The Empty Golf Data Table and the Discipline of Not Inventing Numbers: Notes from an Analysis Week in Nagoya

The Empty Golf Data Table and the Discipline of Not Inventing Numbers: Notes from an Analysis Week in Nagoya

It was 3:58 in the morning on a Wednesday. A seventh-floor apartment in Nakamura ward, Nagoya. The left monitor showed an elevation map of a golf course in the Tokai region. The right monitor showed a CSV file opened in a plain text editor. Sixteen columns: average driving distance, fairway percentage, greens in regulation, proximity to the hole, scrambling rate, putts per green in regulation, and four Strokes Gained columns split by area. The header row sat there, neat and complete. Beneath it was nothing.

The tournament began on Thursday. I reopened that file four times in two and a half hours, each time with a different assumption about where I had gone wrong: wrong file path, wrong date format, wrong character encoding, wrong time zone on sync. On the fifth attempt I opened the system log. There were no errors. No data had simply been written.

I sat still in front of that empty table for a long while. In seventeen years in this trade I had seen data contradict me, watched my models collapse, predicted six of ten final rounds wrong in a single season. I had never faced an empty table right before a tournament teed off. And I realised that what unsettled me was not the missing numbers. It was the very clear, very specific urge rising in my head: fill it in. Anything. An estimate. An interpolation. A belief.

This piece is about that moment. About what I call the discipline of null handling, and why it matters more than any model I have ever built.

---

Context: how a week of golf data gets built

To understand why an empty table is an event, you have to understand how an analysis week runs. I work for the Japanese market, where audiences follow both the Japan Golf Tour and international events, and where the reading culture around numbers differs sharply from how a general audience elsewhere absorbs them.

At the lowest layer sits shot-tracking. The PGA Tour has operated ShotLink since the early 2000s, recording the coordinates of nearly every shot in nearly every round and converting them into distance and proximity metrics. In Japan the equivalent infrastructure is not as broad; many events carry only round-level aggregate data, hand-recorded by volunteers, with cross-round error large enough that a single model cannot be applied uniformly.

The second layer is independent data providers and aggregation platforms, where analysts pull data to re-run. The third layer is me. I do not generate raw data. I read it, verify it, place it beside tactical context, and write down what it permits me to say and what it does not.

The central instrument of the trade is Strokes Gained, developed by Mark Broadie of Columbia Business School, published in a 2026 study and popularised through his 2026 book. The core idea is elegant: every shot is scored by the difference between expected strokes before and after the shot, against the tour average. That allows a three-metre putt and a 320-metre drive to sit on the same scale.

The four standard areas are Off the Tee, Approach the Green, Around the Green and Putting. The first three combine into Tee to Green. Each area has sub-metrics: approach distance, green-hit rates by distance band, sand save rate, putting inside three metres.

Precisely because every area has sub-metrics, an empty table is not simply missing numbers. It is sixteen distinct holes, each one fillable with a different assumption, each assumption entirely plausible-sounding.

That is the architecture of temptation in this profession. It does not arrive as one large lie. It arrives as sixteen small guesses, each defensible.

I lived the reverse in 2026, at twenty-four, when I started doing data analysis for a Nagoya football club just relegated. I built an expected-goals model by hand from video and missed a four-match losing streak because I failed to weight home advantage correctly. My model was right four times in ten. I sat through the footage again and understood something that remains the foundation of everything I write: raw data carries no meaning on its own. Meaning is manufactured by the context the analyst is responsible for constructing.

In 2026 I worked as a data contributor for a Nagoya sports outlet. In a major match I computed pressing metrics and concluded one team was pressing well. I ignored the opponent's running distance after the seventieth minute. That team lost from a winning position. I published a public self-criticism and since then every analysis of mine carries a running-intensity chart in fifteen-minute bands.

Those two lessons combine into one principle, and that principle collided with the empty table at 3:58 on Wednesday: when data hides its face, margin of error becomes the guide. The catch is that the error must be declared, not disguised as data.

---

Layer one: technique and data — the line between inference and fabrication

When the technical table is empty, the first question I must answer is: what exactly am I missing, and can that gap be substituted from elsewhere?

In golf analysis there are three kinds of gaps, different in nature, demanding three different responses.

The first is mechanical. The data exists somewhere; it simply has not reached me. A tournament has ShotLink but the provider has not finished syncing. This gap is solved by time, not by guessing. I wait.

The second is structural. The data was never created. A rural Japanese course may host a regional event with nobody recording shot coordinates. No ShotLink, no TrackMan, nothing but an aggregate leaderboard on the clubhouse wall. This gap cannot be solved by waiting. It can only be solved by downgrading the question: instead of asking why a player lost strokes on approach, I ask how that player's scoring distributes across the front nine and the back nine.

