Empty Cells, Full Desk: The Night the Cricket Data Pipeline Returned All N/A

At 3:07 a.m. on a Tuesday night in Liverpool, a file sat open on my desk called stage2_input_deconstruction.csv. Fourteen columns. It took four seconds to read the header: title, source, type, core viewpoint, information points, entities, time sensitivity, source quality. Every cell carried the same three characters — N/A.

The column marked information points carried a number: 0.

The kettle was boiling in the kitchen. The cursor blinked in the first empty cell, and I felt the exact urge that rises in every data desk at that hour — the urge to fill. One team name, one player, one format, four numbers. The file would suddenly look complete. Charts would appear. The columns would go green.

I switched the kettle off and left the file empty. This report is about that decision.

Cricket analytics has settled into a two-stage pipeline over the past six or seven years. The first stage pulls content out of a source — headline, outlet, date, fact points, entities. The second stage breaks that content down: format, technique, team structure, commercial logic, governance, risk. The pipeline exists because every major outlet and every franchise now has a data desk, and every desk has the same complaint: source quality is never uniform. The two-stage structure was the way to price that variance.

The trouble is that when the first stage comes back empty, the second stage finds itself in an awkward position. I had a framework with eight dimensions, five risk flags, three scenario cases and a rating system — and not one fact to put in it.

Cricket is unusually exposed to this, and the reason is arithmetic. A T20 innings produces roughly 250 to 300 ball-events; an ODI clears 600; a Test runs past two thousand. The density is so high that a desk always has some number within reach, and the availability of a number is easily mistaken for evidence. Football throws up about 900 pass events in a match; cricket produces four thousand in a day. In a sport where data is that cheap, an empty cell feels unbearable — and unbearable things get filled first.

I first felt this in June 2026. The Champions Trophy final, June 18, The Oval. Pakistan 338/4, India 158 all out, a 180-run margin. I was 24, three months from finishing an MS at Liverpool John Moores, and had talked my way into a freelance coding pass at a new-media startup through a friend. Across four re-watches I logged all 96 balls of India's top-order collapse — Mohammad Amir's 6-0-16-3 spell mapped by pitch, seam position and release height. It became a nine-part newsletter, "Lines of Least Resistance", with hand-drawn zone grids. 2,400 subscribers in six weeks.

I set myself a rule then that still holds: no tactical claim in print without at least 30 coded balls behind it. I took the match apart to see which part was still warm — and the parts that had gone cold, I did not write about.

The following year the same template travelled to a different sport. In the summer of 2026 a digital outlet paid me £150 a match to code all 64 World Cup games in Russia — 4,300 possession sequences, half-space heat maps, a spreadsheet I still call grid_v4. I filed the last chart on July 15, the day France beat Croatia 4-2. Three weeks later I carried the identical template to Edgbaston. India lost the first Test by 31 runs; Virat Kohli made 149. I mapped England's seamers against him and found 71% of deliveries in the channel outside off. Same grid, different ground. That piece drew 400,000 reads.

The success taught me a dangerous lesson. I learned that method travels further than subject — but I did not learn that method travels without its content. It does not. The grid transfers; the data that fills it does not.

September to November 2026, the IPL in the UAE — three stadiums, no crowds, a biosecure bubble. Delhi Capitals hired me as a remote performance analyst, and I worked from Liverpool on a four-hour time difference, filing dew models at 3 a.m. Night games in Dubai were chased down in 63% of cases. Sharjah's square boundaries measured 58 to 62 metres. Three stadiums, zero crowd: the model heard what the noise used to hide — with empty stands, 340 field changes were audible on broadcast, and nobody was logging them. Delhi finished second, losing the final to Mumbai Indians by five wickets on November 10.

The bubble taught me to write about variables nobody films — dew, heat load, silence, the 3 a.m. spreadsheet. My sentences got shorter and more numeric. That habit was waiting for me on that Tuesday night, when my first instinct on seeing the empty file was to ask which invisible variable was behind the emptiness.

