Fifteen Metres Decide Medals: Re-reading Swimming Data Through a Map of Limits

Fifteen Metres Decide Medals: Re-reading Swimming Data Through a Map of Limits

A column that does not exist

In June 2026 I sat in front of two screens in a Brisbane apartment, watching the heats of a national swimming meet through the official timing feed. My model gave one swimmer a 98.7 percent probability of reaching the 100 metre freestyle final. She swam exactly as the model predicted: distance per stroke 0.04 metres above her personal average, stroke rate steady at 47 cycles per minute, breakout after the opening 15 metres exactly on the mark. Four consecutive swims, no sign of decay.

Then the organisers published the results of a relay she had swum third leg in. Her feet left the block 0.02 seconds before her teammate touched the wall. The whole team was disqualified. My spreadsheet has 42 variable columns. None of them is called “changeover two hundredths early”.

The model was not wrong about her. The model was wrong about the world she was swimming in: a world with officials standing at the corner of the pool, with the cold wet feel of the wall under a searching foot, with a teammate screaming in the next lane, with a decision that has to be made in 1.2 seconds and no coach can make it for you.

Kazan is the day I learned that a 99 percent probability can still die on the betting table. After the day Germany collapsed in Kazan, I set myself one rule: every time I present a number, I must also state what that number cannot measure. The rule followed me into swimming, and it turned the first 15 metres of every lane into a private obsession.

Swimming: the most measured and most misread sport

Swimming has something football will never have. The result is a single number, produced by a machine, with no interpretive intermediary. There is no argument about whether a touch counted as an assist. There is no case where the same shot is logged as 0.08 or 0.12 expected goals depending on the provider. Time is time. At international level, electronic timing is wired directly into World Aquatics scoring, and results reach the database within minutes.

Because the data is clean, amateur analysts assume swimming is the easiest sport to predict on the Olympic programme. I think the opposite is true. Clean data is not complete data. A result sheet gives you time. It does not give you how a swimmer handled the first 15 metres underwater, how they took the turn, how they distributed breathing across the race. The gap between the number that exists and the mechanism that produced it is where betting money is mispriced, and it is where I earn a living.

This year is a genuine annual season, with no Olympic Games. That changes how the whole system should be read. In an Olympic season every meet is a springboard and every athlete peaks around July. In an annual season we see a longer, flatter and therefore more honest arc: national championships from February to April, a block of physical accumulation, international racing in mid-year, then a regional circuit at the end of the year. The physical current and the tactical current of an annual season move more slowly, and because they move slowly they expose things that a medal final hides.

As a reporter covering the Australian market, I sit close to one of the two strongest swimming nations on earth. Australia runs one of the most brutal selection systems anywhere: a place on an international team is decided by times swum inside a single week. That produces a measurable kind of pressure. Based on my experience tracking meets in this system over five years, the gap between morning heats and evening finals can reach 1.8 percent in the 200 metre events but only around 0.9 percent in the 50 metre events. Two events, two psychological mechanisms, one results sheet.

On the other side, I keep in contact with coaches in Vietnam and follow the national championships, where the data layer is far thinner. Results exist. Split data, stroke rate data and breathing distribution data largely do not, or are not published. That asymmetry between two swimming nations is part of the story.

Four metrics I use, and one I refuse to use

My swimming framework has four groups. The first is breakout time after 15 metres, measured from the moment of entry or push-off to the first time the head breaks the surface. The second is stroke rate and distance per stroke, which must always be read together as two axes of one coordinate system. The third is 50 metre split structure and the variance between splits. The fourth is relay exchange data, block reaction time and changeover margin.

The metric I refuse to put into a model will surprise some people: any measure of an athlete’s media or social media popularity. I call it the inspiration index, and I treat it as noise, not signal. An athlete can be celebrated ten times more than a rival and still lose by 0.3 seconds at the third turn. Once a model starts using the inspiration index as an input, it stops describing a lane and starts describing a press conference.

Numbers have no gender, but the people who read them do. I learned that the hard way, and it is why I lock down sourcing for every variable I use.

