The Season of N/A: When Cricket's Archive Comes Back Empty

It is two in the morning in Bangalore. A spreadsheet is open with a single row in it. Eight fields. Seven of them read N/A. The eighth reads Unclassified. And at the very bottom, the only field holding a value at all — Domain Label: cricket_world. That is the whole thing. A complete cricket report, or whatever it was meant to be, has collapsed into eight cells and one word: cricket.

This is not a scorecard. It is the output of a Stage-1 extraction pipeline whose only meaningful answer is cricket_world. I opened it more than once, because I have learned to recognise an empty shell quickly.

In July 2026, after the Lord's final ended — England 228 for 7, India 219 all out, a nine-run defeat — the desk asked me for 150 words of wire copy. I filed 1,400 instead: on Harmanpreet Kaur's 171 not out off 115 balls at Derby three days earlier, and on the fact that nobody on the desk had watched that innings live. That file ran at 2 a.m. and travelled further than anything I had written. What I learned that night is that an empty field often tells more truth than a filled one, because it tells you what nobody bothered to fill.

The same lesson has come back, this time delivered by a machine. And it is not merely a technical accident. It is a small sample of a much larger structural problem in cricket coverage: we have built systems to read material that, in many places, was never written.

Context: one pipeline, eight fields, zero information

Stage-1 had a simple brief. Take any cricket article, record its title and source, identify its type, produce a one-sentence summary, determine the author's stance and the article's purpose, list its information points, identify the entities involved, assess time sensitivity, and weigh source quality. Eight tasks, eight fields.

What came back: no title. No source. Type Unclassified — meaning classification never happened. Summary empty. Author stance N/A. Purpose N/A. Information points an empty list. In the entities field, an instruction sentence. Time sensitivity unassessed. Source quality resting on a template phrase telling the reader to judge from source fields that do not exist.

Reading that row makes one thing clear: the failure is not in the analysis. The failure is upstream. When the input is empty, every downstream field fills with its default, and the default is N/A. The term means not applicable; in practice it becomes a polite stand-in for we did not look. On a dashboard the row appears complete. The process appears successful. The information content is zero.

Cricket is the heaviest data-producing sport in the world — a record for every ball, a scorecard for every innings, a page for every series. That is precisely why it is also the hardest to extract from. Scorecard templates have changed repeatedly over the years. Live-blog pages render through JavaScript, where the raw HTML contains no story at all. Domestic season results still sit, in many places, inside scanned PDFs. In the women's game, these layers stack on top of one another.

— Root: The 2 A.M. File from Lord

Read the shell field by field, because every empty cell tells a different story.

Title and source returning N/A is the crawler-and-extractor layer. If no name was recorded at all, the pipeline may have reached the page and still failed to select the main content. On cricket sites, the main content is rarely inside one clean container; it is scattered among score widgets, live-update feeds and advertising blocks. When the selector misses, you are left holding an empty string.

Article Type Unclassified is worse than N/A. N/A says the information was not found. Unclassified says the classification engine ran and could not commit to any category. And when it cannot commit, it does not fall back to a general category — it falls to an empty one. That is not unusual for news text: a match report, an analysis piece, an interview, an auction report, an institutional statement all have different anatomy. But it raises the real question: why is there no general fallback category that says I read this, it is news, I simply could not determine the sub-type?

The empty summary and empty information points are the most serious layer, because they show the pipeline never read the material. Summarisation is the easiest task a language model performs. Given actual text, it would return at least one sentence. An empty list means there was nothing to carry forward.

The sentence that never ran

The most revealing field is the entities field. In place of results, it holds this sentence: identify from the information points above.

Read it two ways, and both are troubling. First, it is an instruction, not a result. In any properly built pipeline, such a sentence should never appear in final output; it belongs in the prompt or template, sent to the model and then replaced by the result. Second, it proves the field above it — information points — was already empty before this sentence was meant to run, and nobody stopped the emptiness; the process simply moved on.

This small-looking detail is the entire diagnosis. When instructions and results start occupying the same field in a system, assume some part of that system never ran. It is a template-string bug, and it is a design omission, not a data gap.

Three failure modes, three layers

From the available diagnostic, three probable causes can be ranked by likelihood.

Most likely is upstream crawl or extraction failure. Main text was not selected from the source page, the extractor returned an empty string, and every layer beneath filled in defaults. This is the most common failure in cricket coverage because a single page often carries three or four different technologies at once — static report, dynamic scorecard, live blog, embedded audio and video. A selector written for one will silently return empty-handed on the rest.

Second is the classifier short-circuit. In many pipelines, summary, stance and purpose hang off the type branch: first decide the article's type, then fill the remaining fields accordingly. When the type comes back Unclassified, the branch is never found and the remaining tasks are abandoned without being attempted. This is silent null propagation — one empty field empties the next, which empties the next.