The answer sat in the source. When a Stage-1 run returns only a domain label — cricket_world in this case — with no entities, no format keyword and no timestamp, one of three things is always true. First: the source text ran under about 200 words, a headline, an image caption, a fragment of a feed. Second: it was paywalled, and what the extractor saw was a wall. Third: the content was never text — video, embed, podcast clip.

Cricket has a fourth variant, and it looks the most innocent. Wire copy is often rewritten from the scorecard feed. Scorecards carry numbers, not sentences. Run an extractor over that copy and it knows the subject is cricket, because overs, runs and wickets are right there, but it finds no statement it can convert into an information point. The result: domain label populated, everything else blank. That is not a pipeline failure. That is the pipeline being honest.

The real failure starts afterwards.

A file that returns empty is a diagnosis, not a verdict. In ten years of watching data desks, the biggest error I have seen is not people drawing wrong conclusions. It is people reading an empty input as an incomplete piece of analysis rather than a broken line. The empty file was saying: resend me. The desk heard: guess for me.

There is a technical point underneath this that has to be stated plainly, because without it the rest is noise. Without a format, a cricket number is not a number. An economy rate of 8.2 is roughly par in T20I; the same 8.2 in ODI is about 2.4 runs an over expensive. A strike rate of 145 is fine in T20, extraordinary in ODI, and a different sport entirely in Test cricket. An average of 40 is ordinary in Tests and rare in T20Is. One figure, three formats, three different verdicts. So when the format reads N/A, the entire player-analysis section collapses — the basis of comparison is not there.

I saw this most clearly in the summer of 2026, when grid_v4 came from Russia to Edgbaston. A football grid does not sit straight on a Test match. Every variable had to be recalibrated — pitch moisture, the seam on a Dukes ball, May-to-August temperatures, session length. The grid stayed; its scale changed. Assuming the data will travel with the grid is the most expensive habit in the trade.

There are four ways a desk fills an empty file, and I have watched all four reach print.

Plausible ranges. The desk says: no name, but the format is probably T20, so take the average as somewhere between 28 and 32. This is the cleanest lie available because it makes no specific claim — just draws a band wide enough to fit nearly anyone. The problem is that a point picked from inside the band later becomes "data".

Borrowing from the adjacent match. No entity found, so the team's previous game is lifted whole and its numbers pressed onto this one. This is especially dangerous in cricket because pitches change and the same bowler looks like a different man two weeks later.

Inferring an entity from the label. The domain label says cricket_world, so the desk assumes a franchise or league story and builds around whichever tournament is currently live. This is the most common method and the hardest to catch, because the resulting story is internally coherent.

Substituting sentiment for a number. Where there is no statistic, in goes pressure, momentum, confidence. I know this habit more intimately than any other, and it irritates me most — "the momentum shifted" is a placeholder, not a mechanism. Which over was the difficult one, who lost the field restriction, whose run rate quietly collapsed: that is the information that does work. Sentiment cannot stand in for it.

Which brings me to my favourite question: what does an empty file cost?

I price decisions by what they cost, not by whether they worked. Resending an empty file costs a delay — four hours, maybe a day. Filing a filled-in wrong file costs in instalments: a correction, a source who stops picking up the phone, an auction board that bids on the number, and a reader who stops trusting your figures. Auction night is the best laboratory for this logic. Every year, crores are bid on numbers with no format tag attached — a domestic T20 strike rate, an international average, last season's economy rate. Purse arithmetic is unforgiving: a team has a fixed purse, and every buy consumes a slice of it. When that slice is spent on a figure whose scale has not been established, what was bought is not a player. It is an assumption. I have no hesitation saying that making India's second XI is the hardest selection job in world sport, and the reason is not a shortage of talent but a failure of pricing.

A number printed without a source is a mispriced asset. It acquires a market value, and then people start treating that value as the truth. This is where I distrust my own instincts most — the habit of deferring to price. Auction values, market odds, implied probabilities: these are signals, not verdicts. There is information in them and there is hedging in them. The question always has to be asked — who set this price, on what book, and what were they hedging against.

Three invisible variables sit in the pipeline, and all three were present in that empty file.

