The Empty Column: The Cricket Data That Was Never Recorded

Pune, three in the afternoon. Two files open on the laptop. On the left, a scanned scorecard from a first-class season — four columns beside the bowler's name: overs, maidens, runs, wickets. Nothing else. On the right, the same sport's other face: a ball-by-ball file from a T20 match with more than four hundred columns — line, length, shot, field position, run value, wicket probability, even the crowd noise on that delivery.

One sport. Two universes.

That morning a franchise analyst had emailed me a simple question: in the 2026 season, how many balls did this spinner bowl between overs twelve and fifteen, and what was his economy in that block? I opened the file, looked at the four columns, and replied: nobody knows. The information was not lost. It was never written.

India's cricket archive is not a census. It is a broadcast log. Where the cameras were, every ball is recorded. Where they were not, cricket happened, wickets fell, careers opened and closed, and the database carries no trace of it. I write this because we keep measuring the health of the game by what we counted rather than by what we chose to count.

The archive was built on the scorecard, not the ball. When CricInfo began in 2026 it was, at root, a scorecard collection — scoreheets typed by hand and passed around email lists and Usenet groups. The unit of that era was the match, not the delivery. That difference is structural, not technical. A scorecard is an aggregate, and aggregation destroys interaction. When all you hold is a sum, you cannot say who was beating whom inside it.

Ball-by-ball encoding became standard in international cricket through the 2000s, and with the IPL from 2026 it became industrial, because franchise economics demanded ball-level granularity. Pricing a player at auction required knowing what he does in the powerplay and what he does at the death. A scorecard cannot tell you that. So the grain of cricket's database — the fineness to which information exists — grew along the map of money. Where the money was, every ball is on file. Where it was not, the same four columns are still waiting.

In the summer of 2026 a broadcaster's data desk sent me to Russia: six weeks, sixty-four matches, an expected-goals model and a pressing index built from scratch in a rented flat. I flew back to Pune and stripped the football out of it — expected runs per ball by phase, control percentage, a pressure index from dot-ball clusters. Two Indian digital desks rejected the framework outright. The third ran it through India's 4-1 defeat in England, where my top-order false-shot rate read 27% inside the first ten overs. I learned plenty from that trip, and one thing more useful than any framework: I borrowed xG from Russia and learned that data travels better than tactics — but only when its granularity can travel with it. The first casualty of that framework in India was exactly this: in matches with ball-level data the model ran; where only a scorecard existed, that 27% line could not even be drawn.

So the question is not how much data we have. It is how many empty columns sit inside our data, and whose names are next to them.

I sort those empty columns into three kinds, and each has its own politics.

First, the empty column of time — cricket before 2026. Scorecards exist here, but they are not uniform. Some records carry full bowling analyses, some only wickets and runs. Catch credit goes to the keeper in one source and to slip in another. Run-out and stumping conventions shift season by season. All of that can be cleaned. What cannot be recovered is this: no player from that era will ever have a phase profile. A man who played his entire career in 2026 is, in any modern model, an entity with a name, an average, and nothing else.

Second, the vertical empty column — domestic, associate and women's cricket. The Ranji Trophy has been played since 2026 and its scorecard record largely survives. But ball-by-ball? That exists only where the match was broadcast or where an encoding agency was contracted. A large number of domestic matches have no video, no tracking, no ball-level record at all. In the women's game the fault line is sharper: the WPL began in 2026 and arrived with a full ball-by-ball corpus, while a large share of the domestic women's matches before it still stands in the four-column era. Associate cricket is thinnest of all — the ICC has pathways and tournaments, but ball-level records across several divisions remain irregular.

That asymmetry is not innocent. The national selection pipeline fills itself from the numbers that exist, and numbers exist where the cameras are. The camera is not a technical device; it is a selection filter. A spinner with two hundred first-class wickets across ten untelevised seasons has wickets but no profile. A bowler with forty wickets in a televised league has a profile. Scouts buy profiles, not wickets. Which is where I say the uncomfortable thing: the IPL auction does not buy players, it buys data profiles — and a player with no profile is unavailable in the market even while he is present in it. The purse is distributed against measurable quality, not quality.

Third, and least discussed — the empty column of dimension. This is what we fail to record even while recording the ball. Fielders' exact positions until recently, release height, batter intent, running speed between the wickets, weather, grass, wind — and the largest omission of all, the crowd. In March 2026 the calendar went dark. I spent the shutdown encoding all 757 IPL matches played to that point, ball by ball, and built a home-advantage model. When the league moved to the UAE that September, a franchise gave me a remote contract to test it live. With empty stands the measurable toss-and-chase home edge collapsed toward nothing, and the dew factor I had been quoting for a decade turned out to be mostly crowd noise and narrative. The empty stadium study showed me that silence has a measurable weight — and that weight sits in none of our ball-level columns.

