The Matches That Come Back as Empty Files

It is 2:47 in the morning in Dhaka. I open the file on the laptop. Twelve columns are ready — date, venue, innings, over, bowler, batter, runs, wicket, extras, review, scorer, feed. Below them, zero rows. The cursor blinks in the first cell. I do not move my hand for roughly two minutes.

On paper, the match happened. Fourteen thousand people sat in the Mirpur stands, two strangers argued at the drinks break about which bowler should take the death overs, and when the boundary came off the last ball the whole ground stood up at once. But the file in front of me says: nothing happened.

The cricket was there. The record was not.

There is no scorecard for a single match in this piece. What follows sits one layer above the scorecard — the layer where it is decided which games enter the ledger, and which stay outside it forever. Cricket arguments are usually about results. This one is about the raw material arguments are made of.

1. The World Cup is on in Russia. I am a second-year student at the University of Dhaka with a stopwatch, a legal pad and an old laptop. I watch all 64 matches and log PPDA, xG and shot maps into a Google Sheet within 90 minutes of each final whistle. Bangladesh are not in that World Cup. I still log all 64, because I want to know what it feels like to hold an entire tournament in front of me at once.

That sheet still exists. It was my first school. What it taught me is that a dataset becomes complete only when somebody pays for it to be complete. In FIFA's ecosystem every pass, every pressing trigger, is purchasable within ninety minutes of the final whistle. That is not a gift of technology. It is a business decision.

Asian cricket is not short of money. The region supplies the largest share of global cricket revenue, and a single franchise season in this part of the world can pay a cricketer more than his entire first-class career. But the density of money and the density of record are not the same quantity. The assumption that the floor is highest where the budget is highest is simply wrong.

Tournament pressure widens the gap. I always read a tournament as a load story — who arrives tired, who has played three series back to back, whose knee is taped. That calculation needs over-by-over logging. In European football the log is produced daily. In South Asian domestic cricket it often is not, which means rotation decisions rest not on good data but on whose voice carries furthest in the room.

I built an index. I call it the Coverage Floor Index. I keep the definition crisp so that it is easy to attack me: for a defined competition-season, it is the share of scheduled senior matches whose ball-by-ball record is available, with a fixed minimum field set (batter, bowler, over, runs, wicket, extras), in a public or licensed feed within 72 hours of the last ball.

Let me be explicit: this is a proxy. It does not tell you how accurate the record is, whether selectors use it, or whether the scorer spelled the name correctly. It answers one question — is the match being converted into a form a model can eat?

From my own hand-logged files and public feeds, I can only give a band, not a number, because my sampling is structurally biased: I have mostly logged matches I could watch, and I could watch them because they were logged. That loop is the real trap, and it is the method's trap, not one writer's. The band still helps. Men's international cricket sits close to the ceiling. Men's domestic first-class and List A in South Asia sits a tier below. Women's domestic competitions sit a tier below that. Age-group and district cricket sits near the floor.

I do not call this gap "hidden talent." It sounds beautiful, and beautiful sentences cannot be measured. I say: where the model does not look, the budget does not go. A selection model is a machine that eats one specific unit. If a competition does not produce that unit, the model is blind to it no matter how much ability is on the field.

Picture a young fast bowler taking eight wickets in a first-class match in Bogura in April. A record everyone standing on that ground will remember for life. But there is no ball-by-ball file, so he never enters the pool a franchise analyst scrapes. Six months later, the auction. His name is not called. That is not a defeat. That is a player who never even waited outside the door, because nobody told him the door existed for him.

Every match still has a human keeping the count. The district scorer with a carbon-paper ledger, who goes home for lunch and turns back mid-meal because a wicket has fallen. The format he writes in will never be read by an API. He has no spreadsheet, so his work cannot be measured, and work that cannot be measured does not get paid more.

The 612-match spreadsheet I built in 2026 — empty stadiums, home win rate falling from 43.1 percent to 34.6 percent — was published the same month a Dhaka sports desk laid off nine writers. I started a free Sunday class for them, teaching them to read football data, because football data exists. For cricket in this region that class could not have taught anything new. It would have taught how to build a story out of an empty file, which is journalism, not statistics.

The table remembers what the highlight reel forgets. The reverse is also true: where there is no table at all, the reel becomes the only memory.

My toughest opponent here sits inside me. Several analysts I respect argue that the spread of data is not the improvement of data — scale adds noise. The biggest selection errors in South Asian cricket over the last decade did not come from missing information; they came from leaning on information that existed. A fully logged league hundred has, more than once, outweighed an unlogged first-class fifty and cost someone a place. Three good matches are never a trend, and the people who break that rule usually pick the matches that happen to be on video.

The argument is fair. I concede it plainly, and I publish my own model's failure condition: if the Coverage Floor Index rises while selection variance does not, then my index is measuring the wrong thing. On that day, the index is void.

Correlation and cause stay separate here too. Coverage grew because money grew. The same money improved pitches, funded physios, cut travel. The improvement is present wherever coverage went, but the cause is the same currency. Crediting data would be an unclean inference — easy to say, hard to prove.

So what do I watch next? Three things. First, whether any South Asian board releases a full domestic season of ball-by-ball data on an open feed — release meaning the whole season, not a marketing campaign. Second, whether women's domestic and age-group coverage clears a minimum floor, because that is where the largest invisible share of talent sits. Third, whether the 72-hour window holds, or gets buried under another "coming soon" placeholder.

I do not model players. I model the spaces between them. And the space gets wider every season, because the fixture list grows and the coverage does not.

The file is still open. Zero rows, twelve columns. The question is not for the file. It is for those fourteen thousand people, and for the scorer who wrote eight wickets onto carbon paper. Whose season does this empty row belong to?