Nine Filters Behind a Single Athletics Result

Nine Filters Behind a Single Athletics Result

On 1 August 2026, at the National Stadium in Tokyo, a thirty-one-year-old Chinese sprinter walked into lane four for the 100 metres semi-final. The board showed 9.83 seconds. One line beneath, in a smaller font, was the wind reading: +0.9 m/s.

Most spectators read the top line. I read the bottom one first.

The bottom line says the mark is record-legal. It also says the conditions were not generous. For anyone who works with performance data, the word "legal" carries as much weight as the number itself. An athletics result is not a bullet fired straight into a record book. It is the final output of a long chain, and that chain has at least nine filters a reader must pass through before trusting what they just saw.

This is a map of those nine filters.

When the data shattered

I started keeping athletics and football records at seventeen. In the summer of 2026 I sat in front of a screen for the entire tournament in Russia, building a spreadsheet for every Japan match. Against Belgium in the round of sixteen, Japan held 55 per cent of possession but registered only seven touches inside the opponent's penalty area, against twenty-one for Belgium. I published an analysis on my personal blog arguing that pushing the defensive line high in the closing minutes was a structural error. The piece drew a wave of angry responses.

That night in Russia, I watched the data shatter in front of me. Not because the numbers were wrong. Because I had assumed a spreadsheet was enough to explain a match.

Two years later, the 2026 season was suspended for four months. Unable to attend Yodoko Sakura Stadium to watch Cerezo Osaka, I rebuilt a dataset from old match footage: 1,240 pressing situations from the 2026 season, converted into PPDA, the number of passes a team allows before applying its first defensive action. When the league returned, I predicted Cerezo would decline without home advantage. They finished fourth; my model had them second. I rewrote the model and added a variable I had previously dismissed as noise: crowd effect.

An empty stadium, and the numbers still full of noise.

In the summer of 2026 I spent three weeks on the European Championship and found a team scoring four of its six goals from designed set pieces, against a tournament average near 28 per cent. I placed that beside the 38 set-piece goals of a Bundesliga club in the 2026-21 season, where the coach used running-position and ball-landing data to design training drills. A small football site republished my analysis; it reached roughly fifteen thousand readers.

In early 2026 I covered the winter transfer window for an online football magazine. Screening more than two hundred players moving from the J-League to Europe, I found a correlation coefficient of 0.67 between kilometres run per match and successful adaptation in the Bundesliga. One midfielder, at 11.8 kilometres per match the highest in the league, made my shortlist and later moved to Germany on loan.

Three errors, three corrections. Since then, whenever I look at a result, I do not allow myself to stop at the first number.

Filter one: the physical conditions of a mark

Athletics is the only sport where the validity of a record depends partly on the weather. World Athletics rules require a headwind or tailwind no stronger than +2.0 m/s for sprint, long jump and triple jump marks. Beyond that threshold, a performance is flagged as wind-assisted and cannot stand as a record. The +0.9 m/s in Tokyo sat comfortably inside the legal zone, yet still saved a sprinter hundredths of a second compared with a headwind.

Altitude is the second variable. At stadiums above 1,000 metres, thinner air reduces drag and sprint records cluster. A responsible analyst always records venue altitude alongside the mark, the way an accountant records the currency.

The track surface is the third. Modern stadiums use synthetic surfaces with different base and top layers. After 2026, some newer tracks allowed athletes to reach higher peak speeds, which makes comparing marks across decades more fragile than people assume.

Shoes are the fourth. Since carbon-plated shoes became standard on the roads, World Athletics has had to regulate sole thickness and the number of rigid plates in road racing shoes. A carbon plate does not run for the athlete, but it changes energy return at every foot strike, and over a marathon that difference is multiplied more than forty thousand times.

These four physical variables form the first filter. When someone sends me a mark without wind, altitude, track and shoe context, I treat it as incomplete data. Incomplete data cannot be compared.

Filter two: the athlete's personal curve

A mark only means something when placed on the athlete's own performance curve.

Three indicators must be read together: personal best, season best, and year-on-year rate of improvement. The third is the least discussed and the most useful. For a mature sprinter, annual improvement typically sits within a few hundredths of a second, and the curve flattens after twenty-seven. When an athlete suddenly improves at three times the historical annual rate, I mark that row in red. Marking is not concluding. Marking means more data is required.

Age and event form another pair. The peak window for sprints usually falls between twenty-four and twenty-nine. Middle and long distances mature later, often twenty-six to thirty-one. Throws peak latest, sometimes at thirty-three. A standout result at twenty-two in the shot put means something very different from the same mark at thirty: the young athlete may still improve, while the veteran is fighting the curve.

