When the Data Falls Silent: The Map of Limits in Elite Swimming Analysis

In a March morning at the Brisbane Aquatic Centre, I sat in the stands with a spreadsheet open, and on screen was a line of numbers that made me pause: a swimmer finished the 200m freestyle 1.4 seconds faster than the previous meet, but their average stroke rate had dropped by 3.1 per minute. For most spectators, that is good news. For me, it is a question without an answer. A faster time is always easy to celebrate. But when stroke rate falls while the time improves, what is happening underwater could be technical refinement, or it could be a swimmer conserving energy for a target not yet revealed, or it could simply be a day when the pool was cold enough for the body to operate unusually efficiently. Three explanations, three completely different consequences, and the numbers do not tell me which one to choose.

That is why I am writing this. Not to tell the story of a triumph that fell within prediction, but to talk about the moment the numbers fall silent — the most dangerous moment in my profession. Kazan was the day I learned that a 99% probability can still die on the betting table. But there is another lesson I learned later, and it is the one that has shaped how I write about swimming today: the most dangerous thing is not a wrong number, but an empty dataset dressed as a complete one. A spreadsheet with no data can still present as beautifully as a spreadsheet with data. And when the numbers fall silent, the least disciplined person in the room will be the loudest.

Numbers have no gender, and in swimming, they have no style either. They cannot distinguish a start that flinched from excitement from one that was executed with discipline. They cannot measure the feel for the water, which any coach at St Peters Western will tell you decides a third of the final result. Numbers have no gender, but those who read them do — and those who interpret them carry even more bias than that.

I entered this field from a different foundation. Nearly thirty years of observing the sports industry, including a long stretch as a sports betting analyst in Brisbane, taught me one thing about the Australian market: people love swimming for national emotion, but they bet on it with blind faith. There is no xG for swimming. There is no integrated model that substitutes for intuition. And that is both the opportunity and the trap. Five years of deep swimming coverage have given me enough data to know what data I am missing — and knowing precisely that is the entire job.

I do not believe in emotion. I believe in a data chain longer than your emotion. But I also know that the longest data chain can still break at the exact point you need it most. This article is a map of those break points.

Context: why swimming data differs from every other sport

To understand why swimming analysis falls into the "fake numbers" trap more easily than football, you need to understand one structural feature: swimming is a sport of numbers that appear transparent but are actually ambiguous. A football match has hundreds of discrete, labelable events — shots, passes, tackles. Swimming has only one continuous time series, and everything you want to know is compressed into a single curve.

Take Ariarne Titmus's 400m freestyle result. The final time is the clearest number, but it is the sum of four laps, one start, seven turns and one finish. When you break it down, each segment has a different data resolution. The start has sensors on the block, giving you a reaction time accurate to a hundredth of a second. The middle swimming depends on stroke rate and distance per stroke — two metrics that devices capture but with large error margins depending on the pool, the camera and the person labelling. The finish is affected by that specific meet's sensor system and that specific lane's water.

In other words, a swimming results table looks like a complete dataset, but it is actually five datasets of different quality stitched into one timeline. That is why I always tell my editors: never present a swimming table without noting which segments are sensor-derived and which are video estimates. Because when you remove that footnote, you are selling an empty dataset under the guise of a complete one.

The second context is the home-and-cycle problem. Swimming has no "annual season" in the football sense. It has four-year cycles cut by odd-year world championships, and within each cycle are meets of entirely different priority. A swimmer can swim poorly at one continental meet in July because it is a disguised training session, then swim brilliantly at another continental meet months later because that is their peaking point. Looking at the results, those two outcomes can look identical in numbers. In meaning, they sit at opposite ends of the spectrum.

I once called this "the problem of the timeline that does not say what it is." A 1:47 in the 200m freestyle can be a worrying result if it is a designed peak for a major target, and a step forward if it is a mid-cycle test. Results tables do not distinguish the two. But the coach knows. The swimmer knows. And the bookmaker, if smart, should know too.

The third problem, and the one I want to spend most of this article on: swimming is where the media's storytelling pressure collides head-on with the silence of data. Football has enough data to tell a rough story that stands up. Swimming does not. There, a good performance in a single swim can be amplified into a big story simply because there is nothing else to say that day. And because swimming is Australia's national soul sport, the storytelling pressure here is many times greater than elsewhere.

The core: a chain of evidence — the points where numbers can decide

I want to move from the most solid data zone to the most ambiguous, because that is the only way to draw an honest map of limits. For each zone I will state what evidence it rests on, and where its error lies.

Zone one: stroke rate and distance per stroke

This is the foundation of all technical swimming analysis, and also where most people misunderstand most. Stroke rate and distance per stroke have a classic trade-off: swimming faster usually means increasing stroke rate, but increasing stroke rate usually reduces distance per stroke if technique does not keep up. Final speed is the product of these two divided by a constant. Sounds simple. But it is an equation with two unknowns, which is why it generates endless debate.