The third is cognitive. The data is sufficient, but it does not answer the question I am asking, and I have not noticed. This is the most dangerous kind, because it does not look like a gap. It looks like a conclusion.

At the technical layer I start by building a comparison table: four Strokes Gained areas, corresponding raw metrics, and a margin-of-error column. If a cell is empty I mark it empty. I do not insert an industry average, because the industry average of a Japanese course differs from a Florida course in grass, humidity, altitude and competitive culture.

| Metric group | Function | Minimum condition to use | |---|---|---| | SG by area | Locate lost strokes | Shot-level data available | | Putting stability | Exclude hot streaks | At least ten rounds | | Course profile | Estimate fit | Grass, humidity, altitude data | | Nine-hole split | Detect physical decline | Hole-level data |

The table looks simple. The notable column is the third. When every cell in that column reads no, then the metrics in the first column, however elegant, are not permitted to support any conclusion.

That is the entire content of the discipline of null handling.

Every number is a confession that has not yet been written down.

That sounds like a line of prose, but it is an operating rule. When I look at a metric I always ask what it is confessing about the conditions that produced it. A high green-hit rate on a narrow course confesses that the player controls trajectory. The same rate on a wide course confesses only that the player is playing safe.

---

Layer two: player and form — the sample-size problem

If the technical layer asks why, the form layer asks how long, and how many times.

The biggest temptation here is linear extrapolation from a short run. A player putts well for four rounds and a story about rediscovered feel appears instantly. In reality four rounds is a sample where random variance easily exceeds real signal.

My method is to always place a short run beside a longer window. If putting over the last four rounds is far above the ten-round baseline, I record two possibilities and choose neither: a genuine technical change, or statistical noise. Only when the short run repeats at round six or seven do I begin to lean toward the first.

I do not believe in luck; I believe in cultivated probability.

“Cultivated” is the operative word. Three consecutive good weeks is not the same as four good rounds. A cultivated run leaves traces in practice data, equipment changes, setup adjustments, or repeating course conditions.

At the player layer I track four axes: world ranking position — the system established in 2026 that allocates major exemptions; the age curve, which in golf decays unevenly, with clubhead speed usually going first and green-reading usually lasting longer; injury risk, which I read only from indirect signals such as withdrawals and abnormal scheduling changes; and major-championship record, where conversion from contention to victory is a separate variable that regular-season form cannot infer.

When all four axes are empty, the only honest sentence is that the question lacks the data to be answered. I know that is unexciting. But in seventeen years I have never seen anyone lose credibility for saying they lacked data. I have seen plenty lose it for speaking too early.

One example stays with me: Hideki Matsuyama won the 2026 Masters, becoming the first Japanese man to win a major championship. Before that week, one wave of analysis insisted he was ready; another insisted he could not hold his nerve on Sunday. Both waves lacked the same variable: how he would handle a weather-delayed schedule and forty years of national waiting. The result confirmed neither wave's method. It only confirmed that what cannot be measured still exists.

---

Layer three: tournament system — weight is not prize money

A common mistake is judging a tournament's importance by its purse. Prize money is the easiest metric to read and the most misleading.

Professional golf runs in four tiers: the four majors — the Masters, the PGA Championship, the U.S. Open and The Open; the PGA Tour's signature events; regular season events; and team events, whose scoring structure and selection logic differ entirely.

What separates tiers is not money but three other things: world-ranking points allocated, field strength, and historical weight.

Elimination is the key to the transfer market.

I borrow that line from how I read player markets, but it holds in golf. When assessing a young player, what matters is not what he has won but which exemptions he still lacks, and why. Lacking points is a scheduling problem. Lacking ranking is a capability problem. Those demand completely different strategies, and only system-level data distinguishes them.

Risks do not end with injury. I use a seven-row risk matrix: technical, competitive, psychological, injury, career, governance and systemic. The last row is the one I use least and need most, because systemic risk has a published effective date, which means it can be planned for. Psychological risk has no effective date at all.

When the data is empty, the first row of that matrix is analytical risk: the chance that I manufacture an unsupported conclusion and turn it into a headline. That is the most serious risk in this trade, because it leaves no trace. An invented number does not corrupt a data file. It corrupts the reader's trust, and trust has no column to record it in.

---

Layer four: landscape and governance — a split that has not closed

This is the layer where technical data cannot help and where writers most easily slide into declaration.

Governance in world golf this decade has centred on one event: the arrival of a new tour backed by Saudi Arabia's public investment fund, which staged its first event in June 2026, and the framework agreement announced on 6 June 2026 between the established system and the new one. The power structure is still forming, which means every analysis at this layer must be written in the conditional.

With the Japanese market, this layer has a direct consequence: where domestic events sit in the ranking system determines whether a young Japanese player can build a career at home or is forced to move. That is a long-horizon question no single season can answer.

