When the Wind Column Is Empty: The Architecture of Verification in Athletics Data

5:40 a.m. in Osaka. The first Midosuji subway train is still twenty minutes away, and in a small apartment in Nishi ward I open a spreadsheet with 148 rows. It is a men's 100 metres result list from a meet in the Kansai region. The time column is full. The athlete-name column is full. The bib-number column is full. The referee's notes column is full. Exactly one column is completely empty: the wind-reading column.

I sit still in front of that blank for a long while. Not because I do not know how to fill it in, but because I know exactly what happens if I do.

In athletics, a 100 metres mark is recognised for record purposes only when the wind reading does not exceed +2.0 metres per second. If the organisers failed to record a wind reading, the fastest mark in that file becomes a number that cannot be published. It still exists on the stopwatch, it is still announced over the stadium speakers, it is still written into the official protocol. But it can never enter a ranking list, can never become an official personal best, can never be used to calculate world-ranking points.

The stadium is empty of spectators, yet the number is still full of noise.

There is a very natural reflex for a sportswriter: when you see a blank, fill it with a guess. The wind that day was probably around +1.2. The air was still, the stadium was enclosed, so it probably did not change much. That is the reflex of a storyteller. I do not do that job. I do the job of going back through the file after the story has already been told, to check whether the story holds up.

On the night of Russia 2026, I watched the data fall apart in front of me. I was seventeen, logging every match of the Japan national team, and I realised something that has shaped my entire working method since: most of the conclusions the public trusts most are built on empty cells.

This article is about athletics. But it approaches athletics the way an auditor approaches a balance sheet: before discussing performance, you discuss the integrity of the inputs.

Context: the densest data architecture in sport

Athletics has one of the oldest and most detailed measurement systems in sport. Everything is quantified: time to the hundredth of a second, distance to the centimetre, wind to a tenth of a metre per second, altitude to the metre, implement weight to the gram, release angle to the degree.

Precisely because the data is so dense, athletics is a sport where false or incomplete data can survive for a long time unnoticed. A single race can contain hundreds of data points, and one blank field in exactly the wrong place can bring down every conclusion drawn from that race.

Based on my experience following competitions, from high-school meets in Osaka prefecture to Diamond League rounds broadcast on television, I keep seeing the same pattern: spectators remember the mark; analysts remember the conditions. The mark is what gets printed on the scoreboard. The conditions are what decide whether that scoreboard means anything.

The athletics season runs on a fairly stable rhythm. The indoor phase lasts from January to March and usually serves as a technical test for sprints and jumps. The outdoor phase runs from May to September, with Diamond League legs across Europe, Asia and North America. The peak usually lands in August, when the World Championships or a major Games dominate attention. In parallel runs the road-racing branch, with the major marathons in spring and autumn.

This rhythm creates a feature that anyone analysing athletics has to accept: athletes do not race for a mark every time they step on the track. Sometimes they race to test tactics. Sometimes to regain racing feel. Sometimes simply to fulfil an obligation to a sponsor or a national team. That means a slow race does not necessarily reflect poor form, and a fast race does not necessarily reflect a leap in ability.

I work in a two-stage process. The first stage extracts events: what is in the article, who appears, which numbers are cited, which dates are mentioned. Only the second stage analyses in depth across nine dimensions: performance, athlete condition, competition structure and qualification, event landscape and national strength, rules and anti-doping, team and training systems, risk landscape, shift signals, and finally the data requirements still outstanding.

It sounds heavy, but the core principle is simple: the second stage may only conclude on the basis of the information points the first stage has extracted. No information points, no conclusion. And when the first stage returns an empty set, the only correct answer is the one analysts least like to say out loud: insufficient information, cannot assess.

That moment is the subject of this article. I collect mistakes, classify them, and from that I learn where a team is heading. Athletics is the same: I collect empty cells, classify them, and from that I learn which dataset is still usable and which one has to be thrown out.

The wind column: the cheapest and most powerful field

Start with the blank that woke me at 5:40 a.m.

The wind reading in athletics is not a supporting detail. It is the legal condition of the mark. An athlete who runs 100 metres in 9.85 seconds with a wind of +3.4 metres per second still has a competition result, is still ranked in that competition, but that mark can never be a record. Conversely, the same athlete running 10.12 into a headwind of -2.8 metres per second still has a record-eligible mark, even though it is nearly three tenths slower.

This is the point spectators overlook, and the point news reports tend to overlook with them. The news wants the striking number. The wind reading makes the striking number complicated. So the wind reading gets pushed to the end of the piece, or dropped entirely.

