An athletics report with no names: the silent failure inside sports data pipelines

At three in the morning, a JSON file landed in the newsroom inbox. Every field was populated: title, source, article type, author stance, time sensitivity. Only the content section was empty. No athlete, no discipline, no mark, no competition date. A flawless description of something that does not exist.

I read it twice, then a third time, because professional habit makes me believe the fault sits in my eyes before it sits in the file. But the file was not wrong. It was simply hollow.

The nine-dimension analysis generated from that file did the first thing it could: it refused to work. Every cell carried the same line — insufficient information. No venue, no discipline, no starting line. A report about athletics with not a single metre run.

Then it named the only thing still alive in the file: the domain label, athletics — and the report itself doubted whether that label was a default value rather than a classification result.

To see why this is worth writing about, you need to know what the pipeline was supposed to do. The extraction step had to pull out athlete names, disciplines, marks, wind direction and speed, venue altitude. The analysis step had to place those fragments onto the sport's coordinate system: world records, qualifying standards, world rankings, the three-per-country quota. A mark can only be read as true ability when the wind stays at or below +2.0 m/s, when the venue sits under 1,000 metres, when the shoe is not the carbon-plated model with a 40 mm sole that was legitimised a few seasons ago.

That step also had to answer harder questions: where the athlete sits on the career age curve — sprinters mature between roughly 24 and 29, middle-distance runners later, throwers later still; how often they have raced this season; whether there is any sign of an abnormal performance spike that should be cross-checked against anti-doping screening.

None of those cells had data to fill it. But one detail deserves a longer pause: the template instructs the analyst to list entities “from the information points above”, while the information points above are empty. A self-contradicting instruction, inside a single file.

Three hypotheses were offered, and all three lead to the same place. Either no article was ever passed to the extractor — a fetch failure. Or an article existed, but the model returned a schema-correct empty object — a parsing failure. Or the article exists but has nothing to do with athletics, and the domain label is a leftover default.

What makes all three equally worrying is the shape of the file: it is valid. It clears every formal validation gate. A file that reports an error stops the line and someone has to fix it. A file with every field populated but no substance travels straight to the next stage, and the next stage does exactly its job: it produces a report that looks very much like a report.

The real risk of an empty record is not that it lacks data, but that it carries enough form to be trusted.

That report rated its own risk as high. High, but as analytical risk, while athletic risk remained undetermined. In information-governance terms it is high — because an empty object is precisely the kind of input most likely to coax people into inventing plausible-sounding content.

One asymmetry kept me thinking. When a report cannot name a single doping signal, a skimming reader takes away “no doping problem”. The correct reading is “no data at all”. Those two sentences are worlds apart, and nothing in the output format distinguishes them.

At the same time, the two dimensions that lost the most are the two few people watch. The rules and anti-doping dimension lost its ability to cross-check ten-year sample storage and biological-passport anomalies. The industry-economics dimension lost the entire transmission line from shoe technology down to the mass running market. For a story about women's sport, that is usually the only section capable of saying something new.

There are playing careers that never appear on a transfer list, only in the silence between two touchlines.

Based on my experience covering meets indoors and outdoors, I have grown used to a smaller version of this disease. A mark goes into the headline, while the wind reading is pushed to the final line. A carbon-plated shoe is praised as a coaching breakthrough, while the sole thickness goes unmentioned. At the data layer, today's failure is the same disease shifted up one notch: we accept a fully populated record instead of checking whether it has anything inside.

The reflex is to blame the extraction model. The schema is designed to reward completeness, not content. The newsroom is designed to reward publication, not stopping. Between the two there is no gate at all. A two-field rule — title non-null and at least one information point — would almost certainly eliminate this entire class of failure. It is cheap. It simply never ships, because what it catches is invisible by design.

More telling still: that report behaved correctly. It did not invent an athlete's name, did not conjure a sprint, did not pin a medal on anyone. It chose to say it did not know. In a trade where a wrong headline can outlive a season, being willing to say you do not know is the most mature response this pipeline managed to produce.

There are matches whose score is never recorded, because people chose to forget. And there are records that are never questioned, because every field of them has been filled in.

I write to recover the names that history edited out of the line-up. But a name can only be recovered if it was written down somewhere first. If a report about women's sport can travel the whole pipeline without anyone noticing it is hollow, then who is actually reading the data that remains?