The Suncorp lesson: data first, emotion after

In 2026, aged 37, I was the only female analyst in the press room at Suncorp Stadium in Brisbane before Brisbane Roar met Melbourne Victory. I published a prediction that Melbourne would win despite trailing 1-0 at half time, based on expected goals of 2.4 against 0.6 and running distance of 112 kilometres against 98. A male commentator smirked: “Sweetheart, football isn’t mathematics.” Melbourne won 2-1. I wrote a detailed breakdown on my blog, dissecting every phase with the data itself, and it spread through the Australian analytics community.

That episode gave me a non-negotiable professional rule: open with the numbers, close with a raw data table so anyone can check the work. I removed the phrase “I think” from my analytical writing and replaced it with “the data indicates”. But I learned a second, less discussed lesson. When the numbers are on your side, people can still reject you for a reason that is not in the spreadsheet. If you want to last, you have to accept that some people will never read column three.

Data has no gender, but how a number is received does. A man presenting a 2.4 expected goals model is called cold and rigorous. A woman presenting the same model is called someone who does not understand football. I do not use that difference as a complaint. I use it as an input when designing an article: state the source, state the method, state the limits, and let the reader decide.

Anatomy of the first 15 metres

In freestyle and butterfly, a swimmer may stay fully submerged for the first 15 metres after the start and after every turn. Backstroke has a similar limit. The rule, introduced to stop athletes swimming entire races underwater, accidentally created a second event contested entirely beneath the surface, and that second event appears in no official results sheet.

In a 100 metre race there are four 15 metre underwater segments: one after the start, three after turns. In a 200 metre race there are seven. The total underwater distance of a 200 metre swimmer can reach 105 metres. More than half the race happens where no spectator can see it and no column records it.

Three factors decide the quality of that underwater segment: the number of dolphin kicks, the amplitude of those kicks, and the timing of the transition from kicking to the first stroke cycle. A properly executed dolphin kick generates forward impulse with far less drag than any arm stroke at the surface, because the whole body sits in relatively undisturbed water. That is why short-course world records in butterfly and freestyle keep falling to athletes with exceptional kicking foundations.

Break out too early and you trade hydrodynamic advantage for vision and breathing rhythm. Break out too late and you trade oxygen for instantaneous speed, and you pay for it in the final metres. The optimal point differs by athlete and by swim within the same meet, depending on accumulated lactate. That is why I treat 15 metre data as the most valuable and hardest to collect data in swimming.

I record kick counts and breakout points from high-resolution video at 60 frames per second for every swimmer I care about across a five-day meet. It is manual work that cannot be fully automated, and it takes about seventy percent of my analysis time at a major meet. Most analysts skip it because it is slow and produces no tidy number.

The negative split trap

Official results split a race into 50 metre segments. A 200 metre race gives the reader four numbers. A 400 gives eight. Add reaction time and you have a string that looks scientific and is very easy to misread.

The most common error is drawing a fitness conclusion from split structure. People say: this swimmer accelerated at the end, so she has a better aerobic base. That is a flawed inference about mechanism. A faster final segment can come from three very different causes: energy conserved early, a rival fading in the next lane, or simply a first segment swum too slowly relative to the swimmer’s real capacity.

I once analysed a dataset of several thousand 200 metre swims by national-level female swimmers over five years. Negative splitters — those whose second 50 of each 100 was faster than the first — showed no average performance advantage over positive splitters. What separated the groups was not fitness but physiological phenotype. Negative splitters tend to be slow starters; positive splitters tend to be fast starters. Two phenotypes, two training strategies, one performance distribution.

So when you see a swimmer with the fastest closing 50 in the lane, you do not yet know anything about their fitness. You know one certain thing: they finished. The rest is a story you are telling yourself.

There is also a technical blind spot in 50 metre splits that few notice. Turns sit inside each segment. A swimmer with brilliant turns can compensate for weaker straight-line speed, and a total time will never tell you that happened. At elite level the turn differential between an excellent and an average turner can reach 0.25 seconds per turn. Over seven turns in a 200 metre race that is nearly 1.8 seconds — the gap between a medal and a place off the podium.

Stroke rate and distance per stroke: two curves that never meet in the middle

Every stroke cycle couples two variables: how fast you turn the arms and how far the body travels in one cycle. Their product gives speed. Mathematically, infinitely many combinations produce the same speed.

The danger is that the two curves are not independent. Pushing stroke rate too high shortens distance per stroke because the body never finishes gliding. Stretching distance per stroke too far lowers rate and creates dead spots between cycles, making speed oscillate like a saw blade.