I once built a comparison table for more than four hundred men's 100 metres and 200 metres results across Asian competitions over three consecutive seasons, purely to answer a very narrow question: when the wind reading exceeds +2.0, how much faster does a mark get on average?

The result forced me to rewrite the opening of my whole report. The average difference in my sample sat between one and two per cent of the time, depending on the athlete group. It sounds small, but in an event where the gap between gold and bronze is sometimes four hundredths of a second, one to two per cent is the entire difference between a final and a flight home after the heats.

From that, I adopted a personal rule: every conclusion about speed must come with a wind reading, or must state explicitly that no wind reading exists. No exceptions.

But the wind column is only the first layer. The verification architecture of athletics has more layers than that.

The second layer is altitude. Thinner air above a thousand metres reduces drag and creates a measurable advantage in sprints and long jump. Some venues are famous for this advantage, and experienced analysts always attach venue altitude when assessing an unusually fast mark.

The third layer is equipment. Carbon-plated shoes deliver a proven energy-saving benefit, especially in distance events. A marathon mark set on a flat course, in low temperatures, with a professional pacing group and high-technology shoes cannot be compared directly with a championship marathon on a hilly, twisting course with no pacemakers.

The fourth layer is the track surface. Newer synthetic tracks differ in their rebound, and some stadiums are deliberately designed to be fast. This is an advantage held by the organiser rather than by the athlete's legs, and it has to be deducted when comparing across venues.

The fifth layer is split data. In sprints, reaction time is a decisive field. In middle and long distance, splits per lap or per four hundred metres reveal how speed was distributed. An evenly fast mark does not mean the same thing as a mark made fast by a closing surge, and two races with identical finishing times can tell two completely different stories about conditioning.

When split data is missing, the analyst is forced to lower confidence in every judgement about the underlying physical base. This is a silent error, because nobody sees it on the results sheet.

The personal-best curve: the test nobody wants to run

Moving from conditions to the athlete himself.

The strongest tool in an athletics analyst's hands is not the season ranking. It is the year-by-year series of personal bests.

A single mark is a point. A series of marks is a line. And in most cases the line is what tells the truth. A career-trajectory table separates three kinds of athletes who currently share the same mark: the one improving steadily, the one who has hit a ceiling and is moving sideways, and the one who has just produced an anomalous jump.

In my framework there is a warning threshold I apply fairly strictly: if the improvement in a personal best within a single year exceeds roughly three times the athlete's own historical annual gain in the preceding period, the file deserves closer review. That threshold is not a verdict. It is a request for explanation.

There are plenty of entirely legitimate reasons for a jump: a new coach, a move to a better-resourced training group, recovery from a long-term injury, new equipment, a shift from combined events to a specialist event, or simply a natural phase of physical development. The problem is this: without the series, the analyst cannot separate the legitimate jump from the jump that needs investigation. And if you cannot separate them, you are not allowed to conclude in either direction.

This is where the concept of event-specific peak age becomes useful.

Sprints usually peak between twenty-four and twenty-nine. Middle and long distance usually peak later, between twenty-six and thirty-one, because pacing experience and tactical understanding offset the decline in raw speed. Throws usually peak latest, between twenty-eight and thirty-three, because muscle mass and rotational technique take years to accumulate.

Knowing where an athlete sits on that curve completely changes how you read their mark. A flat result at twenty-three is a good or neutral signal. The same flat result at thirty-one is a signal of decline. A jump at twenty is ordinary. The same jump at thirty-two needs explaining.

Alongside the trajectory sits the injury record and the competition calendar. The highest-risk indicator in my framework is withdrawal from competition in two consecutive seasons. One withdrawal can be a minor problem. Two in a row usually signals an unresolved injury, an unnamed psychological barrier, or a conflict with the coaching staff.

Based on my experience following competitions, I notice that the media generally treats a withdrawal as a short line at the bottom of the piece. For me it is often the most important line in the piece, because it is the only data point that tells you what state the athlete's body or head is in.

Qualification: two doors, one narrow corridor

Athletics competition structure has a feature that sets it apart from most team sports: an athlete can earn the right to compete without going through any national selection at all.

The current mechanism runs on two parallel channels. The first is achieving the qualifying standard within a defined window. The second is accumulating enough points through the world ranking, built from results, placings and the level of the competition.

