N/A: The Cricket File That Came Back Empty — And the One Thing It Got Right

2:07 a.m. The rain in Delhi had stopped, and the only sound left was the hum of a neighbour's air conditioner. On my screen was a file with eleven fields. Ten of them said “N/A”, in capitals, with no comment attached. The eleventh said two words: cricket_world.

For twenty-six years I have begun with the document, not the claim. The habit was set in the winter of 2026, when I laid 41 pages of inquiry records alongside 19 phone-tape transcripts and read them in sequence — because truth hides in the order of documents, not in their tone. That night the file arrived carrying no argument at all. It carried an emptiness, and the emptiness was so cleanly stated that ignoring it would have been the harder choice.

The most honest sentence about any story can be: we could not establish what happened. Even that honesty needs a frame to be recognised. That night the frame itself was lying broken in front of me — and a broken frame is also a document, provided you are willing to read it.

In a transfer window, the cricket newsroom changes shape. Retention lists, right-to-match rules, purse arithmetic, agent calls, injury updates, and over all of it a thick sediment of rumour. The reader is drowning and does not need more rumour; the reader needs a reliability filter. That is why nearly every large outlet now runs a quiet machine — a pipeline that picks up each published article, breaks it into fields and prepares it for downstream analysis. Title, source, type, one-sentence summary, author stance, information points, entities, time sensitivity, source quality.

The pipeline looks cleaner than it is. Cricket is the worst possible input for it. Cricket sites run on live blogs, on scorecard widgets, on a DOM that is rebuilt every over. An extractor looking for paragraphs finds a box that says “loading”. The parts that matter most — run rate, required rate, who bowled the twelfth over — arrive through scripts and never as text.

This has worried me for a while, and the worry is old. On 20 February 2026, in a Mumbai hotel ballroom, I watched eight franchises with a five-million-dollar purse each buy 78 players in under seven hours. Mahendra Singh Dhoni went first, to Chennai, for 1.5 million dollars, and I filed 900 words on deadline and a 2,400-word scene piece by two in the morning. For that auction I had built a one-page card per name — Ranji average, age, injury history, previous season's strike rate. Auction day taught me that a transfer window has a heartbeat, not a deadline. The card worked because a human was filling it, and the human had been at the nets that morning.

The difference is exactly there. Behind every field on that card stood a person taking a risk of being wrong. Behind this new file, at two in the morning, stood nobody.

I read the file twice. The first time to see what had survived. The second time to see what was missing. The second reading was the useful one.

The title field was N/A. Without a title there is no identity. You cannot deduplicate, you cannot link the piece to an older file, and six months later you cannot prove this was the item that moved the market. A title is not decoration; it is a passport. The file had lost its passport.

The source field was N/A. That is the first link of the audit trail, and it had snapped. I could not say whether the article belonged to a newspaper, a website or an agency. On a beat that runs on board-adjacent access, the source field is not a formality. It tells the reader whom you read and whom you skipped.

The type was “Unclassified”. That is not a classification; it is a failure wearing a category's name. The summary field was empty, which means the compression step never ran. The author stance was N/A, which means the reader could never know whether the text was a report, an opinion, or a polite translation of a press release. On this beat, that is the difference between journalism and a favour.

The list of information points was empty. That was the actual cargo. Every other field exists to carry it. No cargo, empty truck.

Then came the field that changed the meaning of the whole file. Against “entities involved” was written: identify from the information points above. That is not an entity. That is an instruction. This is the loudest clue in the document — a field holding a command where a result should sit. The schema fell back to its template, which means the step that was supposed to fill it never executed.

Time sensitivity had not been assessed at all. Against source quality sat the same kind of instruction: judge from the source fields — when there were no source fields to judge from. Twice, in two different places, an imperative survived into the output. That is not chance.

Missing data and a broken process are different illnesses, and they present differently. When data is absent you get null, sometimes with a reason attached. When the process breaks you get an instruction sitting in the output — as though a blank form had been submitted with “write your name here” still printed on it.