The analyst’s territory is how those curves deform under fatigue. As lactate accumulates, swimmers drift toward higher rate and shorter distance. It is almost automatic, requiring no conscious decision. If a swimmer holds distance per stroke steady while raising rate in the closing metres, they own what I call technical reserve.

Tracking many women’s events at international meets, I keep seeing the same pattern. Winners of 100 metre events are not the ones with the highest stroke rate. They are the ones with the smallest decay in distance per stroke between the first and second 50. Average decay among winners tends to sit around 3 to 5 percent; among the athletes behind them, 8 to 11 percent. The absolute value of distance per stroke has little predictive power. Its durability does.

This is the kind of conclusion that makes me uncomfortable, because it cannot be verified from public data. You need someone counting every stroke cycle, or a computer-vision system tracking the shoulder joint. In an annual season, national meets rarely have that. I have to accept that most of the insights I care about most cannot be presented as a table a reader can check.

The 200 metres: an allocation problem, not a speed problem

A 200 metre race is the most interesting event in the sport because it forces an allocation decision before the race begins. Swimmer and coach must choose an energy distribution across four 50 metre segments, and that choice is encoded in the body before the whistle.

Broadly there are three strategies: aggressive early (fastest opening 50, then steady decay), even distribution (minimal variance between segments), and conservative start (economical first 50, explosive third and fourth).

All three can win, and none has an overwhelming mathematical edge. What decides it is the fit between strategy and physiological phenotype, plus how rivals in adjacent lanes allocate. Choose an aggressive start when the swimmer next to you also starts aggressively and has a better sprint foundation, and you have walked into a race where your win probability collapses.

In women’s 200 metre freestyle, one repeat champion at major championships produced a distinctive allocation: fast first 50, noticeably slower second, third close to the first, and a fourth that was the second fastest segment of the race. That shape creates two tactical rest points inside a race lasting under two minutes. Rivals must decide: follow on the second 50 and burn, or hold rhythm and let her break away on the third. Neither option is good.

For the market this has direct application. When a model quantifies a 200 metre race as a single distribution, it ignores the possibility that a swimmer owns several distributions depending on the situation. You are not betting on a fixed athlete. You are betting on a set of tactical decisions that may occur, and that set changes with lane draw.

Relays: where the model dies

The event where my spreadsheets fail most often is the relay. A team’s time is not the sum of four individuals. It is the sum of four individual times, minus or plus three changeovers, plus the mutual dependence between legs.

An effective changeover happens when the incoming swimmer’s hand touches the wall exactly as the outgoing swimmer’s feet leave the block. Timing systems record block reaction time, but the decisive moment is the interval between the incoming swimmer initiating the final turn and actually touching. In that window of a few hundredths, the swimmer waiting on the block must commit.

That phase lag cannot be measured by any field in an official result. It depends on the noise of the arena, the sightline of the waiting swimmer, whether the pool water is clear or murky, and whether the incoming swimmer touches with the right or left hand. None of these variables appears in any commercial database, which is why relay betting models carry far more uncertainty than individual models.

Over five years covering national and international meets, I have recorded a significant share of relay teams disqualified for changeover faults, and most were not technical failures by the athlete but errors in the timing decision. For a dominant team, this risk is usually underpriced in the market, because the public only sees four names and four personal bests.

That is a very concrete information edge. When you know the real gap between the best team and the third-best in a relay is around 1.5 seconds, and you know the disqualification risk at each changeover is an independent variable that does not scale neatly with talent, your model produces very different odds from the market. I have profited from exactly this gap several times in recent seasons.

The world lane map: three axes and one trough

Over the past four years the world map of swimming has shifted. The United States holds overall depth. Australia holds women’s freestyle and backstroke. China has risen in men’s sprint and medley events. France has built a new centre in men’s medley. Canada and Italy remain strong in technical women’s events.

What is interesting analytically is the speed at which these axes move. In the men’s 400 metre individual medley at the Paris 2026 Olympic Games, the world record fell by a margin far beyond the typical improvement of a world record. An improvement beyond normal bounds implies a change in mechanism, not a lucky day.

