The Empty Column in Athletics Results and the Audit Nobody Runs

The results sheet printed all twelve columns. The athlete name column was blank. The distance column was blank. The time column was blank. The wind reading column was blank. In the status line at the bottom of the page, the system still displayed one word: valid.

That sheet sat on the table during a technical meeting in Nagoya on November 14, 2026, when a competition-data supplier presented its new pipeline for the national-level athletics results system. The chief engineer said the system had met the standard, because every data field existed in the right place. I asked the inverse question: if every field exists and no field contains anything, what exactly was tested? He replied that content verification sits in a later stage. That later stage did not appear in any schedule I was shown.

Three months on, an export file from a regional athletics meeting reached me: 412 rows, structurally complete, with 61 rows carrying no value in the wind column. Not one row was flagged as an error. The null rate was 14.8 percent, and that rate appeared nowhere in the acceptance report.

Athletics runs the strictest data system in the entire sports industry, and it is the only sport where a performance can be stripped of its status because of a single secondary column. A 100m record exists only when the tailwind does not exceed +2.0 m/s. A road shoe is legal only if the sole does not exceed 40mm; track spikes have been capped at 25mm since the 2026 season, subject to a market-availability condition tied to a fixed date. Athlete samples must be stored for ten years to allow retrospective analysis. These conditions are not paperwork. They are the line between a record and a training session with spectators.

In 2026, in Vienna, Eliud Kipchoge ran a marathon under two hours. The 1:59:40 mark was never ratified as a world record, and that was the correct conclusion: the course was staged for a single objective, pacers rotated in formation, the conditions were not open to all challengers. The official record at the time remained the 2:01:39 Kipchoge set in Berlin in 2026. Same athlete, same distance, more than two minutes apart — and the entire gap lives in footnotes almost nobody reads.

My question starts there. A sports data system runs in three layers: collection — timing devices, officials, federations submitting results; extraction — turning text and tables into a structured data object; publication — pushing into newsrooms, rankings, and commercial data feeds. A failure in layer one usually makes noise. A failure in layer two stays silent, and that is the far more dangerous class.

The extraction layer can return an object with the correct structure, every field present, and nothing inside it. In technical documentation this phenomenon has a name: a silent pipeline failure. The output object conforms to the schema, clears every automated validation step, and travels downstream carrying no content at all. It resembles a passport with all its pages intact and no photograph: heavy in the hand, unidentifiable when opened.

In the 412-row file, the fault lay elsewhere and was more troubling: the data was real, only the wind column was missing. A blank column does not break a table. It only makes every mark in that table unauditable. For those 61 rows, nobody can say which performance was genuine and which was the product of a gust that arrived at the right moment.

The strangest thing is never the error itself, but the way people try to explain it. When I asked about the 61 blank rows, the answer was that the wind sensor at that station had failed to connect to the central system. A reasonable answer. But if the sensor lost connection, the correct default is to mark those rows as undetermined, not to leave them blank and publish as usual. The difference between those two choices is not technical. It sits with whoever is accountable for an empty cell.

Here I have to state my method plainly, because every conclusion that follows rests on it. When I meet a dataset with a high null rate, I do not conclude fraud. I build a simple model: let p be the probability a record suffers a silent failure at the extraction layer, q the probability a downstream operator fills the empty cell with a plausible but unverified value, and k the number of independent copying layers. With p = 0.15, q = 0.4 and k = 3, the probability that at least one fabricated value reaches publication is 1 − (1 − 0.06)³, roughly 16.9 percent. That number proves nothing. It says only that with three copying layers, a small fault at the top can become a printed fact at the bottom, and the probability of that is not small.

I often ask: where did this money come from, and what did it do on the way? Applied to competition data, the equivalent question is: who pays for completeness? A data supplier sells coverage, not accuracy. A feed with 100 percent of fields filled sells to more customers than a feed with 85 percent filled and 15 percent marked undetermined. Contracts do not say correct; contracts say complete. And once a contract says complete, an empty cell stops being a scientific problem and becomes a commercial one.

Downstream of that chain sit three customer groups. The first is the press, where a results story needs enough figures before the page goes live. The second is rankings and athlete profiles, where a blank field leaves a hole in a competitive biography. The third is the commercial data feeds serving analytics and betting markets, where a blank field can suspend a line. All three share one pressure: fill it and move on. None of them has an incentive to write undetermined.

And here is what I consider the most serious consequence, far beyond one missing wind column. Across the whole control architecture of athletics, the absence of a signal is always read as the absence of risk. An athlete with no adverse test result is treated as clean. A file with no recorded whereabouts violation is treated as compliant. A mark with no equipment annotation is treated as a mark of the body. No document states that those readings hold only when the data is complete. Samples are stored for ten years, biological passports are tracked as a series, medals can be revoked years later — that entire retrospective machinery is worth something only if the original record exists. An empty cell leaves nothing to analyse retrospectively.

In 2026 I stopped following matches and started following vials. When I cross-checked a club's compressed fixture list against the testing log of the anti-doping authority, the first problem I hit was not a positive sample. The problem was eleven cases with no collection date, and in the summary report those cases were folded into the category of no abnormality detected. An empty cell had been read as a negative result. That is why I wrote the report on statistical analysis rather than testimony: the probability that six players sourced the same protein supplement from the same sports clinic was 0.7 percent, and that number forced the disciplinary committee to open a file.

Safety is not about never being caught; it is about never leaving a trace. In the data world, an empty cell is the safest object in the entire system: it is never audited, because there is nothing to audit. It is never questioned, because there is nothing to question. And it still counts toward coverage.

I do not want this piece read as an accusation against any individual. In the 412-row file I found no person paid to delete a wind column. What I found was a process in which no step is accountable for an empty cell, a contract that measures only completeness, and an acceptance report with no line about error rates. Together those three produce a system capable of generating results sheets that look immaculate and cannot be verified.

There is one argument on the other side that I consider correct, and I have to put it on the record before concluding. An honest blank is better than a wrong entry. An editor who publishes nothing harms nobody. Automated extraction has cut transcription error to a level humans cannot reach — a decade ago, misspelled athlete names and transposed result columns were routine in every newsroom. If every meeting were required to supply full wind, equipment, track and weather data, youth competitions in districts with one timing device would vanish from the data map, and those are precisely the athletes who most need to be recorded. I have heard this argument from data people I respect, and I have no technical rebuttal to it.

But that argument leads somewhere different from where the industry has taken it. If missing data is acceptable, then publishing the fact of missing data must also be acceptable. The problem is not the existence of an empty cell. The problem is an empty cell with no label. A results sheet reading wind: undetermined is an honest sheet and remains usable. A results sheet with a blank wind column published as usual has transferred the risk to the reader.

Athletics has learned to audit performance at the athlete layer: wind, altitude, equipment, stored samples. It has not learned to audit the pipeline that produces those records. A two-field gate — a title must exist, and at least one quantitative data point must exist — would remove most of this silent failure class at near-zero cost. A simpler metric still is the null rate, published openly per meeting and per month, the way injury counts are published.

All I do is connect the dots — and count how many people deliberately draw them wrong. But before anyone can be counted, there have to be enough dots on the page. If a sports data system could say plainly that it does not know, would this industry still need numbers filled in to occupy the space?