These two channels produce two completely different types of season strategy. Those who hit the standard early can control their schedule entirely, even racing sparingly to preserve energy for the main event. Those chasing ranking points must race densely, travel constantly, and often accept highly competitive meets in mid-season.

The physical cost of the two paths is not the same. An athlete who races twelve times in a season to accumulate points enters the main event with an accumulated-fatigue base quite different from one who raced five times.

In my framework this is the most frequently ignored variable when assessing form. People compare personal bests while forgetting that one athlete has run twelve races and the other five.

At national level, one selection model stands out for its severity: one race decides everything. Under this model, a place at a major championship is awarded on the outcome of a single trials meet, regardless of prior marks or status. The consequence is that a reigning world champion can be absent from the biggest stage if they have one bad day.

The model has clear merits: it is transparent, easy to communicate, and removes all argument about criteria. It also has a clear flaw: it treats one afternoon as the measure of four years.

At international level, one mechanical constraint creates particular pressure: a maximum of three athletes per country per event at the World Championships. This constraint turns countries with real depth into environments where internal competition is harsher than the final itself. An athlete who finishes fourth at a strong nation's trials may hold a better mark than an entire delegation from another country, and still stay home.

This is information the results sheet never shows. It only appears when the analyst reconstructs a country's whole qualification structure.

I consider this one of the biggest blind spots in sports media generally, and in Asian athletics media specifically. We celebrate an athlete clearing the standard, but we rarely spend space explaining that in another country an equivalent mark would not even earn a reserve place.

Event landscape: four shapes and how to recognise them

An athletics event, at any moment, usually falls into one of four landscape shapes. Recognising the shape matters more than remembering the champion's name.

The first is single-ruler dominance. One athlete holds the lead by a wide and stable margin across several seasons. In this shape the analytical question is not who wins, but whether the gap is narrowing or widening, and whether the chasing group is getting younger.

The second is a two-horse race. Two athletes share almost all victories across several seasons. In this shape head-to-head data becomes more important than absolute marks, because both usually compete in identical conditions.

The third is a wide-open melee. Nobody holds the lead across two consecutive seasons. Here consistency matters more than the best mark, and major championships tend to produce a high rate of surprises.

The fourth is generational transition. A group of peak athletes passes its career peak within a short window, creating a gap the younger cohort has not yet filled. Here the value of a major championship lies not in absolute marks but in identifying who stabilises first.

Recognising these shapes is fairly mechanical: take the season's top ten marks, sort by nationality, sort by year of birth, and look at the structure. If one country holds four of the top ten and the group's average age is under twenty-five, the landscape is tilting towards dominance with a foundation. If the top ten spans eight countries with an average age above twenty-eight, the landscape is in transition.

At national level, the world power map has a fairly clear structure. Sprints are usually dominated by several Caribbean nations alongside the United States, thanks to school sports systems and a sprint culture built from secondary school upward. Distance events are heavily shaped by East African nations, where natural altitude training combines with a group-training tradition to produce a continuous stream of athletes. American jumps and throws have unusual depth thanks to the collegiate system. European throws maintain a stable position through club structures and national training centres.

For Asia, the picture has two notable bright spots. The first is race walking, where several Asian nations hold world-leading positions through centralised selection and high training volume. The second is women's throws, where cycles of success have lasted many years.

In the men's 100 metres, the 9.83 seconds set by Su Bingtian at the Tokyo 2026 Olympic Games is a data point of particular analytical value. It proved that the speed limit for Asian athletes is not biological but lies in the quality of the training system, the quality of competition, and the density of international exposure. At the same time it raises a question the data has not answered: why, after that milestone, the number of Asian athletes running under ten seconds did not rise correspondingly.

That is the kind of question I like. It has no ready answer, and it demands system-level data rather than individual-level data.

Rules and anti-doping: where blanks are most dangerous

This is the section I have to write most carefully, because it is the easiest to misread.

My anti-doping analytical framework runs on a specific set of variables. The athlete biological profile, with blood and urine markers tracked over time, exposes anomalous fluctuations that training or altitude cannot explain. The athlete's whereabouts schedule, with missed-filing violations, is another indicator. Long-term sample storage allows old samples to be retested with new technology, leading to medal reallocation years later. Association with previously sanctioned coaches or doctors is a variable to watch. And finally, anomalous performance jumps.

These five variables work like a net. None is sufficient on its own. But when several deviate from the normal range at the same time, the file moves to a higher level of review.

The problem is this: if the input contains no doping-related fact, no athlete name, and no suspect indicator, then all five variables have no input value. The net is never lowered into the water, and therefore catches nothing.

