The Ledger of the Empty Cell: Cricket Analysis's Most Expensive Mistake

A sheet landed on my desk last week. Eleven rows, eleven headings: format, match nature, venue, player technique, squad structure, rankings, league economics, rules and governance, risk, market expectation, and the industry transmission line. Beside every row were the same words: "Not available — insufficient information." At the top hung a label, cricket_asia, and at the bottom a note: the information points are empty, therefore no conclusions can be drawn.

What stopped me was not the sheet. It was the finger that was itching to fill it. After forty-one years of watching this desk, I can say with some confidence that cricket's most expensive analytical errors have not come from empty columns. They have come from columns somebody filled, because leaving them blank looked like a failure of effort.

Three floors of the information chain

Cricket's information chain stands on three floors. The first is the match played on the ground. The second is the data that comes off that match. The third is the decision built on the data — a selection, a bid, a rating, or a single line in a newspaper. My work sits on the third floor. And the third floor's biggest enemy is the thing that never visited the first: the question of format.

Test, ODI and T20 are not three moods of one game; they are three separate sets of denominators. A fifty-over powerplay and a twenty-over powerplay share a name and nothing else. Analysis that begins without a settled format is not selling a metric. It is selling an estimate at a metric's price.

This is why the pipeline keeps two stages apart. The first stage breaks the material down — which match, which format, which three hard facts. The second builds conclusions out of those facts. When the first stage returns empty, the second has nowhere to sit. The sheet in front of me was the second stage's report, and its entire substance was that the first stage had come back with nothing.

One reason for that nothing is visible in Indian domestic cricket. The Ranji Trophy's knockout rounds get coverage; much of the group stage survives on scorecards alone. At a selection table that gap is at its most expensive — the player with twenty televised innings and the player with four sit on the same sheet. That is not a shortage of talent. It is an asymmetry of information.

The arithmetic behind an empty cell

I went back to the sheet and looked for the column nobody had dared to fill. Zero verifiable units. Every conclusion needs at least one unit a third party can walk over and check — a line of a scorecard, a date, a name, an over number. Zero units means zero checkable claims. And a claim that cannot be checked is more dangerous than a wrong one: a wrong claim gets caught eventually, an uncheckable claim lives forever.

The second calculation concerns format. In data, every claim has a denominator. "His strike rate is good" is a sentence, not a claim, until you say in which format, in which phase, and against which baseline. Cricket has three native units built for this work: runs per ball by phase, wicket probability, and required rate against par. None of them can be called to the witness stand until the format is fixed. A model imported from outside cricket — football's xG is the familiar example — does not simply sit down here; it has to be translated, and the translation rule has to be written before the result, not after it.

The third calculation is the price of replacement. Replacement value is not a statistic. It is a question about who else was available. If the squad list is empty, the question cannot be asked — and when the question cannot be asked, there is no answer either.

In the 2026 IPL auction I tried to do exactly this. From the Dubai sportsbook desk I built a replacement-value model on 2026 IPL and Syed Mushtaq Ali data. The model rated Krunal Pandya's T20 strike rate of 148.7 and economy of 7.1 as underpriced; Mumbai Indians retained him, and in IPL 2026 he made 243 runs at 142.1. In the same audit I underrated Washington Sundar's powerplay overs — an error I logged, because a model's job is not to be right but to be checked.

The other face of that discipline appeared at the 2026 World Cup in Russia. Germany lost 0-2 to South Korea; the model had shown 1.9 xG for Germany against 0.3, with 72 percent possession. The market cut Germany's rating by 18 percent. I did not change the long-run model. I re-read twelve months of international data and treated the result as variance. A result audits a process; it does not earn the right to rewrite one.

Add one more calculation — the date. Every claim should travel with its date, or the analysis becomes a witness for its own result. I have read plenty of pieces stating that a model had called it in advance, where the date sits after the match. That is not forecasting. That is recall, and presenting recall as data belongs to the same family of errors as filling an empty column.

Source tier belongs there too. A board release, a reliable reporter's dispatch, general media copy, and an account chasing traffic can all write the same sentence, and they are not worth the same. When the source tier is not recorded, confidence cannot rise above low.

The one risk logged in the pipeline report is not a sporting risk. It is analytical: the risk that an engine manufactures a full output from an empty input, and that output reaches a decision. The report named the risk and then did the thing models rarely do. It refused.

The genuinely counter-intuitive part

The industry's default runs the other way: more data is better, faster output is better, and filling a blank is evidence of work. Part of that rule is sound — twenty years ago the shortage was scorecards, and now there is a glut. But the moment a model stops declaring what it did not have, every sentence it produces comes under suspicion. Writing "not available" on a blank sheet is not weak analysis. It is a strong record.

Some honesty is needed here, or the whole argument flips over. Not every blank is a hidden signal. Sometimes the data simply is not there, and no story comes out of its absence. I have logged the cases where the consensus was right and the model was wrong, and those cases deserve printing, because the writer who finds an inversion every single week is eventually just finding his own habit. The inversion here is not that the sheet was broken. The inversion is that the sheet worked — and correct work does not get applause.

The forward signal

I keep a ledger of every assumption I have ever had to retire. Its shortest entry reads: never treat an empty column as a zero, and never fill it from somebody else's sheet.

Next time a scorecard or a squad sheet reaches you with one column blank, the question is not which number to put in it. The question is who has the nerve to leave it blank. And the harder question after that: when it is finally filled, will you be able to say how much of it was measured and how much was guessed?