The Truth of the Empty Column: The One Field Without Which Every Cricket Model Is Incomplete

The Truth of the Empty Column: The One Field Without Which Every Cricket Model Is Incomplete

On the evening of 12 February 2026, in the auction hall in Bengaluru, one column on the screen was doing all the work: impact per crore. The sheet our Gujarat Titans data desk was working from carried more than two hundred names, and one of them sat apart from the rest in that column — David Miller, thirty-two years old, near his base price, at ₹3 crore. The bid came, the name was ours. On 29 May 2026 at the Narendra Modi Stadium, in a seven-wicket win over Rajasthan Royals, Miller's season closed at 481 runs with a strike rate above 140. One column of a spreadsheet had turned into a trophy.

A different kind of document landed on the same desk last month. Twelve pages, eight sections, tidy tables, and a single value repeated in every cell: N/A. The title said "deep analysis". No format, no team, no player, no date — every label present, every value missing. On first reading it looked like an empty document. On second reading it was the most honest report that reached my desk that week. A report in which every line says N/A cannot lie to you. A report in which every line carries a confident adjective can — and usually does.

The scorebook knew the truth before the crowd did. The trouble is that cricket content now records the crowd before it opens the scorebook, and files the recording under data.

Context: one field that carries the weight of the whole building

Cricket is the only sport in the world where four different games run under one name. Test, ODI, T20 and The Hundred differ in scoring rate, phase structure, bowling economics and success benchmarks as sharply as cricket differs from hockey. The first condition of any cricket claim, therefore, is that it declares its format. Without a format tag a number is not wrong — it is meaningless. A meaningless number is more dangerous than a wrong one, because a wrong number can be caught, while a meaningless one floats in the air for years.

The T20 skeleton is fixed: the first six overs of fielding restrictions, overs seven to fifteen of containment, overs sixteen to twenty of acceleration. The ODI skeleton runs on two new balls and the final ten-over surge. The Test skeleton runs on session-by-session attrition and declaration timing. Each asks a different question. T20 asks why the spinner came on in the seventh over; Test asks who won the fourth session. A report with no format cannot ask either. It can only fill eight sections with N/A.

My own training on this was blunt. I entered cricket through academia in 2026, began a DPhil on the social history of Indian cricket as a Rhodes Scholar in 2026–2026, and completed it in March 2026. Back then I thought history was the hardest discipline because its sources arrive broken. After a decade on data desks I understood that history at least has sources. Data has only claims that are introduced as sources.

Core: six lessons that counting taught me

Lesson one — without a format tag, comparison is a category error.

A T20 finisher striking at 180 and a Test opener averaging 45 are two professionals in two different sports. Putting their numbers in one table is like dropping a swimmer's seconds and a marathoner's hours into the same column. Bowling economy makes the error plainer: 7 an over is excellent in T20 and meaningless in Test, because a Test bowler is measured by cost per wicket, not by run rate. Any report that says "his economy is 7.4, so he is expensive" has lost the format in its first clause, and the sentence collapses the moment the format goes missing.

I have a working example. A digital desk sent me to Russia for the 2026 World Cup with a laptop and a scraped dataset. Before the opening match I built an xG and PPDA model across 32 teams and ranked France fourth: six goals conceded in seven games, possession ceded to opponents. On 15 July 2026 at Luzhniki, France beat Croatia 4–2 with 34 per cent of the ball. That night several people wanted the word "upset" from me. I refused and filed a piece titled "Possession Is a Lagging Indicator". Since that day the word "deserved" has been banned from my drafts unless a metric stands behind it.

Lesson two — labels intact and values missing is itself a diagnosis.

When field names survive intact and the values inside them are blank, there are only two causes: either the article body never entered the system, or the source was never open at crawl time. Either way the result is the same — analysis is impossible. The real risk is not that analysis is impossible; it is that once an empty skeleton starts moving downstream, a human compulsion to fill it begins. Eight empty boxes invite filling, and what gets poured in is rarely true, only true-sounding.

Here I borrow the principle of DRS umpire's call. The technology shows the ball's path, but if the ball lands inside a defined margin, the on-field decision stands. Cricket has accepted that measurement carries an error bar and that inside that bar a human call is not inferior to a machine. Data journalism never learned this. We publish model output without publishing the error bar. No figure enters my copy without a footnote on sample size and collection method, and that habit makes my filing ninety minutes late. Editors have learned to expect the delay, and I accept it.

