The Empty-Sheet Tax: In Cricket Analytics, the Most Expensive Mistake Is the One That Throws No Error

At 2:44 the screen went green. The pipeline finished its run, logged status: success, and returned the JSON. One line in it was useful — Domain Label: cricket_asia. Everything else was empty. Article title N/A. Source N/A. Type unclassified. One-sentence summary blank. Author stance and purpose N/A. And the list of information points — the list the entire downstream structure was supposed to be built from — came back with zero elements.

The system did not say anything had broken. It said: done.

Inside a franchise analysis room, that sentence translates into something very concrete. The matchup board does not get built that day. The meeting happens anyway, because the meeting calendar does not wait for data. And when there is nothing on the table, the arguments come from memory — "this lad made runs here last time", "he looked in rhythm during that spell". Memory has no error rate, which is why it never loses an argument.

Context

Asian cricket writes about itself more than any other sport I can think of. Board releases, franchise media teams, Urdu sports pages, transcripts of Tamil commentary, Sinhala newspaper reporting, Bengali digital desks, and on top of all that the English scaffolding of wire copy, which has usually already been filtered once through the original language. Stage-1 is the layer that turns this pile into structure: sentences into claims, claims into evidence, evidence into timestamps. Everything downstream is built on it — the argument chain, the strength of evidence, the direction of bias, time sensitivity, citability.

When that layer returns zero, four causes come first, and they need to be told apart because the cure is not the same.

One: the body text never arrived. Anti-bot walls, JavaScript-rendered pages where the text is assembled in the browser rather than on the server, paywalls, redirects to a consent page. The content exists; the capture failed.

Two: the language and encoding wall. The extractor was written for English and the source is Urdu, Sinhala or Tamil; sentence segmentation collapses, encoding degrades, and the whole document reads as "empty". In the cricket-Asia context this is the most likely technical cause and the least inspected, because the log shows 0 nodes rather than encoding error.

Three: selector drift. The site is redesigned, the old CSS path now matches zero nodes, and the pipeline quietly produces an empty structure instead of throwing.

Four: the input was hollow to begin with — only a title or a URL reached Stage-1, not the text.

There is a fine point inside the third cause worth stopping on, because it is a matter of discipline more than technique. The empty list threw no exception. Which means the pipeline does not contain the assertion that should say: "information points = 0, therefore this run has failed, therefore nothing is sent downstream." That silence spreads into every later stage. Source quality has to be assessed — there are no points. Time sensitivity has to be assessed — there are no timestamps. Entities have to be identified — the instruction says "identify them from the information points above", and there are no points above.

This is where what I call the empty-sheet tax begins. What cannot be assessed gets written into the report as "not available". Technically true. Practically a lie, because a reader does not distinguish "not available" from "not found" — and reads both the same way and moves on.

Core

In 2026 the truth lived in a column of numbers. I was 33 and I talked my way into Lancashire's first dedicated video analyst post — nine thousand a year, part-time, in a converted storeroom at Old Trafford with a VHS deck, a borrowed laptop and a kettle. I hand-coded 1,847 deliveries from sixteen County Championship matches, including the Aigburth fixtures played in my own city. Lancashire's seamers averaged 24.1 when the keeper stood up to medium pace on dry pitches and 38.6 when he stood back. The club made the post permanent within a month.

That number was trustworthy because it was precise — that is half the story. It was trustworthy because the counter-case sat in the same paragraph. Now imagine an event that never happened: if a few of the Aigburth matches had silently returned zero rows, the average would have printed anyway. And it would have been wrong in a specific direction, because keeper position on dry pitches was exactly the condition separating 24.1 from 38.6. The missing rows would have fallen precisely on the variable the whole argument rested on.

This is the point, and it should be stated plainly: not found and not there are not the same thing, and the extraction layer reports them identically. A wrong column gets caught, because someone notices a discrepancy and complains. A missing column does not get caught — it just changes the decision, and decisions never explain their own ancestry.

In 2026 I joined Mumbai Indians for a season as opposition analyst, building a death-overs matchup board around Wankhede's short square boundaries. Across sixteen matches their death-over economy fell from 9.6 to 8.2 runs per over, and they beat Chennai Super Kings by 23 runs at Eden Gardens for their first IPL title. I watched the final from the analysis room, not the stands.