In that case the mechanism was a better-executed underwater phase, particularly in butterfly and freestyle, combined with exceptional kick amplitude that is not available to pure butterfly specialists. It is the clearest illustration of a claim I will repeat: new records in modern swimming tend to be built where the results sheet does not look, not in the straight-line swimming at the surface.

In the men’s 100 metre freestyle, the world record at Paris 2026 broke the 46.60 second barrier for the first time. Reading that race, the standout was not the closing segment but the structure of the opening 15 metres: a long underwater phase, high maintained velocity, and a later than usual breakout that lost no momentum. That is precisely what models built on prior results miss, because the underwater phase never appears in official split data.

The trough of the map remains Southeast Asia, and I say this as someone watching both shores. The region produces a handful of outstanding individuals but has not built a data system to find and nurture talent at scale. Meanwhile, at the top of the sport, training data is collected daily, standardised quarterly, and fed back into the programme as part of the cycle.

Vietnam’s lane: results without the data columns

I have followed Vietnam’s national championships and regional games for years, and I keep noting the same gap. Results exist. Data does not.

Take one example. If I want the 15 metre breakout time of Vietnam’s leading male 800 and 1500 metre freestyler, I have to time it myself from broadcast video. If I want his stroke rate and distance per stroke at the fifteenth length versus the fifth, I have to count. No domestic database publishes these metrics in a downloadable, analysable form. That says nothing about the athlete’s ability. It says something about an analytics infrastructure gap.

The consequence is that important resource allocation decisions in regional swimming are often made on a feel for potential rather than on a distribution of improvement probability. Who should be invested in for the next four years? The technically correct answer is whoever has the steepest improvement slope on the best technical base, not whoever has the best absolute time at fifteen. But measuring slope requires monthly data series. Those series do not publicly exist.

I also have to admit something I remind myself of whenever I speak on this topic. Being in Brisbane with access to international databases does not make my judgement about young Vietnamese talent more accurate than anyone else’s. I see results. I do not see the five a.m. sessions in a pool with no hot water. The data gap runs both ways, and the direction where I lack data is the direction that matters more for developing human beings.

In developed swimming nations, the scouting network both finds genius and produces losses that are hard to see: families relocating across a country for one training place, fourteen-year-olds pushed into a twenty-year-old’s training load, shoulder injuries accumulating in no spreadsheet. I do not want that data system imported wholesale into a swimming nation still building foundations. That is why I keep a limits section at the end of every analysis, even when it makes the article less attractive.

When the market prices a number that does not exist

I price things for a living. My daily job is comparing model probability with market-implied probability through the odds, and finding where the two diverge enough to be worth money.

In swimming, most mispricing comes from one source: the market treats the most recent best time as the entire story. A swimmer who has just broken a national record is priced as if that performance is their permanent state, while my data shows most swimmers swim within one percent of their best only a few times a season. The gap between peak and baseline is one of the strongest predictive variables in the sport and the market barely uses it.

The second source is the market reading relays as the sum of individuals. I explained above why that is structurally wrong. There are moments when a relay team’s odds are set well below true probability simply because no individual on it tops a world ranking. My model learned this over several seasons, and the most expensive lesson was Kazan.

The third source, and the one that cost me most early in my career, is mispricing tail events. The market prices disqualification risk as nearly impossible, while historical frequency is far higher. Backing a small-probability event is not a good strategy if the payout is not commensurate; but avoiding the other side once the risk premium has been compressed is a valuable defensive decision.

I do not believe in emotion. I believe in a data series longer than your emotion. But precisely because I believe in long series, I know they contain points the model cannot see, and in swimming those points cluster in exactly three places.

Three blind spots of the spreadsheet

First: the opponent. Swimming is a sport where someone else’s result directly shapes your decision, yet standard prediction models drop this variable because it is nearly unquantifiable. A swimmer alone in the water allocates energy completely differently from one who is 0.4 seconds behind a rival at the second 50. A coach cannot choose the allocation during the race. There is one decision, made in an instant, and the results sheet records only the consequence. My model predicts the optimal allocation, not the actual decision.

Second: undisclosed physical condition. In an annual season, athletes often compete in a heavy training block, and mid-season times do not reflect peak capacity. A model using mid-season results to predict end-of-season results will be systematically wrong. I have watched swimmers undervalued all season produce improvements far beyond any linear projection at the decisive meet. The cause is usually a taper cycle known precisely only to the coaching staff.

