The Empty Notebook and the Full Ground: When the Extraction Pipeline Eats the Cricket Story

Sitting in the press box, eighteen years into this beat, what I encountered that afternoon from Brisbane to Auckland was not about cricket. It was about the cricket that was supposed to be written and never was. Two empty fields, one half-formed tag: cricket_asia. That was all. No title, no source, no one-sentence summary, the list of information points silent. This is the moment journalism's oldest fear stands up in front of you — the story existed, the ground was full, the crowd was roaring, and our machinery decided it was noise and cut it away. At fifty-five this stings more than it once did: I keep two notebooks, one for scores, one for sound. The day the first notebook comes back empty, what survives is not journalism but a template.

What has happened here is an extraction failure in technical language, but its real name is noise classification. The fields above are all empty — Title, Source, Type, Author Stance, Information Points, Entities, Time Sensitivity. The system has decided the text that arrived is not news but rubbish. When the navigation bar, the cookie notice and the paywall placeholder all arrive in the same container as the page body, the classifier throws out the entire page. In cricket coverage this problem turns sharper still, because a large share of our material is written in a language that does not look like 'data' to a machine — 'in yesterday's match', 'last week', 'according to sources'. Where English text shows clean dates and figures, cricket reporting in Hindi and other Asian languages often carries chronology in relative words. The extractor marks it time-insensitive and drops it, and the whole story disappears.

I remember the Mumbai auction of 2026. In that ballroom I entered all eight franchises' purses into my ledger — every bid, every remaining purse, every retention clause. Nobody said then that this was 'data'; everybody said it was 'cricket'. If that same ledger is fed into a system today and the system calls it 'unclassified', I have only one answer: you have not lost data, you have lost the story of the family whose son was changed by that bid. That is my first foundational point — information points are not merely information; they are the physical traces of the world we cover every day. When they vanish, what vanishes first is the thing corporate reports call the 'human angle' and what I call the man in the ground.

This failure has three layers. The first is technical: a missed body selector. If the page's main content is not in an article tag but in a div whose class is story-content or post-body, and your extractor only looks for article, it will find nothing. Then the length threshold — if you treat anything under 300 words as rubbish, short reports, briefs and fast agency copy all fall outside. Briefs matter enormously in cricket: 'so-and-so injured, out of the series' is thirty words, but it changes the entire eleven. If your pipeline calls that noise, your team-selection coverage will always be half-told.

The second layer is linguistic. Phrases like 'according to sources' are treated as weak in Western journalism, but in Indian cricket reporting this is an established style. In my own notebook I sat on a tape for nine days during the spot-fixing spring of 2026 because a young stringer's source was at stake. Every sentence of that tape began with 'according to sources' — if that text had gone through an extractor that dismissed it as 'unreliable, no concrete entity', my entire investigation would never have reached paper. My ethics of source protection and the machine's need for entity recognition collide at exactly one point: whether or not to name the source. The collision is sharpest in cricket journalism because behind every 'source' sits a fate — selection, contract, visa, family.

The third layer is systemic, and it is the deepest. When an entire report is left 'unclassified', what is really left unclassified is not the report but its audience. The voices of those 90,000 people at Eden Gardens in 2026 who did not rise even after defeat — those voices never enter any field. They are not information points, not entities, not time-sensitive. My second notebook exists precisely for this: what does not get recorded in the scorebook must be recorded somewhere else. If our pipeline reads only the first notebook, it cannot understand cricket, because cricket is half in the numbers and half in the noise that sits outside them.

A counter-question arises here, and I want to put it plainly: is it even right to treat this empty output as a 'failure'? To some extent, no. An empty output is sometimes an honest answer — a way of saying 'I do not reliably know what is in this text, so I will say nothing.' This is like the reporter who does not publish when a source cannot be confirmed. I did exactly that in 2026 and my editor was furious. But there is an important difference: a reporter explains in the empty space 'what I could not learn and why I could not learn it', whereas this output is pure silence. Silence and ambiguity are different things: ambiguity tells you how far you got, silence tells you nothing. A good system should be in the second state. It should say: 'The page was found, it held 1,240 words, but the body selector caught only 40; therefore I extracted nothing.' That is an answer a human can act on. Nobody can act on 'N/A'.