Third is the template-string bug, direct evidence of which sits in the entities field. When a template fails to clear itself, an instruction sits in the output dressed as a result. It is the smallest bug here and the easiest to catch. It required only reading the output once. Nobody did.

All three share one property: none of them failed loudly. No error surfaced, no warning printed. That is the greatest danger. A system that fails loudly gets fixed; a system that fails quietly keeps earning trust.

Silence is not a data type

An old habit from reading scorecards is useful here. A scorecard has fields that look empty — not out, and did not bat. They look identical and mean opposite things. Did not bat means the batter never reached the crease. Not out means the batter arrived, held on, and the innings ended around them. Two empty cells, two different lives.

A machine's N/A does not distinguish between them. That is the real damage. When a default value climbs onto a dashboard wearing the clothes of a finding, the decision-maker concludes the system looked and found nothing notable. In fact the system did not look. The distance between those two conditions is, to my mind, the most expensive distance in the cricket-data industry.

Which is why every pipeline needs a state it currently does not carry: not read. It is distinct from empty. Until that state exists, every report that looks successful is probably the report in which nothing was read.

The corpus nobody wrote

In June 2026 I was placed on the Russia World Cup night desk — thirty-one nights of somebody else's world. I had no feel for that tournament, and I spent thirty-one nights sorting its copy. In those same nights, three women's cricket pitches died in the queue. Two months earlier, in May 2026, I had been one of eleven accredited reporters at the one-off Women's T20 Challenge at the Wankhede, an exhibition match played before an IPL qualifier. The gap between the two rooms broke something in me. I stopped filing almost entirely and reread every notebook I had kept since 2026.

Then came 8 March 2026. The Melbourne Cricket Ground. Australia 184 for 4, India 99 all out, beaten by 85 runs. That day 86,174 people were inside the MCG, still the record attendance for a women's cricket match. Two weeks later India locked down, stadiums emptied, and the outlet I wrote for closed its women's sports vertical. I was furloughed for eleven weeks.

In that silence I re-watched four years of footage and found the same flaw in every Indian chase: middle-overs scoring rate. It became the tactical spine of my first book contract — an argument, not a tribute.

I recite this history because the empty shell adds something to it. Automated extraction cannot work until there is something written to extract. And the pre-2026 record of Indian women's cricket sits largely in scanned sheets, hand-filled scorebooks and local newspaper clippings from matches for which no video was ever made. No pipeline will pull anything out of that, because there is no structured data there to pull.

With the Women's Premier League from 2026, Indian cricket finally acquired a structured, regular, machine-readable data source. Which produces the paradox. The most easily extractable document in Indian cricket is the auction sheet — names, base prices, sale prices, dates, franchises, all in tidy cells. And yet the column least often printed after any auction is the unsold one: the names of players who carried a ten-lakh base price and were picked by nobody.

How to set a canary

If you run a pipeline like this, add three checks today. One: count the character length of extracted body text. If it is zero, or under a hundred characters, the problem is not in the model but in the crawl or extraction layer — fix it there. Two: inspect every field as though it might be an instruction. If a field contains verbs like identify, extract, take from above, the template was never cleared. Three: run one article through twice and compare outputs. If both runs return all N/A, that is not not applicable. That is did not run.

The value of these checks is that they measure the existence of data, not its quality. And in cricket coverage today, the biggest problem is not quality. It is existence.

The contrarian turn: the field shown, and the field withheld

It is easy to conclude this is all technical — fix the crawler, rewrite the template, job done. My first conclusion is the opposite. A transfer fee is a public number; the private suitcase is the real story. The figure that gets printed is always the one whose printing costs nobody anything. Every transfer window, the entire market chases the sale price — public, clean, easy to extract. The number that never fits a field is the same player's morning bus, the month without a contract, the fixture that was cancelled. No pipeline catches those, because nobody publishes them.

Second, and more uncomfortable: the empty shell is itself an answer. One reason the source may have come back blank is that the article did not exist, or its substance was never written in a form a machine could lift. The failure is not only in the code; it is in the corpus. And corpora are not multiplied by code. They are multiplied by whoever writes the 1,400 words nobody commissioned.

It is a mistake to think that writing simply resumes after thirty-one nights spent inside someone else's world. It took me three weeks of near-silence, and in that silence I recorded sixty hours of interviews — with domestic players nobody had ever asked about their technique. That is my private archive, and it will never appear on a dashboard. But any system that wants to read Indian women's cricket properly will have to build archives like it first.

Forward

So for now, that row. Seven N/A, one Unclassified, one cricket_world.

It is not the outcome of a great discovery, and trying to make it one would be exactly the error I write against. But it is a small mirror. The day India's domestic women's calendar, its match-fee payment dates and the unsold list from every auction are published as public, structured data, pipelines like this will finally have something to fill themselves with. Until then, whatever output arrives will be the result not of the game but of our silence.

And I will keep writing the pause after the roar — where the athlete becomes a person again, and where the real ledger begins.