This is the point I want to stress in bold, because it is the core of this entire article: an empty result in doping analysis does not mean a clean file. It means an unassessed file.

In analytical practice there is a very common systemic error: treating the absence of evidence as evidence of absence. That error is dangerous in every field, and most dangerous in sport, where an analyst's conclusion can affect a person's reputation and career.

The other branch of this section is technical competition law. Athletics has a dense and strict technical rulebook, and most controversial incidents in a season originate there.

Current false-start rules are close to zero-tolerance: a single start before the gun leads to immediate disqualification. Lane infringement, even by a single step, is enough to void a result. In relays, exchanging the baton outside the designated zone is a common technical fault that frequently decides the overall outcome. In jumps and throws, rules on attempts and technical faults are detailed to a degree spectators struggle to follow without familiarity.

In pole vault, equipment specifications and pole parameters form another branch, where technical disputes can surface late and carry large consequences.

The common thread in the technical branch: every dispute leaves a trace in the data. A voided result appears in the protocol. A relay-zone fault appears in the officials' report. The analyst does not need to guess, only to read the right document.

And when the document does not exist, the only correct action is to write into the file: not assessable.

Training systems: four models and the trap of copying

After performance, individual condition, competition structure and technical law, the next question is always: what produced this athlete?

World athletics contains four athlete-development models with fundamentally different logic.

The first is the centralised state or national training centre model, where athletes are developed within a system with a clear pathway, stable resources and high training-volume demands. Its strength is the ability to produce a continuous stream of athletes in technically complex events. Its weakness is dependence on a small number of decisive experts, and limited internal competition at grassroots level.

The second is the collegiate model, where athletes develop alongside study within a dense internal competition system. Its greatest strength is competitive density: a young athlete can race twenty times in a season against rivals of the same age. Its weakness is limited specialised training volume and recovery time.

The third is the altitude model, where athletes live and train long-term in high mountain regions, exploiting natural physiological advantage and a group-training tradition. Its strength is producing distance athletes with outstanding aerobic bases. Its weakness is dependence on geography, which money cannot replicate.

The fourth is the school-based model tied to local competition culture, where junior school-level meets act as the main talent furnace. Its strength is producing large numbers of sprinters with technique honed early. Its weakness is a low conversion rate from junior talent to elite athlete.

What matters for the analyst is never copying a model across cultural contexts while ignoring the base.

I once made exactly this mistake. In 2026, when the pandemic suspended the season for four months, I sat at home and built a self-made dataset from old match footage. I logged more than one thousand two hundred pressing situations from a club in Osaka to analyse the passes allowed before pressing began.

When the league returned, I predicted that club would drop in form because the absence of home crowds would hurt the pressing game I believed was their foundation. The result: they finished fourth, below my predicted second.

I was wrong. I logged the error and added a new variable to the model: crowd influence. Not because I was certain that variable was right, but because the data showed I was missing a piece.

The stadium is empty of spectators, yet the number is still full of noise. That noise is pressure, expectation, refereeing error — variables analysts tend to push out of the spreadsheet because they are hard to measure.

In athletics the same lesson applies directly. A performance-prediction model built on European data will fail systematically when applied to an Asian athlete with a different schedule, different climate conditions and a different relationship with the coaching staff. The error is not in the formula. It is in the untested cultural base beneath it.

At team level, one risk category is rarely mentioned but carries huge weight: key-personnel risk. A head coach leaving can collapse a four-year cycle. A strength specialist moving to another team can alter an entire group's injury baseline. A leading athlete stepping away can dissolve a whole training group, because athletics training groups are usually built around a few individuals capable of driving intensity.

None of this appears in the results sheet. But it decides the results sheet of the season after next.

The risk map: twelve cells and one dangerous blank

I maintain a risk matrix of several categories, each with specific items, each scored for level, probability, impact and mitigation.

The competitive risk category includes direct rivals improving, schedule changes removing exposure opportunities, adverse weather at the main event, and psychological pressure from entering as a leading contender. This category is usually well assessed, because it shows up clearly in data.

The doping risk category includes exposure from supplements of unknown origin, from therapeutic use without a valid exemption, from whereabouts-filing violations, and from relationships with previously sanctioned personnel. This category is usually under-assessed, because it does not show up on the results sheet until it is far too late.

The injury risk category includes repeated injury history at the same site, sudden increases in training volume, congested racing schedules, and age. It requires medical data outside analysts rarely hold.