That board was not a philosophy; it was a list of forced choices. At 34 for 3 the captain has three fields and only one of them is honest — a deep point on the short square side, or blocking the middle, or pushing a fielder up and burning the last powerplay over. Choosing between them rests on columns of paper: which end the seamer is operating from, his release height, how worn the pitch is, which batsman hits squarer under death pressure. If one of those columns had come back N/A — say the end — the board would still have run. It would have picked a field, just the wrong one, and the reason would never appear on screen.

At the auction desk the same disease is easier to see. In 2026 I left full-time coaching and started a paid tactical newsletter from a back bedroom in Aigburth. By the 2026 mega auction I had 4,200 subscribers and my role-scarcity model was pricing "finisher who bowls two overs" at 2.3 times base. I live-coded all 169 sold lots from Bengaluru. Rajasthan Royals paid ₹11.5 crore for Jaydev Unadkat, and I wrote that night: this is a scarcity premium, not a talent verdict. Before the season, not after.

An auction is a market with a heartbeat, and the heartbeat is fear. The real divider is not the player, it is the purse — the fifth bowler's price is set when the fourth has already cost too much. But here is the part modellers rarely write down: if I had 169 names and no prices, the model would still have run. It would have produced a number. And that number would have said nothing about the empty cells.

That is the danger of partial success, and it is worse than total failure. Total failure makes noise — nothing arrived, someone writes, someone asks. Partial success prints a table. Nobody can see the hole in a table.

During the 2026 bio-secure summer I consulted remotely on Lancashire's red-ball planning and coded all 60 IPL matches from my Liverpool flat, watching on a 90-minute delay to skip the ad breaks. My sheets said chasing sides won 62 per cent of matches against 48 per cent in 2026, and that spinners found late drift in empty grounds because bowlers could finally hear their own feet. I published "The Silence Tax" in October; three franchises asked for the raw files.

Go back to that 62 per cent. A sample of 60 matches. If three games had quietly dropped out — a timeout, an abandonment, a scrape failure — 62 per cent would still have printed. Nobody would have noticed, because the absence of three matches is not written on any document. A percentage printing is not the same as a percentage being right. That is why I publish ranges rather than points, and put the sample size before the conclusion, not after.

One football comparison earns its place here, provided it is tested at the actual constraint. A side with 60 per cent possession and three box entries has the same disease as a 62 per cent win rate with three matches missing. In both cases the fault is identical — nobody checked the denominator. The analogy holds only that far. If it stretches into importing football biomechanics onto a seamer's spell, cut it.

Contrarian

The reflex is to call an empty result a bug. Some of it is. The more useful reading is that an empty cell is the most honest object in the pipeline. An empty cell is at least not lying. The industry's real problem is not missing information; it is fabricated completeness — the cells filled with a plausible default so the table looks finished.

And there is a social asymmetry nobody writes in technical language. The analyst who says "I have nothing" gets flattened in the room by the analyst who says "my model says" — even when the second model is running on three-quarters of a dataset. Silence loses arguments. Confident completeness wins them, and the bill arrives six months later at the end of a season, when nobody can trace where the error began.

The second contrarian angle is the label. cricket_asia is itself a problem, because it does not separate two different things: is the subject Asian cricket, or is the publisher based in Asia? In the first case the analysis runs on a sports-journalism frame — bowling, field geometry, sample, conditions. In the second it becomes a question of media geography and business — who is buying coverage in which language, who holds the rights in which market. Both analyses are legitimate. Running both off one label is not, because right now there is no basis for deciding which applies — and that is the most dangerous state of all: not ignorance, but the pretence of having settled it.

Takeaway

I learned more about pressure from a whiteboard than from any dressing-room speech, and that whiteboard had one rule, which I will repeat here: do not read an empty cell as a zero. An empty cell is a question; a zero is an answer; fill both with the same colour and the whole board becomes a lie.

So what comes next. My threshold is written in advance and I will not move it. If the fetch log shows a non-empty body arrived from the server and the extractor returned zero nodes anyway, that is selector drift — the text exists, the path is blocked, and it should be opened. If the body itself was empty, then the story does not exist, and any article written on it would be invention, however elegantly it reads.

The code was simple; the game never was. And the next time a sheet looks complete, ask: how many cells in it were empty, and who decided to call them zero?