There are three plausible causes and the order matters, because the fix lives in a different place for each. First and likeliest: the fetch layer returned nothing. Scorecards and live blogs on cricket sites are built by scripts; when the content selector misses, the extractor receives an empty string and every field below settles onto its default. Second: the classifier called the type “Unclassified” and the branch terminated there, because summary and stance extraction hung off the type branch and never got a fallback path. Third, and the one that stings: the schema was filled from template defaults without any field being executed individually.

All three share one cheap test. Run that single article on its own and check whether the extracted body length is zero. If it is, the repair belongs in the fetch layer. Adjusting the model or its prompt is wasted work. A pipeline whose front door is shut cannot be fixed by renovating the rooms inside.

The newsroom consequence is that an empty extraction does not stay empty. Empty space gets filled, and it gets filled first by the loudest voice in the room. The desk falls back to the group chat, a name surfaces, a number attaches itself to it, and by morning somebody's byline is standing on it. That is how a transfer window actually operates: when information is scarce, rumour is the fastest object in the building.

Remember the night of 2 April 2026. The Wankhede, 33,000 people, Dhoni's 91 not out closing a six-wicket win in 48.2 overs. At 11:40 p.m., forty minutes after the finish, I was in the corridor outside the Indian dressing room, having spent the last hour collecting 27 short interviews — stewards, a groundsman, two players' fathers. At 11:40 the trophy was already history; the room was still deciding its meaning. The scorecard had told me what happened. The room was telling me what it meant. A match pipeline has two fields, and that night my file held neither.

On the pauses in rooms I will say only this. I stopped counting minutes the night I started counting the pauses in a room. In a newsroom, the pause is the blank field — the place where the file stops before it says anything, and nobody bothers to fill it.

On 8 January 2026 India lost the opening Test at Cape Town by 72 runs. At 2 a.m. IST I filed 3,200 words tracing the collapse back to a nets session I had watched three days earlier. It drew 1.4 million reads, my largest audience in eighteen years. It worked because a page of my notebook was reserved for nets detail — a field I had added to my own template after being burned often enough. Every good template is a scar; every bad template is a scar someone forgot.

The scorer's pencil in the stands remains the document I trust most, and not because of its accuracy but because of its corrections. A scorecard can be amended — a name, a run, a catch. A website cannot. An extraction pipeline does not even build the field in which an amendment could live. The scorer's pencil told the true one, because it was allowed to rub things out.

Which brings me to the part of this I actually care about. An empty file is always better than a lying file, because an empty file can be audited later.

From the outside this reads as a machine failing. I read it the other way. In the entire chain, this is the one component that did not lie. It looked, found nothing, and said so. That is humility, and on this beat humility is rarer than talent.

In 2026 the dangerous documents were never the ones with blanks in them. The dangerous ones arrived with a tidy story already written on top — every field filled, every name settled, every date impeccable. When a file is too clean to argue with, the argument is being had with you. A pipeline that returned a confident summary in place of this empty shell would look like a triumph on somebody's desk this morning, and it would be manufacturing one unsourceable line a week.

The deeper failure sits upstream and it is human. We read less, because forty articles arrive every hour and the day has twenty-four. So we hired a machine to read for us. And then we did not trust its silence. All three things happened, and the third cancelled the first two.

N/A is a verdict too. It says the machine looked and found nothing, which is not the same as saying there is nothing. Treating those two as identical is how a whole tournament can vanish from coverage because one selector did not wait for the JavaScript to land.

And the entities field is the sharpest joke in the file. A human would have typed a name and then doubted it — is it this spelling, this district, this player who was with another team last season. The template typed an instruction into its own answer and doubted nothing. That is the difference between a junior reporter and a schema: the junior reporter knows that he does not know.

Three checks before the next batch, all of them cheap. Log the extracted body length for every URL, and if it is zero, stop asking the fields below for anything. Make “Unclassified” a terminal error rather than a silent pass, so that it makes a noise. And search every output for imperative sentences — any field containing identify, judge or extract is a bug, not a finding.

The next transfer window will bring more pages, faster, and the same pipeline will be asked to read them. The question is not whether a machine can read cricket; it will, and sometimes well. The question is whether anyone in the room can still read it, and whether, when a field comes back blank, they will rush to fill it or stop to ask why it is blank.

Leave the blank line in the ledger. In November, that line is the one that saves you.