Lesson three — 1,214 Ranji overs: the discipline of a manual count.

In June 2026, some months after Gujarat's maiden Ranji title, the Gujarat Cricket Association handed me a part-time analytics contract to code the state side's 2026–18 season by hand. Domestic grounds carry no ball-tracking, so I logged 1,214 overs off livestream footage into a spell-load index. Gujarat's seamers were conceding 3.4 an over in their first four-over block and 5.9 in their second. I published the methodology on a new sports site — the International Communication degree finally earning its keep — and was mocked for "counting overs". By the quarter-finals the second-spell economy had fallen to 4.6.

That episode gave me a rule I have not broken since: I counted 1,214 overs before I trusted a single model. A model tells you what you taught it. A manual count shows you what you never thought to teach it. The second-spell collapse was no secret; it was written in the scorebook and nobody had read it.

Lesson four — sixty empty stadiums taught me what a crowd actually does.

Through the 2026 IPL, played behind closed doors in the UAE from September to November, I worked remotely from Ahmedabad for a franchise data desk, testing whether empty stands flattened toss and dew advantages. Across 60 matches in my count, the chase win-rate rose to 58 per cent. That number is arresting and it proves nothing on its own. Sixty matches, two or three main venues, one season — a large sample for a season and a small one for a decade.

Sixty empty stadiums taught me what a crowd does. A crowd does not bowl and does not bat. It presses on marginal calls, it lifts a fielder's voice, and above all it sets the pitch of the television commentary. With the stands empty the numbers were still talking — the crowd is data too, even when the seats are empty. We simply never entered that data under a column name.

Lesson five — Adelaide's 36: a three-part post-mortem.

19 December 2026, Adelaide. India bowled out for 36 in 21.2 overs. Josh Hazlewood 5/8, Pat Cummins 4/21. That night I coded all 21.2 overs frame by frame and charted where each batter's front foot landed. The work finished at four in the morning, and only then did I write a single word.

Since that night every collapse I write about goes through one template in three parts: conditions, decision quality, execution. I do not publish a post-mortem until all three carry numbers. The template showed me that 36 was not one technical failure. In the first session the pitch was changing bounce; in the second India's front foot kept landing in the wrong place; in the third the shot selection was the one that had worked in the first innings. Three separate failures recorded in one scorebook. Any analysis that folds them into a single "batting collapse" is convenient and wrong.

Lesson six — at an auction, price is not a measure of talent.

At the February 2026 mega auction I ran an impact-per-crore model instead of a runs model. It starts from the role a squad has left empty and prices the available talent against that role, rather than ranking career numbers. It flagged Miller as the most underpriced batter in the pool — thirty-two years old, three ordinary seasons behind him, and a franchise needing middle-order experience. He went for ₹3 crore and finished with 481 runs above a 140 strike rate in a title-winning debut season.

If the story stops there it is promotion, not analysis. So it also has to be said that the same model placed three names near the top of that pool whose seasons did not pass 150 runs. I am not naming them, because the confidentiality terms of that contract still apply — and that same clause limits me here to what is verifiable. I do not chase outliers; I wait for them to repeat. Miller's season happened once. If the impact-per-crore argument works again at the next auction, the framework is proven. One season does not build a model's reputation.

Lesson seven — read India–Pakistan as an event sample.

There has been no bilateral series between the two countries since 2026–13. That does not mean they have not played. It means every meeting has come inside an ICC or ACC event, from the 2026 Asia Cup to the 2026 Asia Cup, including the 2026 Champions Trophy final (Pakistan won by 180 runs at The Oval) and the 2026 T20 World Cup (Pakistan won by ten wickets in Dubai). That is a small, extremely high-pressure, tournament-shaped sample, and measuring it against bilateral-series nostalgia is not a method; it is a habit.

My lineage is Pakistani and my market is Indian, which makes this the easiest subject in the world for me to be sentimental about. So I have a rule: any claim about this rivalry stays inside ICC and Asia Cup matches since 2026–13. No hypothetical bilateral series, no political sermonising. A series that was never scheduled cannot be analysed.

Lesson eight — the second XI audit.

India's bench-strength story has always looked incomplete to me. Look at the domestic pyramid: Ranji's media footprint has shrunk over fifteen years while the IPL has expanded. Everyone says the IPL damaged Test cricket. My count says something else — the real casualty of the IPL was not Test cricket but Ranji's media coverage. Test grounds fill up; the reporters who cover Ranji have thinned out. On the day nobody films a domestic match, the selection argument loses its only independent evidence.