Suppose a female swimmer improves by 0.8 seconds in the 200m freestyle over a month. There are at least three paths to that improvement. Path one: she raises stroke rate by 4%, holding distance. That is fitness-driven improvement. Path two: she holds rate, raises distance by 3%. That is technique-driven. Path three: she raises rate by 2% and distance by 2% simultaneously. That is improvement from both, meaning a difficult and far more credible jump.

The difference between these three paths is invisible in the final time, but it decides everything about whether the result is sustainable. Fitness-driven improvement can evaporate after a taper. Technique-driven improvement is usually more durable but takes months. Simultaneous improvement in both is often the sign of a young swimmer at exactly the right stage of development — and that is where I pay the most attention.

But here is what I learned over five years: stroke rate and distance per stroke are not raw data, they are processed data. They depend on which time window is chosen for measurement. If you measure stroke rate across the whole 200m, you get a different number than if you measure the first 50m and last 50m separately. A negative-split swimmer will have rising rate across laps, and the average conceals the entire structure. When a table gives you only average rate without segments, you are looking at an empty dataset in the guise of a complete one.

Zone two: split structure and pacing strategy

If I were allowed only one type of data to analyse a swim, I would choose 50m splits. Not the final time, not the results table, but the splits. Because splits tell you the story of how a swimmer distributes energy, and energy distribution is where strategy collides with fitness.

There are three classic split models. Negative split: the back 50m faster than the front. This is the model of the confident, well-controlled swimmer. Positive split: the front faster, usually a sign of someone who burned too early. Even split: laps roughly equal, the model of the disciplined, best-prepared swimmer.

On paper, the even split sounds ideal. And in most cases, the world's top swimmers do not swim even. They swim with slight variation — often lap three deliberately faster than lap two, to create a psychological jolt and force rivals to react. This is the detail the results table never tells.

Imagine two swimmers finishing in the same time. Swimmer A swims an even structure. Swimmer B swims with lap three slower and a final lap that spikes faster. In the time, they are identical. In the future, they are completely different. B has greater reserve power but higher misallocation risk. A is more stable but their potential ceiling may be lower if B learns better distribution. The same number, two opposing implications. This is precisely why I never conclude from the final time alone.

Here I want to tell a story from my own direct observation. I once tracked a young Australian female swimmer across three consecutive meets. At the first, she swam a clear positive split and finished fourth. At the second, she swam flatter and finished third. At the third, she swam a negative split and finished second. The results table told everyone she was improving steadily. The splits table told me she was changing her swimming step by step, with control, and that is why I put her on my long-term watch list. A year later, she made a major final.

I am not saying split data predicts the future. I am saying it gives you something the final time does not: the order of change. And the order of change is often more important than the change itself.

Zone three: start and finish

This is the most technically solid data zone, because block sensors and touchpad systems measure to a hundredth of a second. But measurement solidity does not mean meaning solidity. And this is where I often see inexperienced analysts draw conclusions that are too strong.

Reaction time in a single swim is one of the most misleading metrics. It varies widely between swims for the same athlete, because it depends on the specific pool's sound signal, on when the swimmer chooses to react, and on whether they are in an acceleration phase. A swimmer can have a 0.61s reaction at one meet and 0.74s at another without changing start technique at all. If you look at a single swim, you will misjudge their ability.

My rule: never judge reaction time from a single swim. You need at least five swims under the same pool conditions to say anything.

But if you have five swims, reaction time becomes a superb window into an athlete's nervous system. It tells you focus, stress handling and stability under sensory pressure. At elite level, the difference between gold and silver in sprint events often lies in exactly these hundredths.

Finish technique is the same. It is a strangely undervalued skill. In a sprint final, a poor finish can cost 0.1 to 0.15 seconds — more than the gap between gold and bronze in many events. But finishes are not trained separately in many programs. It is treated as a natural skill. That is one of the biggest gaps in modern swimming analysis, and where a nation with good sports science can create a systemic edge.

Zone four: short course to long course conversion

This is the most ambiguous data zone, and it is also the zone betting market analysts exploit most. Short course (25m) has more turns than long course (50m). Because the turn is a separate technique that can be very fast or very slow depending on the swimmer, converting times between pools cannot be done with a simple linear formula.

Common conversion models use correction factors per event and distance. But the factor is built on population averages, not individual swimmers. A swimmer with poor turn technique will be undervalued in long course and overvalued in short course by the average factor. A swimmer with a long body and good underwater push will be the reverse.