I will show the cost of this silence in cricket coverage with a concrete example. Suppose on Monday morning an Australian newspaper reports that a scan on an Indian fast bowler's wrist is worrying and he will not play the first Test. The story is 200 words, has no quotes, only 'it is understood'. If that story passes through your pipeline and is thrown out as 'no entity, no information point', then every team-selection analysis you publish that day will be wrong. You will assume someone else is playing in his place. This is not a small technical glitch; it is the foundation of the entire analytical frame on which squad development, the wage bill and transfer-window talk later rest.

My experience says the cure is at three levels, and all three come from journalistic principle. First — build a list of fallback selectors the way a beat reporter builds a list of contacts. When one source closes, go to the next; when the next closes, go to the third. If article is not found, try main, then class-based selectors, then the largest text block. Second — make the minimum-length threshold context-dependent, not fixed. Agency copy is short, features are long; do not cut both with the same scissors. Third — when nothing is found, report 'incomplete' instead of 'empty'. This is the same difference as between a good sub-editor and a machine: the sub-editor writes on the copy 'something is missing here', the machine goes quiet.

I want to repeat one thing in this whole argument, because I write about Indian cricket from Australia and this is the most honest part of my position: writing from a distance lets you see the shape of the board's power without standing inside it. The cost is that you sometimes play the sound of the ground inside your head and believe it is real. The same danger applies to an extraction pipeline: when the data is empty and you feel there must be something inside, you start filling it with guesswork. I did not do that, because guessing the story from the cricket_asia tag is the same sin as publishing without a source. Treating what was not written as written is cricket reporting's oldest disease — and it has now reached the machines too.

The lesson I took from the tape incident of 2026 applies directly here. At that moment I had a story that could have made headlines around the world, and I waited nine days because a young stringer's future was tied to it. Some could have called me slow. What I learned in those nine days was that the price of a story published in haste is usually paid by someone else. The 'empty report' published in an extraction pipeline's haste is also paid for by someone else — the reader who is startled the next day to read the squad announcement, and the analyst who built an entire strategy on a false foundation.

Now imagine the situation when this system works properly. The same day, the same morning, the pipeline says: 'Page found, 1,240 words, three information points, one entity — the bowler's name, a date, a mention of injury, source: the newspaper's sports department, publication date Monday.' Now work can be done on it. Someone can ask: is this the same foot that was operated on last year? Is this just before the series in which his contract expires? These are the questions that move sports journalism from 'score' to 'story'. And this is where I open my second notebook — because the most important thing in that injury report is not what the scan says; it is who the bowler's father spoke to on the phone that morning in his village.

The longest lesson of my career is this: technology can never replace the story, but if technology eats the story, the damage is real. In 2026 at Eden Gardens I saw 90,000 people refuse to leave after a defeat — that day I felt for the first time that a match report is not just a score. If a system had called that atmosphere 'noise' and cut it that day, I would not even have the memory from which I am writing this piece. Sport journalism's real capital is the details that look unnecessary at first glance — empty seats, unfinished sentences, held breath. If our machines treat these as rubbish, the next generation of readers will get only the scoreboard, not the story.

So my suggestion is clear: fix the pipeline, but while fixing it do not forget what you are saving. The page you are calling 'unclassified' was worked on for hours by someone — a reporter, a sub-editor, a photographer. It has a date, a source, a reason for existing. If your system cannot read it, the fault lies not in the text but in the way you read. And until that way is fixed, returning 'empty' is better than returning 'wrong' — but best of all is that we come back and say where we got stuck.

Tomorrow morning, when the next squad is announced, both my notebooks will be open. One holds names and figures, the other holds the sound that sits behind the figures. I have only one condition: the pipeline this all passes through should at least say what it read and what it could not read. I will do the rest — because a beat is not a list of events; it is a rhythm people live inside, and to catch that rhythm even the machine will have to be honest.