That produces the claim I keep repeating: the hardest team in world sport is India's second XI. It sounds romantic, but its basis is numerical — the selection doors are narrow, replacement capacity is high, and the sample on which a player is judged is short. Together they create an economy in which one match decides a career. That is the actual story of the debate.

Contrarian: the empty report is not the enemy — the full one is

The easy route would be to hand that twelve-page N/A report to the old guard and announce that the data desk won the tactical argument years ago. It is tempting because it sounds true. The problem is that it is only half true. The data desk won the technical argument and inherited the commentary box's worst habit in the process: confidence without a sample. Twenty years ago television said "there is something about this boy"; today a website says "his impact score is 6.2". Neither sentence contains a fact, and both contain total certainty.

The sharpest claim in this piece is that cricket journalism's biggest failure is not false numbers but the language of numbers used without them. "He has good tempo", "he is a big-match player", "there is spin-smell in his bowling" — each sounds measurable and none is measured. A false number can be caught by a colleague. A counterfeit vocabulary cannot be caught, because there is no standard to catch it with.

The second point is about correlation and causation, and it applies directly to that 58 per cent chase rate. It is easy to conclude that empty stands amplified the dew advantage and made chasing easier. The problem is that I could not isolate dew. In the UAE in 2026 dew was present in every match, which makes it a constant rather than a variable. A constant cannot be assigned causality. I had sixty matches and no control group. So what I can say is this: the chase win-rate was 58 per cent; why it was 58 per cent is not something my evidence supports. A journalist who writes "because of the dew" is not running a model; he is dressing a guess in a statistic's clothes.

Third, and least comfortable: it is easy to scold an empty document, but nobody asks the empty document whether its N/A was correct. Seven of those eight sections were right. The format tag was absent, so every metric was invalid. The report had the nerve to say it did not know. In a media culture where saying "I don't know" costs you the shift, the next step is always a plausible-looking inference.

Fourth, a point I have had to apply to my own work: model builders err too, and admit it least. In April 2026 that impact-per-crore model told me the best value in the pool was a 32-year-old middle-order batter. Had Miller's season gone badly, this paragraph would be defending the model. Evidence first, reputation after — that is the only ethical condition of a data desk, and it is the one that most annoys readers, because readers want a narrative.

Fifth, and closest to my own trade: learn to translate into broadcast language. The data desk won the technical argument long ago and then stopped speaking the language the viewer understands. That is a strategic error. "That seamer's economy moves from 3.4 in his first spell to 5.9 in his second" is a fact any Hindi commentator could turn into praise of the same bowler. My job is not to sneer at the commentary box; my job is to hand it a sentence it can say on air.

Takeaway: three mandatory fields for the next round

As this season rolls on I am watching three things, and all three are data-level signals rather than results. First, whether any part of the media returns to the domestic season — independent evidence on Ranji selection is what turns the second XI story from rumour into analysis. Second, whether any franchise at the coming auctions backs the impact-per-crore logic again by letting an expensive name go and bidding on an under-priced one; if the argument works twice, Miller stops being a good buy and becomes a framework. Third, how long it takes for domestic metrics such as spell-load to appear on broadcast screens, because the day they do, a fast bowler's fatigue becomes a number before it becomes an injury.

All three share one name: the honesty that begins with a format tag. A cricket report that opens with the format, then the team, then the date, then the player, is far less likely to lie than one that opens with the words "deep analysis".

My question points forward, not back. The next time an eight-section cricket report reaches your phone, first find which column carries the format. If that column is empty, the other eight sections are worth nothing however handsome the tables. And if that column is filled, ask the second question: who counted those numbers, and after how many overs?

Methodology note

The Russia 2026 xG and PPDA model was built pre-tournament on a scraped 32-team dataset; the goal-based input across seven matches was a small sample and its error bar limited the confidence of the fourth-place call. The Ranji spell-load index covers 1,214 overs from the 2026–18 season, hand-coded off livestream footage; the first and second four-over blocks ran at 3.4 and 5.9 runs per over respectively, with the second block down to 4.6 by the quarter-finals. The 2026 IPL chase win-rate is based on a count of 60 matches in which dew was a constant and therefore cannot be assigned causality. The Adelaide 2026 analysis rests on frame-by-frame coding of 21.2 overs inside a three-part template. The IPL 2026 auction valuation carries its model error range and its list of failed names internally; only publicly verifiable facts are stated here.