When the ShotLink Table Goes Blank: A Night in Nagoya and the Lesson of Data Silence

At three in the morning in Nagoya, I reopened a tournament round's Strokes Gained table and found the SG: Approach column blank in 41 of 78 rows. For the first ten minutes I blamed the connection. Twenty minutes later I blamed the data provider. Forty minutes later I opened the log file and found no error at all: the system had run correctly; the event simply had no shot-level tracking layer. The cause was not in the pipeline. It was in the fact that no device had ever been installed on that course.

That night I understood something I had not been humble enough to accept at 24, when I built my own xG model for Nagoya Grampus: a blank cell is not a zero. A blank means unknown. A zero means known — and known to sit exactly at the tournament average. Those two get mixed together in a great many golf analyses I read every week, and the cost of that confusion is higher than a three-putt on the 18th.

Context: what a round of golf is measured with

In 2026, Mark Broadie — a Columbia University professor — published the Strokes Gained method, and the way a round of golf is read changed at the root. Instead of counting putts or fairways hit, you measure the value of each shot against the expected baseline of the whole field. That measurement needs three things: ball position, distance to the hole, and a baseline expectation built from hundreds of thousands of comparable shots. A scorecard stores none of them.

The infrastructure layer that supplies that data on the PGA Tour is called ShotLink, in operation since the early 2000s, with measuring devices at every hole recording every shot. Japan has its own system, but it does not cover the entire schedule. In Vietnam, most golf stories are still recorded on scorecards — meaning only the final result exists, not the path of the shot. That asymmetry is itself a piece of data: it tells you which questions can be answered in which market, and which questions have to end with the word unknown.

The world ranking, in operation since 2026, is an instructive counter-example. There, a missed cut is a legitimate zero: the player was present, competed, and failed to pass the cut line. The system records exactly that. But if an event is postponed by storms and no round counts, a score of zero would be a perfect lie. The same number, two different meanings, and the reader of the ranking is given no way to tell them apart.

Based on my experience watching rounds, every table I publish has to carry its data source and margin of error in the first row. That discipline came from another stumble: in 2026 my homemade xG model missed the home-venue factor and was wrong in 6 of the last 10 matchdays. The data is never wrong; I was asking the wrong question.

Anatomy of a blank cell

A blank cell in a golf data table comes from one of three sources, and each demands a very different response.

The pipeline blank is the easiest to spot: the data exists but never reached me. The log file will show a missing timestamp or a connection error code. The correct response is to re-run the process, not to write an article.

The classification blank is far more dangerous: the data arrived but was filed in the wrong place. A round played under preferred lies because the course was waterlogged cannot produce a trustworthy approach statistic. If the system files it in the same column as ordinary rounds, I am comparing two things of different natures. This kind of blank leaves no empty cell on the screen. The table still looks complete, and that is precisely the problem.

The third kind is what I met that night in Nagoya: the measurement system never existed at that event. No device, no recorder, no data — and there never will be, however long I wait. This is where elimination earns its keep. With no shot-level data I remove Strokes Gained from every conclusion; I also remove putts per green in regulation, because it does not control for first-putt distance; and I keep only what the scorecard genuinely records: greens in regulation, putt counts, scrambling rate. Those metrics are weaker, but they are honestly weaker. When the data hides its face, the error term becomes the guide.

I built a three-question test to check myself before writing any number. Did this event have shot-level measurement? The answer may only be yes, no, or undetermined. Did the pipeline run during the competition window? The log file answers that, not my intuition. Is this field applicable to the event format? Average putting is meaningless in four-ball match play. If any answer lands on undetermined, the output must be undetermined — rather than an estimated value presented as a measurement.

One case sticks with me. After Hideki Matsuyama won the 2026 Masters, becoming the first Japanese man to win a major, the volume of Japanese-language golf tables I had to cross-check each week rose sharply. Most of them shared a single origin, so they did not verify one another; they merely repeated one another. I had to trace each back to its source file to learn which cell was a real measurement and which was a number copied three times until it looked like three independent sources. Gaps in a table can speak, if we are willing to listen. The trouble is that most tables no longer have any gaps left to listen to, because someone filled them with a value that looks perfectly reasonable.

I once tried to transplant football's pressing language into golf. The idea was seductive: measure pressing intensity in fifteen-minute blocks, and find which golfer holds his attacking rhythm after a bogey. Six weeks later I abandoned it. Golf has no data column corresponding to ball recoveries; the intervals between holes vary in length, and heart rate does not measure pressure on an opponent. The translation failed, and it failed because of the data, not because of my patience. Gegenpressing does not break the data; it breaks my assumptions.

The counter-intuitive angle

The reflex of the analytics industry is to fill the gap. Missing data gets interpolated, a thin sample gets a model, a missing model gets a composite index with weights nobody re-checks. The danger is that an interpolated value and a measured value have the same shape on screen. The reader has no way to tell them apart. After a few rounds of this, the original blank disappears from history, and an assumption becomes a fact.

What did NOT happen often tells the truth more plainly than what did.

The Vietnam–Japan comparison is useful here, but only up to a point. In Japan the gap is mostly technical: the event was not instrumented, though the infrastructure exists. In Vietnam the gap is mostly structural: no measurement layer has been built at all, so there is nothing that could lose signal. The two gaps differ in nature, but if both are collapsed into a column holding the value 0, the resulting error is identical. I keep this comparison only when the divergence between the two sides is large enough to change how a result should be read; on other occasions it is merely a pretty and useless cultural story.

What remains

The most honest column in my spreadsheet is labelled undetermined, and it is longer than I would like. Every number is a confession not yet written down.

Next week, when a Strokes Gained table appears in front of you, read the blank cells before you read the numbers. What matters is not who struck it better, but what we are actually measuring — and what we are pretending to measure.