The Empty-Field Tax: When a Cricket Data Pipeline Returned Nothing and Said Nothing

At 1 a.m. in Sydney the screen did not show a scorecard. It showed a JSON — nine fields. One field held a single word: cricket_world. The other eight were N/A, or instructions that did not know their own address. Against Entities Involved it said: “Identify from the information points above.” Above, the information-point list was empty. Against Source Quality, the same: “Judge from the source field of the information points.” There were no sources. An instruction was pointing at an empty space, and the empty space was pointing back.

I watched that 2026 final at Lord’s on a lagging stream — England 228/7, India 219 all out, nine runs short, Anya Shrubsole 6/46. I paused on every wicket and sketched where the six ring fielders actually stood. That habit gave me a rule I still apply: when a frame shows an empty position, the first question is not what was there — it is who was supposed to fill it and why they did not. The JSON asked the same question.

Context: three floors and an empty basket

A modern sports desk has three floors. The first — Stage-1 — lifts the raw material, where information points are extracted. The second orders them: article type, author stance, time sensitivity, source quality. The third pours language over the frame. Each floor assumes the one below it did its job. In cricket terms this is the toss, the pitch report, then the eleven. If the toss never happened, the eleven is not “yet to be named” — it is unknown. Printing “yet to be named” is not a fact; it is a fabricated fact with a byline.

cricket_world is a tag, not an article. It names the domain, not the genre. A report and a comment piece are read differently: the report asks what the required rate is, the comment asks who scored how many. The difference is not stylistic, it is mechanical — required rate versus milestone. If the pipeline cannot decide whether it is holding a report or an opinion, it cannot price the piece either.

Core: two defects, one root

Defect one: the information points were not extracted, and the system stayed quiet. This is an old error in data work — zero and null are not the same thing. Consider a fielding grid. A fielder is logged with zero stops. What does that tell you? Perhaps he is excellent and the ball never came near him. Perhaps he was asleep. Perhaps the person filling the log was. The grid cannot separate those three — that is the grid’s limit, and I accept it. But a pipeline can be expected to write “nothing was found” differently from “there was nothing”. It did not. The empty list travelled onward as a valid value.

The consequence is direct. Every downstream field is derived from the information points. With zero points, those fields have nothing but N/A — or worse, no idea what they are looking at. That is the moment a pipeline stops doing data work and starts doing guesswork. And guesswork never looks empty; it always looks full.

Defect two: circular dependency between fields. Entities Involved and Source Quality are both defined as “derive from the information points.” Both roots sit in the same basket, and the basket is empty. There is no fallback, no “if empty, do this.” Think of a batting order where every position is defined as “the man after the previous one.” One wicket and the whole card collapses. Stage-1 was designed on the assumption that information points are never empty — an assumption with no pre-condition check behind it.

Now the sample size, because I write it first by habit. This is one JSON, one run, n = 1. I cannot claim from it that the pipeline always fails. What I can say is that in this one run the failure was not loud — it was silent. In this trade a loud failure is a gift; the silent failure is the actual cost.

The cost demands a name. Suppose the editor downstream treats this packet as complete. They write a headline. They write a caption. A story outline takes shape with no subject. And the moment the headline is written, a fact is born with no source behind it. Someone is billed for that omission — on someone’s byline, under someone’s name. I call it the Silence Tax, because it is money paid deliberately: labour that went unpaid, and an error that was paid for.

This is not the first time I have written about work not done rather than work done. In 2026, with stadiums empty, I built a dataset of all sixty IPL matches in the UAE. The question was whether home advantage survives without a crowd. The result: nominal home sides won 47% in 2026 against 55% in 2026. The same dataset showed something else — with the crowd gone, the stump mic opened up. Keepers called field settings aloud; batters called runs. I used it as a primary source. That piece sold for A$250, and the dataset became the pilot for my first lab job. Had the extractor returned an empty list that night, the piece would never have existed — and nobody would have seen the loss.

In the summer of 2026 I borrowed football’s grid the same way — an eighteen-zone pitch map, first on the IPL final, then on the Russia World Cup. Spain logged 1,029 passes and lost on penalties in the round of 16. England scored nine set-piece goals. I sent 900 words to The Roar, was paid A$80, and got 40,000 reads. The lesson was one line and it still holds: the first sentence names the mechanism, never the result. “Croatia’s midfield survived by occupying the space Russia kept vacating” is a sentence. “Croatia were brave” is not. Back to the JSON: what was the mechanism? An empty list accepted as valid. What was the result? Eight N/As. If I write only the second, I am repeating the error I am reporting.

One clarification: no cricketer has a role in this report. No team, no series, no toss. It is about the machinery that produces all of those stories. When the machine goes quiet, stories do not decrease — they appear to be produced, only with an empty foundation. The data desk won the grid-versus-commentary-box argument years ago; what has not arrived is the habit of audit — of printing the result that exists. Since the result here is “nothing was found,” the audit is the article.

Contrarian: the urge to fill the void

The most dangerous line in the whole output is not an N/A. It is this: “Identify from the information points above.” That is an instruction, and it says: because something must be written below, put something above. It looks like diligence; it is a blank cheque. A pipeline forced to fill empty space soon learns to fill in facts that were never there.

This is the same move the commentary box makes daily. Asked for the required rate, it keeps adding hundreds and five hundreds. The difference is that the man in the box knows what he is doing. The machine does not. My professional instinct says be strict here — but my other instinct, the one that runs the other way, says occasionally publish the boring confirmation. Sometimes Stage-1 runs clean and returns a piece with nothing new in it. A model that turns emptiness into achievement quietly destroys its own credibility.

One honest limit: my audit is bounded too. It can say which fields held what, and which field was absent. It cannot say whether the source article was sound, because I do not have the source article. Just as the grid cannot see seam position, wind, or a captain’s gut, this audit cannot locate the fault in the source. And I know this is exactly where a writer would want to thicken the story. I am not thickening it.

Takeaway: next time, a receipt for the silence

On the next run, ask the pipeline for one new thing: a receipt for its own silence. The field that carries information points needs a status flag alongside it — extraction_status: empty, with an error code — so an editor downstream can tell “there was nothing” from “nothing was searched for.” Until that flag exists, part of the trust stays empty. The real question, to be repeated over every silent output: when the machine reports that nothing was found, whose signature goes under the empty space?