Since covering swimming deeply, I have stopped using average conversion factors. Instead, I build a metric I call "turn deviation" for each swimmer: the gap between that swimmer's turn segment and the average of their peer group. This metric is far more stable than converted times, and it tells me whether a swimmer benefits or suffers when moving between pools.

Let me tell you a truth many bookmakers do not want you to know: short course and long course analysis are two separate markets, and most smart money sits in the intersection. When a swimmer goes very fast in short course, the crowd reacts by believing in an imminent record-breaking miracle. The smart bookmaker reacts by lowering that swimmer's value when returning to long course — because they know which conversion factor is being applied, and they know it is being applied wrongly to whom.

Zone five: the taper problem and peaking points

Here I want to go into the hardest part of swimming, the part I consider the biggest unknown in the entire sport's data. In most sports, athletes pursue peak form continuously. In swimming, they pursue peak form deliberately, and deliberately swim poorly in the preceding phase to preserve reserve for the exact peaking point.

This means a swimming results table is often systematically distorted. Major meets sit at the end of the cycle, where top swimmers have reached peak form. Mid-cycle meets often have top swimmers swimming at fifty to seventy percent of real ability. If you do not know which meet is which, you can conclude a swimmer is declining when they are in their biggest foundation-building phase.

I remember sitting with a senior coach in Brisbane who told me something I never forgot: "What you see at a mid-cycle meet is not my swimmer. It is my training unfolding in front of you."

Numbers with no reader are meaningless. And with swimming, the most important reader is not the spectator, not the bookmaker, but the training program itself. A number without a training footnote is a number that can lead you entirely astray.

The contrarian section: when correlation is not causation

Now I want to reach the part I consider most important, and the part I learned after my most expensive mistakes.

There is a widely held belief in Australian swimming analysis that a nation with a strong youth development system, many outstanding junior swimmers and a well-funded federation will inevitably produce sustained senior success. This belief is true in correlation but false in causation. And it is dangerous because it creates a false appearance of certainty.

Look at international data. There are nations with extremely strong youth systems that hollow out very fast at senior level, because they lack elite swimming infrastructure after age eighteen. There are nations with weak youth systems that nonetheless have a few world-class training centres enough to attract talent and develop them directly to the top. The correlation between "strong youth development" and "senior success" exists, but it is dominated by a third variable: the quality of the bridge between the two phases.

In Australia, we have a very special bridge. Centres like St Peters Western in Brisbane, under Dean Boxall, do not just develop young swimmers — they retain them through the hardest transition phase. That is not a natural outcome of youth development. It is a proactive strategy of a few individuals, and it cannot be replicated by increasing youth funding.

Here is what correlation does not tell you. Australia's swimming strength does not lie in the youth pipeline. It lies in the retention rate through the black hole of ages eighteen to twenty-two — a window where data shows most nations fail. If you look only at junior medal tables, you will misjudge any nation's future. If you look at retention rates, you will see everything before it becomes a headline.

I want to go one step further, into the contrarian zone I consider most important in this article.

In my predictive models, I once found a variable with a very strong correlation to a swimmer's long-term success: the average number of years in the same training program. Swimmers who changed programs regularly had more volatile results and shorter peak careers. The correlation was beautiful. I nearly built an entire model around it.

But then I realised this: that variable might be measuring exactly the opposite of what I thought. Swimmers who change programs often do so because they have a problem — injury, conflict, decline. In other words, changing programs does not cause decline; decline causes changing programs. The causal relationship runs backward from my original hypothesis.

That is my Kazan lesson applied to swimming. Kazan taught me that a 99% probability can still die on the betting table. Swimming taught me that a correlation with an 0.8 coefficient can still be telling the story in reverse. In both cases, the number is not wrong. What is wrong is the story I attached to it.

And here is the practical consequence for the Australian betting market. When a famous swimmer moves to a new training program, the crowd reacts by lowering their value — based on the principle that change is risk. But if you know that changing programs is often the consequence of an existing problem, the right question is not "will this transition hurt" but "has the original problem been solved". Two questions leading to two entirely different betting behaviours, and most of the crowd is answering the wrong one.

The blind spots that numbers never illuminate

I want to give a section of its own to the things five years of swimming taught me are data in another sense, but cannot be measured in hundredths of a second.

The first blind spot is feel for the water. No sensor measures feel for the water the way a coach watches a swimmer and immediately knows whether they are "catching" well today. Feel for the water is the composite of hundreds of micro-signals — hand tension, wrist angle, timing of the catch, breathing rhythm. Numbers see the result of all these but cannot decompose them into causes.

This matters to the writer because it creates a gap between what experts see and what the crowd sees. When a swimmer is slow, the crowd says "they are declining". The person with feel for the water says "they are at the wrong phase of the cycle". Two readings of the same number, and one is correct with far higher probability.

