the dataset carries no venue at all, so it is in no ground filter. An IPL ground works; IPL home and away does not, because franchises are not countries.

A side can be asked about as a whole — how often it wins, what it averages, how often it has collapsed, and how those move with the state of its innings: three wickets down in the powerplay, 24 for 3, sixty conceded in the first ten overs. What it cannot be asked for is its own bowling as a record. The figures exist only as the opposition’s innings in the same match, so runs conceded, wickets taken and economy are ODI, T20I and IPL questions and are refused for Tests, which pair nothing. A side is never judged either: there is no verdict over teams, because a verdict weighs a player against qualified peers and a side has no peers of that kind and no batting position. And a team answer has no phase of its own — eleven players bat in the powerplay and a total is not one batter’s — so a phase question about a side is asked about a batter or a bowler instead.

Every question stands alone: there is no conversation, nothing is remembered between one ask and the next, and a phrase like “these matches” has nothing to point at unless it is typed under the answer it means. Under an answer you get the “About these matches” tag, and it is the tag rather than the sentence that narrows the next question to the matches the last one was about — remove it and a sentence pointing at the answer above is refused, with the tag named, rather than answered wider. The tag is offered on a list of one named player’s innings (his hundreds, his five-wicket hauls, his ten best) because that is the shape whose matches you can point at; a career total is not offered one, since its matches are the whole career and “these” would be a heading rather than a narrowing. The narrowing is by match, so “these players” is not a scope — it is “in these matches”, and the player and format carry down from the answer while its other conditions do not. Two scopes deep is the limit: a scope of a scope works, a third is refused.

Facts mode keeps formats and disciplines separate because raw totals and averages cannot be added. Verdict mode can compare unit-free batting, bowling and all-rounder scores.

Scorecards show a player’s latest recorded IPL team, not releases, trades or auction results.

The published dataset currently covers men’s Tests, ODIs, T20Is and the IPL.

A ground names the places to count, so “RCB’s record outside Chinnaswamy” is refused rather than answered with every ground — the words ask for a set this cannot name. The ground itself and the side’s whole record are offered instead. “Away from home” does answer, but it is a different set: the matches a side played as the visiting team, which is not every ground but its own.