The systemic risk category includes coaching changes, budget cuts, sponsorship policy shifts, and volatility in national selection policy. This is usually ignored in purely technical analyses.

And then there is one cell in the matrix I pay special attention to: the cell for unidentified risk caused by missing data.

This cell is usually left empty in professional reports. People list known risks, score them, propose mitigations, and finish. But the empty cell is the most dangerous one, because it represents every risk nobody has thought of yet.

In my practice this cell is always clearly marked as undetermined, with a list of the data fields needed to assess it. I learned this after my wrong prediction in 2026, and I apply it more strictly after every model failure.

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

When the input dataset is empty: what is actually missing

Now I return to the blank from the opening, but at a larger scale.

When an athletics file reaches deep analysis while the extraction stage has returned an empty set, the problem is not in the analysis stage. The problem is that the file lacks its basic particles. In my terminology, those are information points: event name, athlete name, nationality, date of birth, mark with units, wind reading, altitude, competition name, round, placing, personal best, season best, multi-season series, injury record, competition calendar, coach name, training base, and any fact relating to competition law or doping.

The list is long, but every item exists for a reason. No event name, no competitive level. No wind reading, no validity of the mark. No multi-season series, no jump test. No calendar, no accumulated physical cost. No competition and round, no tier. No nationality, no reconstructed qualification structure. No coach name, no systemic risk assessment.

What I want to say here goes beyond one case. In my profession this happens more often than the public thinks. Many widely shared analyses are in fact built on files whose information-point density is so low that the conclusions cannot be verified.

There are two ways to handle it. The first is to fill the blank with experienced inference. This produces smooth, readable writing that is frequently wrong. The second is to state clearly that the file is insufficient for assessment, list what is missing, and stop. This produces dry writing, rarely shared, and honest.

I choose the second. Not because it is easy, but because it is the only way the file can be completed next time.

A statement of "insufficient information" is more useful than a wrong conclusion, because it points precisely at what needs adding. It turns a failure into a task list. A wrong conclusion only creates a wrong belief, and wrong beliefs are very hard to remove.

The counter-intuitive angle: the false cleanliness of empty datasets

Now the part I consider most important.

In modern analytical culture, one habit has settled into the way people think: if a dataset shows no problem, the conclusion is usually written as there being no problem.

That phrasing sounds safe. It is neutral in language. It accuses nobody. It also demands no further work. But logically it is a serious error.

A dataset that shows no problem because it contains no data about that problem is one thing. A dataset that shows no problem after being fully checked is something entirely different. These two statements differ in nature, and in sport that difference relates directly to a person's honour.

I want to push this further by arguing against myself, as I always do before an important conclusion.

Suppose someone argues the opposite. Suppose they say: in athletics, where the data density is so high, where every mark is stored, where every elite athlete sits in a regular testing pool, the absence of any doping fact in an article is actually a positive signal. High testing density creates a natural barrier. It is harder to evade a continuous monitoring system than a fragmented one.

This argument is not weak. It has a basis. And I have to concede that for an athlete who has been in a regular testing pool for years, the likelihood of a long-running violation going undetected is considerably lower than for an athlete outside that pool.

But the argument has three holes.

The first is the timing problem. A dense monitoring system is only effective from the moment it begins monitoring. What happened before does not sit in the data. Long-term sample storage exists precisely for this reason: people know that a clean present does not prove a clean past.

The second is the scope problem. The regular testing pool covers athletes at the very top. Most cases detected in sport originate at lower levels, where testing density is low and where a performance jump may be the only route upward.

The third, and the most important for me as an analyst, is the epistemic problem. An empty dataset carries no information. It carries no positive information and no negative information. It is simply empty. Assigning it a positive meaning is an act of fabrication, exactly as assigning it a negative meaning would be.

In other words, the cleanliness of an empty dataset is a false cleanliness. It does not come from a system having checked and found nothing. It comes from nobody having checked.

This is why I never write "no sign of doping" when the input contains no data. I write "not assessed". Those two words take up space, make the writing less smooth, and leave the reader unsatisfied. But they are correct.

The same logic applies to every other dimension. No injury data, condition not assessed. No competition calendar, physical cost not assessed. No coach name, systemic risk not assessed. No wind reading, mark not assessed.

In each of those sentences, the word "not" does all the work. It separates a conclusion from a blank. It separates an analysis from a summary.