The second blind spot is psychological state under pressure. In sprint events, the gap between top swimmers is often so small that psychological factors become the deciding variable. And we do not yet have reliable enough tools to measure them. A swimmer can be strong in fitness and technique but weak in handling final-night pressure, and the numbers will never show it — until the final comes and they fail for reasons the numbers cannot explain.

Emotion is also data, but we do not yet have enough tools to measure it. I first wrote that line in an analysis of a major final, and it remains the line I return to most.

The third blind spot is coach decisions. This is the variable my models never capture, and I suspect no model ever will. A coach can decide to pull a swimmer from one event to focus on another. That decision usually does not appear in the results table, but it shapes that swimmer's entire career. Numbers see the final result. They do not see the door that was closed.

The fourth blind spot, and the one I consider most underrated, is the scouting network. Swimming, like many sports, both finds genius and creates lottery tickets and broken families. A young swimmer discovered at twelve and placed in an elite program can meet one of two fates. Either they become a star, or they become a forgotten story in a ranking no one remembers. And the numbers, once again, record only the outcome. They do not record the cost.

I work in the betting industry, and I will not pretend my industry is immune to this. In fact, I think the betting industry has a special responsibility not to turn children into instruments. Every number in my table has behind it a person with a gender, with emotions, who can die even if the probability is 99%. That is what I remind myself every morning before opening the spreadsheet.

Direct observation experience: what numbers did not teach me

I have said that thirty years of observing the sports industry taught me more than any statistics course, and I believe it. But I want to make that concrete with what I have watched directly.

When I began covering swimming seriously, I brought all my football-analysis habits. I looked for advanced metrics. I looked for predictive models. I looked for certainty. And I quickly found that swimming does not offer certainty the way football does. It is a sport of coarse data resolution, where the real analytical skill lies in knowing when to conclude nothing at all.

There is one meet I think about often. A young Australian swimmer competed in an event at an international meet and finished well below expectations. All my metrics indicated she was at peak form and should have finished faster. I had prepared an article saying this was a missed opportunity, a statistical shock.

Then I spoke to an acquaintance in the coaching world who knew her program. He told me one thing: she had just been through the highest-intensity training week of the entire cycle, and this was a disguised training meet the program deliberately put her in to test a new technical adjustment.

My numbers did not know this. They saw a poor result and concluded a missed opportunity. The truth was the exact opposite. And that was the day I wrote a new rule in my notebook: before analysing any swimming result, ask what this meet means to that swimmer's training program. If you cannot answer that question, do not write.

I have many times written a line in my analysis sheet: "Is this a model-applicable race, or just an ordinary race?" Nearly thirty years in the profession taught me that, at elite level, sometimes an ordinary race has more analytical value than a final. And vice versa.

Closing: signals for the next cycle

I am not closing this article with a summary. Summaries are for those who believe they have understood everything. I close with what I will watch next, and I invite you to watch with me.

The first signal is turn deviation. I will track it in young swimmers moving from short course to long course in the coming cycle. If their deviation narrows over many swims, that is a sign of systematically improving turn technique. If it fluctuates randomly, that is a sign of an unstandardised technique.

The second signal is split structure at mid-cycle meets. I will look for swimmers swimming negative splits at unimportant meets, because that is a sign of confidence and control. At training meets, most swimmers swim loosely. The one with structure is the one training with purpose.

The third signal is finish frequency. I will look for swimmers with small but steady improvement in their finish segment, because that is an indicator of a technique being trained deliberately — and it often comes with larger improvement elsewhere.

The fourth signal is retention rate. I will track how many young swimmers stay in the same program through the eighteen-to-twenty-two transition. This is a weak signal for immediate results but a strong one for the long term, and it often shows a nation's development direction before the medal table says so.

And the final signal, the one I consider most important, is silence. I will listen to the silence of data. When a swimming result is celebrated as a miracle but comes with no split data, I will be the first to ask: what are we not being shown? When a swimmer is criticised as declining but their program is not disclosed, I will be the one to ask: who is reading this number into which story?

Because this is the truth Kazan taught me and swimming reinforced: the most dangerous thing is not a wrong number. The most dangerous thing is a right number placed into a wrong story. And the wrongest story is one built on an empty dataset dressed as a complete one.

Numbers with no reader are meaningless. Numbers have no gender, but those who read them do. And my task, as a storyteller through data, is not to create certainty that does not exist. My task is to show you exactly where the numbers stop speaking and the human begins — so that when I draw a conclusion, you know how much data, how much inference and how many unresolved limits that belief rests on.

I do not believe in emotion. I believe in a data chain longer than your emotion. But the longest data chain can still break at the very last link, and I want you to be the one who sees where it breaks before you place your trust in it.