In my model the crowd was a noise variable. It turned out to be the signal. A large slice of the home-advantage literature has been written treating that variable as noise.

Now the part where the problem is not missing data but falsely uniform data.

In 2026 the IPL introduced the Impact Player. The rule changes the number of bowling options, changes batting depth, changes the price of an all-rounder, changes the phase distribution of overs. The direct consequence is that a phase-adjusted strike rate computed by stacking 2026 and 2026 into one column is two different quantities sharing one name. The data has not broken; its meaning has. I call this a comparability break — the rule changes, the series quietly splits in two, and the coverage writes the same sentence the next day: strike rates are rising. They are not rising. The denominator is being replaced every three years.

There is a cheap fix, and I am genuinely puzzled that nobody owns it: give every series a version number, and log every rule change as a flag inside the same file. An analyst comparing 2026 with 2026 is comparing two different sports and does not know it.

The largest lesson of this whole exercise is not technical but commercial. Data gaps do not close by themselves, because the marginal cost of collecting data is lowest where data is already collected. Adding one more metric to a televised match is nearly free — the camera is there, the feed is flowing. Encoding a Ranji match with no broadcast costs many times more. Every new metric thickens the centre and leaves the periphery untouched. The more advanced we get, the more unequal we get.

Here I owe a confession, because the man who builds models sees an empty column and wants to fill it, even with material borrowed from another sport. The Russia framework worked spectacularly once, and that success bred a habit — take what looks elegant. That is framework envy. So my order is now reversed: cricket-native measures first, and an imported metric only when our discrete-event structure can carry it.

The honest consequence is that some empty columns should stay empty. Where the sample is small, I fast; where there is validation, I feast. As a data monk, I fast on small samples and feast on validation. An economy index conjured from four scorecard numbers will do a domestic player more harm than good, because it hands the market a precision that does not exist.

Which brings back my hardest lesson. In 2026, aged 38, I was analytics lead at Rising Pune Supergiant in the franchise's final season — the year MS Dhoni was stripped of the captaincy and Steve Smith was handed it. I built a powerplay matchup matrix that pushed Rahul Tripathi, 391 runs at a strike rate near 165, to the top of the order, and kept 17-year-old Washington Sundar in the XI as a specialist powerplay bowler: eight wickets at under seven an over. We lost the final in Hyderabad to Mumbai Indians by one run. With eight overs left, my model had us at 61%. It took a one-run final to teach me that certainty is a fragile model.

I imposed a rule on myself afterwards: nothing filed within 24 hours of a knockout, and instead of a reaction column, a published decision log — selection, matchup, field placement — so readers judge the process, not the scoreboard. Every final I cover still carries a paragraph on what the model missed. Even when we win.

That is why my posture in this whole argument is inverted. The default assumption is that more data means more clarity. In my experience the opposite has held. Cricket's data volume has multiplied over two decades while the economics of the game have stayed just as opaque — because the new data landed where data already was. Growth was not distributed; it was concentrated. Cricket's information deficit no longer comes from a shortage of data. It comes from the unequal distribution of data.

And there is a structural conflict I have to state plainly, because I spent two decades on the franchise and board beat and I know how easily access starts to feel like infrastructure. In Indian cricket the broadcast rights, the match data and the distribution of that data sit inside one ecosystem. The body that sells the rights also decides when the data layer is published. That is not an accusation of corruption; it is an ordinary incentive structure. Numbers that flatter the owner are never in a hurry to be filled, and numbers that raise questions — empty stands, domestic attendance, regional-language reach, the delivery of WPL broadcast promises — never get a contract.

So I keep one strand of my work that needs no locker-room goodwill: public records, broadcast filings, attendance data, and datasets I encode myself. That strand cannot be vetoed by an institution, and it tells the most truth.

None of this means I am against filling the gaps. I am against filling them badly. And one thing I am certain of: the true cost of a data gap becomes visible only after it closes, because retro-encoding gets more expensive every year. Tapes decay, scoreheets get thrown out, the people who were at the match get older and their memories shrink with them. If domestic ball-by-ball data is not contracted within the next five years, the fracture becomes permanent — not merely expensive to fill, but impossible.

Next time you see a strike rate, ask three questions: which series is this from, under which rule, and from what ball-level record. If you get no answer to all three, it is an opinion wearing a number's clothes.

My monitoring signal for this cycle is not on the pitch but in the language of contracts. Watch the next rights cycle for whether ball-by-ball and tracking data becomes a contracted deliverable for domestic matches — not an add-on, a condition. If it does, the fracture starts closing within two seasons and names that today exist only in a wicket count start entering the selection pipeline. If it does not, we should accept that we have chosen a sport in which some careers get counted and the rest get remembered.

And the question I still cannot answer: if a match was played and nobody encoded it, did it happen to the market at all?