An Empty Scorecard Is Still Evidence: Where Cricket's Data Supply Chain Drops the Story

4:30 a.m., Bangalore. The screen said 148 for 4 after 16.2 overs, and then everything stopped for eleven minutes. The feed came back at 152 for 4 — four runs, five balls, no explanation. The official scorecard stamped those five deliveries 'data not available', and by the next morning the same batter was sitting in an aggregate index at his full average, as though nothing had happened. A clock in my head beats the official timestamp: that night my notebook carried eleven dot balls between the thirteenth and seventeenth overs, and not one of them ever reached a feed. This is not a complaint about a server. It is a report on the moment cricket's data goes silent before cricket does — and on the confident conclusion that gets built on top of that silence by the next morning.

In a tournament cycle, that silence is at its most expensive, because there is no time to repair it. The next match is forty-eight hours away, the next report files by dawn, and the same incomplete average is on the table at the next selection meeting.

Cricket's numbers run through a four-storey supply chain. Ground floor: the scorer at the venue, pen moving on every ball. Second floor: the board's official statistics department, updating the file within hours of the close. Third floor: the broadcast feed and the data aggregators, tagging ball by ball — line, length, shot type, field position. Fourth floor: the analytics layer, which means fantasy points, selection models, franchise scouting reports, and now the team's own dashboards.

The crack is not always visible at the top. When the scorer on the ground floor leaves his seat — rain, a long DRS review, a shift change — the three floors above fill the gap in their own ways. One guesses. One repeats the previous ball's number. One quietly drops the over. None of them fills it with what the man who left actually knew.

You will say four runs and five balls do not matter. They matter when the same numbers become a decimal in the net run rate. At the 2026 World Cup, Pakistan and New Zealand finished level on 11 points; New Zealand took the semi-final place because their NRR was -0.106 against Pakistan's -0.43. A four-year campaign sat on a decimal, and that decimal was built out of overs somebody had bothered to record properly. In a sport where a semi-final is decided at the third decimal place, 'data not available' is not a harmless annotation. The same arithmetic governs revised targets under Duckworth-Lewis-Stern: a mis-entered revised target corrupts the NRR, and a corrupted NRR corrupts the table.

Failure one: the format tag disappears. This is the most common and most expensive error in cricket analytics. A Test average, an ODI strike rate and a T20 economy are three different measurements of three different games; pooling them into one column is like compressing a fast bowler's third-spell length and his death-overs yorker into a single score. January 19, 2026, Brisbane. India were 186 for 6; Washington Sundar and Shardul Thakur added 123 for the seventh wicket; Thakur took seven wickets in the match; India chased 328 at 329 for 7 to take the series 2-1, using twenty players. A selection model that had filed Thakur's T20 economy in the same column as his Test bowling does not pick him that morning. When the format tag is missing, the model does not pick the player — the model picks its own convenience.

Failure two: blank records quietly dropped from the aggregate. The real question is not whether a record is blank; it is which source class the blank belongs to. JavaScript-rendered scorecards, paywalled pages, regional-language portals — those three classes die first. The consequence is that domestic cricket, the Ranji rounds and the women's domestic games weigh less in the aggregate than they should, when that data is the only instrument that measures genuine bench depth. I re-ran my second-XI thesis for two seasons on exactly this material — caps, debutants, replacement value. But if more than five per cent of records are blank, the thesis stands on half the data, not all of it. And a record dropped in silence is not logged as 'no signal'. It is logged as nothing at all. That is the most dangerous gap in any aggregate, because a blank record carries no weight — it simply vanishes and leaves the average looking clean.

Failure three: reconstruction from memory. When the official record is empty, the story gets rewritten — from somebody's recollection, somebody's second-hand summary, somebody's 'I remember it'. November 2026, Mumbai. I was the only woman in the room when the bans were read, one among forty-one accredited journalists. The hearing itself was closed to us, so I filed 900 words from the corridor off pages 44 to 61 of the CBI report I had photocopied two weeks earlier: the Cronje phone transcripts, the amounts in rand, the dates. Two rival reporters filed the same story with the wrong number of players named. Mine held, because it had a page number in it. Same pattern, February 6, 2026, Goa. The bid was in my notebook ninety seconds before the feed — Bangalore took Kevin Pietersen for $1.55 million, Chennai took Andrew Flintoff for the same figure minutes later. The rival who had the right number published it four minutes late. A record built from memory usually looks right and is almost always incomplete; a record built from a document usually looks incomplete and is almost always right.

July 23, 2026, Lord's. The Women's World Cup final, seven women among roughly 120 accredited media. India were 191 for 3 in the 43rd over, chasing 229, needing 38 from 45 balls. Then seven wickets fell for 28 runs and Anya Shrubsole finished with 6 for 46. Seven women in the press box, and nobody blamed the over — the room blamed the 44th; I had counted 31 dot balls across the last twelve overs and wrote that instead. It was a strike-rotation failure, not a nerve failure. Had my notebook been empty that night, the easiest story would have been filed by dawn: they cracked under pressure. An empty notebook never stays empty. It gets filled with whichever story is most convenient.

I have to write where I could be wrong. First: not every blank record is a pipeline fault. Some sources are genuinely hollow — a live-score page that carries numbers and no narrative, a video clip with no words. Marking those as 'data not found' is correct; calling it a suppressed record is not. Second, and less comfortable: documents are not neutral either. Minutes that decline to record who was absent are not a server glitch, they are a choice made by power. The post-2026 spot-fixing fallout and the Lodha reforms both show the same pattern — the record protects whoever was in the room to protect it. Third: sometimes the first XI is simply better, and my bench-strength thesis has to lose a round in public. If I turn every blank record into proof of hidden talent, I commit the exact error I audit in others — deciding the conclusion before the evidence.

So I am filing a prediction before this tournament cycle ends, so it cannot be edited afterwards: at least one widely quoted aggregate or selection metric will be traced to a batch carrying more than five per cent blank records, and it will be quietly re-based without a correction note. When that happens, the first question is not what the number was. It is how many other records in that batch were blank, and how many of those were domestic.