Third: officialdom. A recalled start, a butterfly kick judged illegal, a backstroke turn judged illegal, a changeover flagged from an angle the camera never saw. These events are not evenly distributed across athletes and appear in no probability model. At Kazan I learned that a single blind spot is enough to destroy a 99 percent model. In swimming, a race can be destroyed by one hundredth of a second in a place nobody watched.

What I want you to carry from these three blind spots is not scepticism about data. Data remains the best tool we have for understanding a race. What I want is a habit: whenever you see a number presented without its limits, ask what that number is hiding. Most of the time the answer sits in the first 15 metres, on a starting block, or in a training session nobody recorded.

Pricing a person is not a calculation

There is a story I retell often when asked how I look at sport through numbers and still keep respect for people.

In 2026, riding the reputation built at the World Cup, I was hired by a large Brisbane betting firm as a transfer-window consultant. My first task was assessing a young Australian talent loaned by a big European club. The media framed it as the Daniel Arzani valuation race. I presented the data: his average running distance sat below the average for forwards in that league, his dribble frequency was roughly two per match, and he had a history of two ACL ruptures. I concluded the deal would fail.

At first the sporting director objected, saying I saw people as machines. Two seasons later Arzani had played about twenty minutes at that club. I was right about the conclusion, and I was never comfortable with how I was right. In my spreadsheet he was a set of variables. In real life he was a twenty-year-old in a foreign country with two surgeries and a belief being eroded month by month.

Valuing a player is not a calculation; it is a contest between belief and the spreadsheet. I kept my professional conclusion and changed only how I presented it. Since then every valuation report I write has a section listing what the numbers cannot measure: tolerance for pain, the ability to relearn a movement after injury, the ability to live alone at twenty in a city that does not speak your first language.

In swimming that section matters even more. A swimmer spends eighteen to twenty-five hours a week in water — an environment that forbids conversation, forbids seeing an opponent’s face, forbids almost any social interaction for the length of a session. Much of the psychological data observable in team sports simply does not exist here. The feeling of being underwater, the structured solitude of a long session, is data of another kind, and I have no instrument to measure it.

Limits of the data

This is the section I write at the end of every analysis, with the same seriousness as the opening.

First, the metrics above depend on timing quality and video quality. At national level, manual timing error can reach a tenth of a second — enough to reverse the order within a lane. At international level machine error is far smaller, but 15 metre split data is almost never officially published. Most figures in this article come from my own video analysis and therefore carry the subjective error of the analyst.

Second, I cannot measure spirit. I have no variable for a swimmer who cried in the changing room three hours before racing, or slept four hours because of nerves. I once made a correct prediction at a major tournament based entirely on pressure free-kick data and was accused of being mechanical and ignoring national spirit. I answered with a line I still stand by: emotion is data too, we simply do not yet have the instruments to measure it. That is not evasion. It is an admission about the limits of the trade.

Third, and most importantly, swimming data has a structural bias: it records only what happens at meets with measurement systems. Everything that happens in pools without timing systems, in countries that do not publish data, and among swimmers never entered into an international database does not exist in my model. When I say my model rates a swimmer as low probability, that may only mean I have never seen them swim.

Fourth, a professional note. I work for the betting market. Every analysis I publish has a practical purpose, and that purpose is not to celebrate the sport. Read my work the way you read a valuation report, not the way you read a tribute.

What I will watch in the next stretch of the season

The annual season is entering the phase where the physical current becomes clearest, and that is when signals appear before headlines.

I will watch three things. First, the decay in distance per stroke between the last two 50 metre segments among leading 100 metre swimmers; this anticipates outcomes better than any total time. Second, breakout timing in morning heats — a swimmer breaking out later than usual in a heat is often preparing for a faster final. Third, changeover faults in relays during the accumulation block; if fault frequency rises, it signals a disconnect in team organisation and usually predicts failure at the decisive meet.

For Vietnam and the region, the signal worth tracking is the emergence of a generation with better technical foundations underwater. If that happens, it will show first not in the results table but in split structure — middle segments relatively faster than the opening, something only systematic turn and breakout coaching can produce.

I do not know who will win. I know that every race to come will contain roughly a hundred metres swum beneath the surface that no results sheet will record, and most spectators will never see it. If you want to read a swim correctly, start from the darkest part of the lane.