Injury history is the third element. An athlete who withdraws from two consecutive seasons is a high-risk flag in my tracking sheet, regardless of how good the latest mark looks. Recovery curves after tendon, Achilles or hamstring injuries are not linear, and a strong result immediately after a comeback is often the data of a body borrowing against the future.

Filter three: qualification mechanics and competition structure

At elite level, athletics runs on two parallel roads to a major championship. The first is the entry standard published by World Athletics for each season. The second is world ranking points, based on placings and the value of the meets contested.

These roads produce two different strategies. An athlete chasing the standard concentrates resources on one or two races with the best weather, track and pacemaking. An athlete chasing ranking points races more often, accepts slower times and trades peak performance for consistency.

One structural detail is routinely missed: the limit of three athletes per country per event at major championships. This creates what I call the fourth-place effect. In countries with real depth, an athlete finishing fourth at national selection can hold a mark equal to or better than a medallist from another nation, yet never appear on a start list.

The United States selection model is the clearest example of this structural risk. In some Olympic cycles, team places are decided almost entirely inside a single trials meet. A reigning world champion can miss the Games by failing across those three days. No probability model captures this fully, because it converts a stable ability into a binary variable.

For Southeast Asian nations the problem runs the other way. Quota is not the issue. The gap between the entry standard and current ability is, compounded by the cost of travelling abroad often enough to find qualifying conditions. A Vietnamese middle-distance athlete chasing a standard must race overseas several times a year, and every trip carries expense, jet lag and injury exposure.

Filter four: event landscape and the national power map

Every event has its own terrain, and that terrain shifts on a four-year cycle.

In men's sprints, Jamaica and the United States have been central for decades, but the power structure is not fixed. Since 2026, European and Asian athletes have appeared more frequently in major finals. When an event shifts from single domination to an open field, a recognisable signal appears: the gap between the fastest and the eighth-fastest seasonal marks compresses. When that happens, racing strategy moves from chasing records to chasing positions, and the value of starting technique rises.

In distance events, Kenya and Ethiopia remain the two large centres, but the internal structure has professionalised around training groups. The group, not the nationality, has become the unit of analysis. In marathon running, an athlete inside a group with organised nutrition and pacing support carries a higher record probability than an equally talented athlete running alone.

In throws, Europe and the United States retain depth advantages. A country can produce one champion without producing a foundation, and this is the landscape most easily misread: people believe they are watching a sport rise, when they are watching an individual exception.

China holds a specific position in race walking and women's throws, the result of a calculated event-selection strategy that concentrates resources where the training cycle is longer and technique matters more than raw speed. In the opposite direction, the distance-event gap remains unresolved.

For Southeast Asia, the picture must be read at regional level before continental level. A SEA Games gold in the 3,000 metres steeplechase and a place in an Asian final are two different tiers, and merging them into one narrative about regional sport rising damages both the analysis and the training plan.

Filter five: competition rules

Athletics is a sport where the rulebook can erase a performance in an instant.

The current false start rule removes the concept of a second chance entirely. One movement before the gun means disqualification, with no exception. Before 2026, major meets still allowed a restart after a first offence, and that change reshaped the psychology of the sprint events.

In relays, the exchange zone is tightly defined, and any contact or exchange outside the zone results in team disqualification. With four athletes, the probability of a technical error compounds rather than adds. This is why the four fastest individuals are not always the medalling team.

In jumps and throws, each attempt carries rules on preparation time, foot placement, exit direction and landing position. Pole vault has its own equipment specification. A vault pole outside specification does not slow an athlete down, but it can void an entire result after the competition ends, and that has happened at junior championships.

For a data analyst, this is the most uncomfortable filter. No probability model predicts a technical error in the ninetieth minute of a three-hour session. The only defence is tracking each athlete's seasonal technical-error rate as an independent variable.

Filter six: the grey zone of anti-doping

The most important thing an analyst can say at this layer is what they do not know.

Absence of doping data does not mean absence of doping risk. In my files, an athlete with no testing-related information is marked "unassessed", never "cleared". Those are entirely different states, and merging them is the most serious error in the profession.

The current system runs on several externally observable mechanisms. The athlete biological passport tracks blood markers over time and flags changes that training or altitude cannot explain. Samples are stored for years, allowing re-analysis with new technology and the stripping of old medals. Whereabouts obligations mean that repeated missed tests become an independent violation, without any positive sample.

