Twelve Empty Fields, One Filled: An Audit of a Cricket Data Pipeline's Corpse

Thirteen fields. Twelve of them empty. The thirteenth carried four words — Domain Label: cricket_world. Everywhere else the same token repeated: N/A. Article Title N/A. Source N/A. Type "Unclassified". One-sentence Summary blank. Author Stance N/A. Information Points an empty list. Time Sensitivity "not assessed". And in the field named Entities Involved, an instruction sat where a value belonged: identify from the information points above.

That is an output. An output is what a pipeline returns when it has found nothing to return. The field was to be filled from the information points above, and there were never any information points above. This is not an error, it is a job quietly completed — a template string seated in the position reserved for a result, then published in the costume of N/A.

I first met an empty scorecard on the night of 13 March 2026, in the scorers' box at Eden Gardens. India, chasing Sri Lanka's 251, had slid to 120 for 8 in 34.1 overs. I was the second scorer, hand-coding every delivery. Not one cell on the sheet was blank — runs, overs, wickets, bowler breakdowns, all filled. Then the bottles came, the match was stopped, Clive Lloyd awarded it to Sri Lanka, and I sat there holding a complete sheet that could not answer a single question I had.

Within a week I had added three columns. Required rate. Wickets in hand. Bowler spell length. That was the night I learned a full sheet can still be an incomplete one. Pressure is a logged variable; outcomes are just noise.

The question now is how common this thirteen-field table is in cricket's own record. The answer — very common, and we never call it N/A.

Cricket's archive is a building whose foundation is scorers, people whose names never appeared in a byline. Thousands of Ranji and Duleep Trophy matches played before 2026 exist on paper that was never scanned. The ones that were scanned carry innings totals, not ball-by-ball. Ball-by-ball in India is generally traced to television commentary logs of the late eighties, and those logs are full of cells marked unavailable and disputed, cells no modern website displays.

That is my second old habit — provenance before names. Before any number I ask who logged it, under what pressure, and what fell out during transcription. When the answer is "unknown", correct engineering leaves the cell empty rather than filling it with a mean.

In 2026 that habit bought me five months of work. After Hansie Cronje's confession in April 2026, I coded every one-day international India played between 2026 and 2026 — 412 matches — logging no-ball counts, dead-rubber status and dismissal types. No-ball frequency in matches with no consequence ran roughly forty per cent above the sample mean. I handed the method, not the names, to an investigative reporter in Kolkata. The database had 412 matches; the byline had someone else. The road to the CBI report of November 2026 ran through that method, and in December 2026 Mohammad Azharuddin received a life ban. I did not publish the player-level cut myself, because a conclusion that cannot be reproduced is not a conclusion worth printing.

Read this table in the same spirit. Its N/A conceals no name. It reports that no method ever ran behind it.

There are three separate failures here, and collapsing them into one is the real mistake.

The first is an extraction layer returning empty. Much of cricket's web now runs on scorecard and live-blog templates rendered in JavaScript. A plain HTML scraper returns an empty string from them — not an error, an empty string. Every downstream field settles onto its default, and the default's name is N/A. An empty string and an unavailable statistic sit in a table in the same shape, and that is the actual bug.

The second is a short-circuit after classification. "Unclassified" is not a category; it is a report of classification failing. But the branches that extract summary and stance hang off the category. No category, no branch. Rather than falling through to a generic path, the system chose silence. A pipeline that cannot say "I don't know" will publish its own ignorance in the shape of a data point.

The third is an instruction left standing where a result should be. "identify from the information points above" is a work order, not a finding. Its presence in the table says plainly that the step meant to fill that cell never executed. That is process evidence, not a gap in data — like finding "to be entered" written against an over in the scorebook.

Now return to cricket, because all three failures already exist in our game's record.

Indian domestic cricket's lost stretch — no video, no ball-by-ball, not even a newspaper clipping surviving — has been handled by us in the same way. Nobody called it empty. We filed it under "cricket of that era" and moved on. I stopped scoring outcomes the day Eden Gardens stopped mattering. From that night, the match result became just another column on my sheet, as dull and as useless as the rest of them.

On the other side sits the 412-match database. It was incomplete too. It excluded washed-out matches, matches with disputed scorecards, matches where the footage spelled a bowler's name wrong. The difference was that every exclusion was written down — sample size, exclusion rules, and the portion the model cannot see. That appendix was the method, not a boast. This table carries no such appendix, which is why it cannot support analysis, and why it is itself the analysis.

The easy move is to blame the scraper. The scraper is at fault, and it is not the first fault.

When a system cannot tell you what it read, every number it produces deserves suspicion, including the ones that look filled. This is the correlation-causation line: an empty field is evidence of absence of data, not evidence that data never existed. Treating the two as one is like recording a rained-off match as a defeat.

Cricket commits that error constantly. We treat scorecards as ground truth when they have always been the final product of a human process — a scorer, a stringer, a digitisation volunteer, an editor with no time to check. When that process breaks, the scorecard does not look broken. It looks tidy, merely a little empty. A model is only as good as the question you ask it — and the wrong question was asked here: which category the article belongs to, instead of what the article was about.

The most useful thing about this table is not its failure but its honesty. It did not guess. It did not fill empty cells with averages, stale data or the word "likely". A great deal of automated cricket reporting does exactly that, and it is far more dangerous, because nobody recognises it as N/A.

If a pressure statistic appears in a match report this Sunday, ask where it came from and what was dropped on the way. The day cricket coverage grants N/A the same respect it grants a wicket is the day analytics stops being a word to fear.

The signal for the next round is simple: go underneath your favourite number. Do not fill the empty cell. Leave it empty, and write down why it is empty.