---

Layer five: rules and equipment — changes with effective dates

This is the layer I like most, because it is the only one where answers can be looked up precisely.

Three recent markers matter. First, the anchoring ban, Rule 14-1b, effective 1 January 2026 in the major systems; it changed how a group of older players putted and therefore changed their putting data in ways unrelated to skill. Second, driver regulations adjusted repeatedly through rebound and volume thresholds, shifting driving-distance distribution across the whole system. Third, and most consequential for long-horizon analysis: on 6 December 2026 the two governing bodies announced changes to ball testing conditions, under which the ball will be adjusted to travel shorter, applying to elite competition from January 2028 and to all players from January 2030.

What does NOT happen often tells the truth more loudly than what does.

I apply that rule to every rules change. When a new regulation takes effect, the interesting question is not who is affected but who is not, and why. Short hitters who already rely on approach play will be less affected than those who rely on distance. The second group will have to rebuild their technical profile over two to three seasons. The first group needs to do nothing. The silence of the first group is the information.

---

Layer six: public narrative — heat cycles and expectation gaps

Every player has a public story, and that story has a cycle: breakthrough, celebration, scepticism, judgement. The length of each phase does not depend on data. It depends on how much content the media system needs to produce in that window.

That is why I always separate two things: the real player and the player inside the story. They may share a name and a scorecard, but their technical profiles differ.

When a former world number one such as Scottie Scheffler wins repeatedly, the public narrative moves into celebration and starts explaining the dominance through unmeasurable qualities such as nerve or focus. The data usually points to something more ordinary: fewer strokes lost on approach and putting more stable than baseline. Dominance looks like cumulative effect rather than mysterious virtue.

Gaps in the table can speak too, if we are willing to listen.

In public narrative, the largest gap is failure data. Nobody records the metrics of a week in which a player missed the cut unless a story comes with it. That means the data set the public sees is always filtered toward success — a systematic bias that makes cross-era comparison fragile.

---

Layer seven: industry transmission — from the course to the data

Golf is a long production chain, and a change at one end takes years to reach the other. Upstream sits the course economy, equipment and talent development. Midstream sits tours and event operations. Downstream sits media, sponsorship, data and derivative products.

The fastest-moving row is the data and derivatives segment, and it is the least noticed. Once shot-level data becomes a sellable product, an analyst's value shifts from owning data to contextualising it. Anyone can download a Strokes Gained table. Very few know that a Strokes Gained table without a course-conditions column says almost nothing.

---

Contrarian angle: this industry rewards confidence, not accuracy

The structure of incentives in sports analysis leans one way. A declarative piece gets read more than a conditional one. A firm prediction gets shared more than an analysis with error bars. A confident headline gets clicked more than a headline admitting insufficient data.

So, systemically, analysts are rewarded for reaching conclusions, not for reaching correct ones. Those are different things.

I have described my 2026 and 2026 errors. What I have not said is that both were praised. My model was right four times in ten and I received positive feedback for daring to predict. My 2026 analysis missed a physical variable and was widely shared because it delivered a clear conclusion. Nobody rewarded me for saying the data was insufficient.

That is the real structure of temptation. It sits not in laziness but in the fact that the market pays for certainty, and analysts must live inside that market.

My only personal rule is this: when the table is empty, I write about the empty table. I turn the gap into the subject instead of a place to stuff guesses.

Gegenpressing does not break the data; it breaks my assumptions.

That means: when I borrow analytical language from another sport, I do not borrow conclusions. I borrow questions. And a question may only be asked when there is enough data to answer it. Without data on the sequence of holes following a bogey, I am not permitted to talk about character.

The same applies to my views on closed-loop women's competition structures and on early overuse of young athletes. I do not state them as declarations. I let them surface through which systems I choose to analyse and which missing data I choose to point at.

---

Takeaway: signals for the next round

The hardest question is never what the data says. It is whether I am hearing what the data says, or what I want to hear.

That empty table taught me nothing new about golf. It taught me something old in a new way: an analyst's discipline lies not in how much data he can process, but in how much emptiness he can endure before speaking.

The tournament went ahead. I still reported. I wrote about course conditions, scheduling and field structure, and I noted plainly that technical judgement would follow once shot-level data synced. I lost nothing. I lost only the right to sound knowledgeable for two days.

One signal I will track relates directly to the ball testing change taking effect in 2028 for elite play. Over the next three to four seasons, the technical profiles of young players will begin reflecting that adjustment in advance. Those who rely on distance will have to rebuild their approach game. Those already built around approach play will gain relative advantage by doing nothing.

We will see it in the data before we see it on a leaderboard. The question is whether we will then have a full enough file to read it, or whether we will be sitting in front of another empty table.

Data is never wrong; I simply asked the wrong question. But before asking the question, I need to know what I am holding. And sometimes the only thing to hold is well-timed silence.