In some middle-distance events, regulations on testosterone levels and eligibility conditions have produced legal disputes lasting years, with athletes excluded from their own specialist events. This is the intersection of science, law and human rights, and a data analyst lacks the authority to rule. What I can do is record the sequence, separate evidence from speculation, and refuse to turn the silence of data into a verdict.

Filter seven: the training system behind the athlete

A result on the track belongs to a system, not only to an individual.

At least four training models coexist at elite level: the centralised national team, where athletes live and train in a state-funded centre; the professional club, where athletes sign contracts and race to a team calendar; the school model, prominent in the United States and Jamaica, where high schools and universities operate as genuine academies; and the private group, where an individual coach assembles a small squad, often in one event, sharing the cost of altitude camps.

The private group is the most notable development of the past decade, because it lets an athlete from a country without an athletics tradition access world-class training infrastructure. It also makes tracking harder: when a group spends three months a year at altitude, all competitive data is interrupted on a schedule no journalist is told about.

Support infrastructure also determines data quality. A squad with its own sports laboratory knows each athlete's lactate threshold precisely and adjusts daily volume accordingly. A squad without one relies on feel, and feel carries the largest standard deviation of any variable in the model.

Filter eight: the risk matrix

Before every major championship I build a four-part risk matrix: competitive, regulatory, physical and media risk.

Competitive risk is the chance of meeting a stylistically incompatible opponent. In the 1,500 metres, a strong finisher suffers badly when the early pace is slow and the race explodes only over the final four hundred metres. An endurance specialist suffers when the pace is pushed hard from the start.

Regulatory risk belongs to the fifth filter, and it is unevenly distributed: relays and pole vault carry far higher technical-error probability than straight sprints.

Physical risk was described in the second filter, but one point deserves emphasis. In events requiring three rounds across four days, recovery capacity is a separate variable, not fully correlated with personal best. Some athletes hold excellent personal bests and never survive a semi-final.

Media risk is the category analysts most often ignore, though it directly affects results. An athlete who becomes the face of a national campaign spends dozens of hours a month on obligations away from the track. That load appears in no statistical table, but it appears on the track at the end of the season.

Filter nine: hidden data and source reliability

The final filter is the hardest, and it determines the value of the previous eight.

When I read a number, the first thing I do is trace its source. What device measured it, how was it calibrated, who published it, when, and has any independent party confirmed it. For official competition marks, this chain is usually clear. For unofficial metrics such as training mileage or GPS data, the chain almost always breaks somewhere.

There is a category I call unverified training data: information passed along from a training camp, a coaching group or the athlete themselves, without any independent measurement system. Such data has reference value and no comparative value. Placing it beside official competition marks is a methodological error, and it happens more often than anyone wants to admit.

Data does not create stories; it strips the stories other people built. When a number is published alongside a glossy narrative, my job is to test whether the number survives being separated from the narrative.

The counter-intuitive angle: correlation is not causation

For years I have received questions with the same structure. Does running more kilometres make an athlete more successful in Europe? Have carbon-plated shoes inflated records? Are new track surfaces the cause of the fast times seen in recent seasons?

All three questions share a flaw: they convert correlation into causation.

In the first case, the 0.67 correlation I found between J-League running distance and Bundesliga adaptation does not say that running more produces success. It says that high-mileage players are usually the ones European clubs selected for other reasons, and running distance is a companion marker. A club that buys on that number is buying the marker instead of the ability.

In the second case, a rigid plate in a shoe improving running economy is a measurable physical mechanism. But attributing an entire wave of records to shoes is a reduction that ignores three other variables: nutrition, physiologically individualised training loads, and professionally organised pacemaking in major marathons. Four causes running in the same direction does not mean all four explain equally.

In the third case, the error is hardest to see. A record set in a particular stadium does not prove that stadium is fast. It proves the stadium is faster than average, and that good organisers know how to assemble the right people at the right moment. A race with three athletes chasing a record will produce a record, whatever the surface is made of.

My defence against this trap is to write the counter-argument first. If I intend to claim a variable is a cause, I must be able to write a paragraph explaining why it might merely be a passenger. When that paragraph cannot be written, I know I am manufacturing counter-intuition for shock value rather than analysing.

Every probability hides a shock. I only make sure it does not repeat.

Where this stops

Back to the board in Tokyo. The top line is 9.83 seconds. The bottom line is +0.9 m/s. The next four filters sit in the venue altitude, the track structure, the shoe and the personal curve of a thirty-one-year-old at the final year of his peak window.

Walking through nine filters does not make the number smaller. It makes the number heavier. And as the regular season moves into its next rounds, the question I carry is not who will run fastest. It is which of these nine filters will be the one skipped next time.