A Nine-Dimension Athletics Report Came Back Blank: The Most Expensive Silent Failure in Data Work

At two in the morning I opened the report the analysis system had sent back. Nine sections, in exactly the order I had designed four years earlier: event and performance; athlete condition; competition structure and qualification; national landscape; rules and anti-doping; training systems; risk matrix; media and expectation; industry transmission. Every heading was there. Every table had all its rows. Every content cell carried the same sentence: insufficient information. Not one athlete's name. Not one event. Not one mark, one competition date, one meet. The only surviving label was the word athletics. I sat still for a while. In this trade people fear wrong numbers. Wrong numbers can be fixed, at worst with an afternoon of cross-checking. What chilled me that night was a report that was formally perfect and waiting for somebody to fill it in. And in most analysis rooms, the person filling it in is not the machine. The system I use runs two stages. Stage one extracts information points from the source text: names, events, marks, dates, provenance. Stage two builds a nine-dimension report under a single, non-negotiable constraint: every conclusion must state which stage-one information point it derives from. That constraint exists for a very old reason. Without it, a writer will always find a way to say something that sounds entirely reasonable while knowing nothing at all. That night stage one returned zero information points. Facing an empty input, stage two had two options: stay silent, or invent. It chose silence. I read the file three times and then did what I should have done first — I traced the failure. Three branches, none of them provable from the available data. One: no article was ever passed to the system, a collection failure. Two: an article was passed but the extractor returned a schema-perfect empty object, a parsing failure. Three: the source was not about athletics at all, and the athletics label was a stale default. I lean toward the first two, because a human-written athletics piece almost always contains at least one person or one competition. But leaning is one thing and concluding is another. The athletics label survived alone, and by my own rule a single field cannot be treated as verified. I came to athletics later than most colleagues. In 2026 I joined Runner's World, wrote about running, and sat in the editor's chair long enough to learn one thing: people read athletics with their feelings, and people manage athletics with spreadsheets. Those two almost never meet. The day football stopped, I started counting strides again. In 2026, at thirty-one, I was a data consultant for Hai Phong. Reviewing the academy's indices, I found Vu Minh Hieu, a midfielder of modest build, with an average PPDA of 6.8 — the highest figure in the entire development system. I took the sheet into the meeting and asked coach Truong Viet Hoang to give him a chance. In round 17 of the V.League against Hanoi FC, Minh Hieu won the ball fourteen times, made one assist, and Hai Phong won 2-1. After that I stopped writing from reputation. Over four months in 2026, when every league had stopped, I reviewed 2,300 matches from five V.League seasons and three major European leagues to build a composite pressure index combining PPDA, defensive distance and pressing speed. Teams averaging PPDA below 8.5 collected 1.8 points per game, clearly above the rest. I published the model on my personal blog, and it taught me what that blank report repeated: data is only worth something when someone is willing to count, to record, and to record the times they counted wrong. That nine-dimension framework is not a work of imagination. It is an inventory of what an athletics nation must measure if it wants to be believed. Read in its empty state, it shows the cost of every blank cell more clearly than a filled one ever could. The first cell is performance. A 100-metre result means something only with four fields attached: wind reading, altitude, track surface, shoe type. A tailwind above 2.0 metres per second voids a mark for record purposes. Altitude above 1,000 metres delivers a measurable aerodynamic advantage in sprints. Carbon-plated shoes change foot biomechanics enough to move results by several percent. In Vietnam, youth and club-level meets almost never publish wind readings. I once saw a beautiful 100-metre mark shared across dozens of running groups, thousands of likes, and not one comment asking about the wind. A mark without measurement conditions is a claim, not a data point. The second cell is athlete condition, and here Vietnamese athletics data is thinnest. You need a year-by-year personal best curve, a season's best, and the gap between the two. You need age, because every event family has its own peak window: sprints around 24 to 29, middle and long distance 26 to 31, throws 28 to 33. A 22-year-old running a fast 200 metres may still be far from peak; a 31-year-old holding a season's best within two percent of a personal best is evidence of a good training system, not of youth. I also run a personal filter: if an athlete's personal best jumps by more than three times their own historical annual rate of improvement, the file must be cross-checked immediately — not to accuse, but to rule out measurement errors that should have been ruled out first. Nguyen Thi Oanh is the example I use when teaching how to read data without being swept away by the story. Four gold medals at a single SEA Games means a dense competition load across several days, and in physiological terms that is a variable about recovery capacity across different event types. The same fact, two readings: the emotional one produces a legend, the data one produces a capability profile you can plan four years around. The third cell is qualification. World athletics offers two parallel doors: hitting a qualifying standard, or accumulating world ranking points during the qualification window. Each country is capped at a small number of entries per event, usually three. The SEA Games barely use world standards at all, so an athlete can dominate a region and still be far from an Olympic slot. These two systems generate completely different competition strategies, and choosing the wrong one is the most expensive mistake a federation can make — it does not show up in one race, it shows up four years later. The fourth cell is the competitive landscape, which