Timestamp decay. A file dated N/A could be fourteen months old. Cricket news decays fast: an injury story goes stale the moment the series ends, retention chatter is meaningless once the auction has passed. An undated file is not a news item; it is an archive.

The series window. The cricket calendar decides who is playing and who is resting. I have learned that the calendar picks the XI before the selector does — fixture congestion, rest windows, broadcast demand. A file that does not say which series window it belongs to cannot tell you anything useful about a player's form.

Weather and DLS. A revised target after rain, a toss whose value dew has rewritten — these bend results without announcing themselves. I built those models at 3 a.m. in the UAE bubble, and the hardest question there was the same one nobody answers directly: how heavy is the dew.

With all three missing, the question that never gets asked is the obvious one: what is this story for? Does it change a decision, correct a price, or just keep the desk looking busy?

I run five checks before filing, and this file failed every one. Does the source contain a format word — Test, ODI, T20, overs, innings? Is there at least one team and one player named? Does the date fall inside the current series or tournament window? Is the outlet one I can cite by name? Are there more than five information points? If any one of them reads N/A, the story does not start. That is not conservatism; it is cost control.

Now the part that worries me most, and it is not inside the analysis. It is on the reader's side.

The file I received was technically immaculate. Eight dimensions present. A risk matrix with six categories, all filled — even if every cell said unassessable. Three scenarios — worst case, base case, optimistic case — all N/A. An information value rating of one star out of five. A professional terminology section explaining format, null result, information point, RTM, DLS and NOC.

A fast reader glancing at that document could conclude: this is deep analysis. They would not notice that it contains not one cricket judgement. Structural completeness manufactures the illusion of analytical completeness — and that illusion is more dangerous than a filled-in file, because a filled-in file can at least be challenged. Nobody challenges an empty one. They file it.

This is where I start doubting my own method. The portable template is my pride — the thing that ran from Russia to Edgbaston, that recalibrated from a Dubai drop-in to the turn at Chepauk. But when a template is flexible enough to run with no content at all, its flexibility has become a weakness. A frame that can wrap an empty box will never tell you the box is empty.

There is one remedy and it is unpleasant: when the sample is one and the surface is new, publish the variables you cannot yet price instead of a clean verdict. It is the same rule I apply to a fresh Wankhede pitch, where the sample is a single match and the dew is doing more work than the model. The confidence interval has to be said out loud, even when it reads as unfinished.

Because the real danger is not that someone draws a wrong conclusion. It is that someone draws a conclusion that looks right and has no source behind it.

I have a clear side in this argument: the null result is the most valuable output this framework can produce. That sounds strange, because the industry's incentives run the other way. Traffic rewards the full report and punishes the empty one. A desk that sends back "this cannot be analysed" is called slow. A desk that turns ten estimates into a beautiful document is called fast. Six months later, when the estimates fail, nobody remembers where they came from.

The same failure runs through every auction table I have watched. The bidder never sees what the number was built on; he sees only the price. And price, as established, is not information.

I also hold — this is self-criticism — that the null result can become a genre. Three thousand words explaining that nothing can be said is, if you are careless, another form of filling. I read this file twice, and on the second pass some sections felt less like a description of emptiness and more like a performance of it. That is the lesson for any writer who turns restraint into a style: restraint has a limit, and you can recognise it by the moment it starts delivering a lecture instead of a resend request.

So what is the right posture? One: label the empty file NULL RESULT explicitly and do not forward it to decision-makers as cricket intelligence. Two: re-run extraction in the same working session — if the source is a short news item, a single clean pass will probably recover title, outlet, date, three to five fact points and the entities. Three: do not draft a word of narrative before that second pass lands. The third rule is the hardest and the most important.

I know this is not a cricket story. There is no innings here, no spell, no final over. But the firmest lesson of seventeen years of watching matches is this: the most dangerous thing about a match you did not watch is not that you know nothing. It is that you own a very good template.

Next time a report reaches you with every column filled, ask one question: how much of this is data and how much is decoration? And when a report comes back N/A, do not read it as failure — the pipeline is telling you where it broke. The only open question is who pays for the repair: the broadcaster writing the calendar, the franchise filling the purse, or the reader who trusted the number and found it wrong by morning.