The cultural trap: do not import Japanese yardsticks onto a Vietnamese track

One thing I have to say, because it relates directly to my position: a Vietnamese person doing sports data analysis in Japan.

For years I have watched analyses of Southeast Asian athletics and football written by applying Japanese or European statistical standards and operating procedures directly onto local contexts. The result is usually technically correct and practically meaningless.

The Japanese athletics system has distinctive features. High-school meets play an extremely important role in talent identification. Corporate teams act as a support structure after graduation, creating a relatively seamless pathway. Domestic competition density is high, allowing athletes to accumulate experience without frequent overseas trips. And a group-training culture with high intensity is sustained across generations.

Applying that framework to a Southeast Asian context breaks three assumptions.

The first is competition density. A young athlete in Southeast Asia may race only a few times a year, while a young Japanese athlete races dozens of times. This skews every comparison of progression speed, because progress in athletics comes largely from the number of times you race under pressure.

The second is recovery resources. A system with recovery facilities, nutritionists and a full medical team allows athletes to carry higher training volume. Applying the same training plan to a system lacking those resources leads to injury rather than performance.

The third is decision-making culture. The relationship between athlete, coach, family and federation has a different structure in different countries. A technically optimal recommendation can be entirely neutralised because it conflicts with that structure.

Based on my experience following competitions in both contexts, I have drawn one principle: when transferring an analytical model across cultures, separate three layers. The basic physiological layer transfers fairly universally. The training-method layer needs adjusting to resources. The decision-making culture layer must be rebuilt from scratch for each context.

Ignoring the third layer is the most common error and also the hardest to detect, because it is not in the formula. It is in the base the formula stands on.

I also have to concede something about Southeast Asian fans, including me and the people I grew up with. We have emotional reactions that do not match data logic. We can forgive an athlete whose numbers show decline, because we know their circumstances. We can turn away from an athlete whose every indicator looks good, because something feels untrue.

Those reactions are data, not noise. An analyst working in Southeast Asia who ignores them will build models that are technically accurate and useless for prediction. I learned this when I got a club's Osaka finish wrong in 2026, and I keep relearning it every season.

The checklist for a standard-compliant athletics file

Before closing, I want to leave the list of what an athletics file needs to qualify for deep analysis. This is the list I use daily, and the list I send back to colleagues whenever I receive a file with low information density.

On event and performance: the specific discipline, the technical element being analysed such as the block start, pacing distribution, javelin release angle or approach run; the exact mark with units; the wind reading for track and jump events; venue altitude; competition name and round; placing; and relevant national, continental and world records.

On the athlete: name, date of birth, nationality; multi-season personal-best series; current season best; injury and withdrawal history; season calendar; training base; and coaching relationship.

On competition structure: competition name and tier; the athlete's qualification status; national selection rules; entry list; and pressure from per-country entry limits.

On event landscape: the season's top ten marks; the world lead; nationalities in the leading group; and the age structure of that group.

On law and anti-doping: any fact relating to testing, sanctions or biological-profile anomalies; the specific rule area at issue; and the athlete's status in the testing system.

On team and training: coach names; training group; training base; programme type; support-staff configuration; and any recent personnel change.

This list is not long compared with the workload of a major championship. But it is far longer than the level of preparation behind most analyses I have read. And the distance between those two levels is why wrong conclusions survive so long in sport.

Closing: what I will check next season

I will not stop at a list. I will turn it into action.

Next season, before every major round, I will run a quick screen on the three fields I consider most decisive: wind reading, multi-season series, and qualification status. Those three answer three different questions. The wind reading tells me whether the mark is record-eligible. The series tells me whether the athlete is rising, flat, or just produced a jump that needs explaining. Qualification status tells me the physical cost the athlete has paid across the season.

Those three fields are not enough to conclude about an athlete. But they are enough to filter out the files that need close reading and the files that do not yet have enough data to read.

And I will keep writing two words where people usually write a conclusion. Not assessed. That is not avoidance. It is the output of a calculation.

Data does not create the story; it strips away other people's stories. And when the data is empty, the story does not disappear. It only moves into a state of awaiting verification, and it stays there until someone is willing to read it again from the beginning.

The biggest lesson of my nine years of observation, after one wrong prediction about a club in Osaka, after countless times a results table collapsed in front of me, is something simple: the discipline of an analyst lies not in how well they analyse, but in whether they dare to stand still in front of a blank.

The wind column in that 148-row spreadsheet is still empty. I still have not filled it in. And to this day I believe that deciding not to fill it in was the most correct decision I made that week.