in any event splits into four tiers: dominant, title-contending, finalist, and qualification fringe. In Southeast Asia I tend to see Thailand strong in sprints, the Philippines strong in jumps and hurdles, Vietnam strong in women's middle distance and race walking. I must be explicit: that is my professional memory, not verified data from this review. In my notebook, sentences like that are written in a different colour of ink. The fifth cell is rules and anti-doping. The biological passport, whereabouts obligations, endocrine regulations, the three-year nationality-change rule, the 40-millimetre sole cap for road shoes. Without a named subject, none of it applies. And here is the point I want to make slowly: "no issue found" and "no issue exists" are two different states, and an empty report can only ever assert the first. A hurried reader sees the anti-doping cell marked insufficient information and reads it as clean. That is the most dangerous misreading in the whole system, because it converts ignorance into certification. The sixth cell is the training system. Periodisation from a winter base through season activation to a championship peak. Altitude camps — in Vietnam that means Da Lat or Sa Pa rather than the familiar Asian athletics destinations. Lactate testing, GPS load monitoring, three-dimensional motion capture. Above all, a coaching change before a major championship is the single biggest roster risk in any national team, and also the earliest thing data can detect, if anyone bothers to record the date a contract was signed. The seventh cell is the risk matrix, seven rows. The first six rows of that night's report were empty. The seventh had content, and the content was me: the risk of using an empty deconstruction to generate invented conclusions. Level: high. Probability: high. Impact: high. It was the only row in the entire report that received a score, and it scored the process that produced it. The eighth cell is media and expectation. Whenever a young athlete runs fast, someone attaches the label of a legend's successor. My filter is simple: was that label built on value-adjusted marks, and is the sample large enough. Two races are not a sample. Five races in the same month are not an independent sample either. Athletics also obeys a rule I have not seen elsewhere: media heat follows a four-year cycle, cooling in the middle years and flaring close to the opening ceremony. Ignore that cycle and every popularity comparison becomes meaningless. The ninth cell is industry transmission. Shoe technology flows from elite competition to the mass running market within two to three years. Major marathons generate an urban economy measurable in hotel nights and flight bookings. New commercial circuits are competing directly with traditional federation structures. In Vietnam, the running boom between 2026 and 2026 is the clearest example of transmission from elite athletics to mass consumption: hundreds of amateur races a year, and an amateur class whose watch data is far more detailed than that of many national team athletes. Reading all nine blank cells again, I noticed something I had missed. That framework was not broken. It was correct. It was waiting for data. And because it was correct, it was dangerous. A wrong framework gets thrown away. A correct, complete, well-tabulated framework with formal authority gets carried into a meeting, and there, under pressure to say something, someone fills the blanks with whatever sounds most plausible. That is why I call this a silent failure. The system returned an object that matched the schema but had no interior. It passed every structural check, because the structure was fine. It failed only at the one step nobody programs: the step where a human actually reads it and understands that there is nothing to read. Now the part where I argue against myself. Most data people would blame the system. I do not. A system that returns an empty object is an honest system. It refused to invent. The party at risk of inventing is the human behind it, who sees an empty frame and feels the deadline pressing on the void. Data is a mirror. Most of the market looks into it and sees only itself. One pair of concepts deserves separating. A report with all nine sections filled does not prove analytical competence. It proves a good template. The correlation between report thickness and correctness of conclusion, in my experience, is close to zero. I have read forty-page analyses, beautifully formatted, containing not one verifiable sentence. I have also read half a page of handwritten notes, one figure per line, accurate enough that I had to call and double-check. The trade taught me something more uncomfortable. Most of what we call a data crisis is really a collection crisis. If the source article reached the system as nothing but a headline, one extra full-text retrieval step would turn the entire report from useless into fully analysable. One pipeline cycle. Nothing tragic. But if nobody notices, an entire batch of reports comes back identically blank, and nothing in their structure reveals it. I have made exactly the mistake I am describing. In 2026 I wrote a piece praising a young athlete on the basis of two competitions in one month. It was widely shared, and I had to rewrite the ending once the full season's data arrived. Since then, every piece I write carries a paragraph stating which data is missing. Not for self-defence. So the reader knows where I am standing on solid ground and where I am standing on an assumption. With that night's report, my footing was obvious: an empty foundation. And I chose to say so rather than fill it with three plausible names and two plausible marks. The signal for the next cycle is concrete and cheap. First, a two-field integrity gate at stage one: the title must be non-null and the information-point count must be at least one, otherwise the run is blocked. That alone eliminates this entire failure class. Second, track the null rate of information points per batch as a data-quality indicator, not as a technical footnote. Third, for Vietnamese athletics, a minimum convention: publish marks together with wind, altitude, surface and shoe type. No grand system required. One extra line on the entry form. People call me a data monk. A monk does not need a cathedral — only the truth. And the truth that night was compact: there was nothing in that file. My job was to say exactly that, then go back to step one and fetch the source.