When the Data Table Returns Zero: Vietnam's Football Analytics Infrastructure

In January 2026, I sat in a small apartment in Osaka and opened three statistical platforms at once to cross-check the event data of all 26 group-stage matches at the 2026 ASEAN Cup. The goal was specific: reconstruct the pressing curve of Vietnam's national team under head coach Kim Sang-sik and compare it with the Philippe Troussier period.

The result was not what I had pictured.

The first platform returned complete event data for 17 of the 26 matches. The remaining nine offered only the scoreline, the line-ups and the list of cards. The second and third platforms covered all 26 matches, but the number of shots they attributed to Vietnam in the same match differed by four. Same match. Same nominal definition of "a shot."

It took me two days to trace the cause. The cause was not football. It was record-keeping.

"On that night in Russia in 2026, I watched the data shatter in front of me." I was seventeen then, sitting in front of a screen, logging every phase of play for Japan's national team, convinced that if I transcribed enough, the spreadsheet would speak the truth on its own. Seven years later, at a Southeast Asian tournament, I ran into the same phenomenon. This time I was not surprised, but I understood something else: Vietnam's problem in the data story has never been a shortage of numbers. The problem is that we cannot tell a zero apart from a blank.

V.League 1 currently has 14 clubs playing 26 rounds, roughly 182 matches per season. The J1 League I follow every week has 20 teams across 38 rounds, roughly 380 matches. Put the two figures side by side and the reflex of most analysts is immediate: fewer matches, therefore less data, therefore harder to analyse. That conclusion is not wrong. It simply puts the emphasis in the wrong place.

What determines analytical quality is not the total number of matches but the density of recorded events inside each one. A match in the Premier League or the Bundesliga can generate three to five thousand labelled events: every pass with its origin and destination coordinates, every duel with its outcome, every off-ball run by a full-back, every freeze frame capturing the position of all twenty-two players. For most public platforms covering the V.League, that figure usually sits between two hundred and four hundred events per match, and almost all of them are on-ball events: goals, shots, key passes, fouls, cards.

The gap is not that one league is inferior to another. It sits in the collection infrastructure. Positional data requires calibrated multi-angle cameras, a semi-automated tracking system, and a labelling team working to the same code all season. Without those three things, every platform is forced to infer, and each platform infers differently. That is why two providers produce two different shot counts for the same match.

In Japan, where I live and work, the J.League publishes event data to a unified standard, with positional data for the majority of matches. An independent analyst like me can download it, cross-check it, and reproduce someone else's results. In Vietnam, that act of reproduction is practically impossible at the public level.

The national team adds its own constraint. The second round of Asian qualifying for the 2026 World Cup gave Vietnam only six matches, closing in June 2026 with a third-place finish in Group F behind Iraq and Indonesia. Six matches, plus a handful of friendlies in training camps. That is the entire official match database for evaluating a two-year cycle under one head coach. Philippe Troussier left the job in March 2026. Kim Sang-sik took over in May 2026 and led the team to the ASEAN Cup title in January 2026. Those three milestones are clearly recorded. The chain of evidence between them is alarmingly thin.

"An empty stadium, yet the numbers are still full of noise." I first wrote that line in 2026, when the pandemic suspended the J.League for four months and I had to rebuild Cerezo Osaka's dataset from old footage. Back then I thought the noise was a temporary consequence of playing without crowds. Later I understood: the noise exists even when the stands are full. Refereeing pressure, public expectation, pitch quality, a congested schedule — all of them are variables, and all of them are pushed out of the spreadsheet because nobody measures them.

VAR arrived in the V.League in 2026. It was a step forward for fairness and, at the same time, a new source of noise: the number of interventions shifts from one refereeing crew to the next, and no body publishes that data to a continuous standard. When I tried to put an "VAR intervention" variable into my model, I could not find a source coherent enough. I dropped it, which means my model is blind to a factor capable of changing a match result.

The core of the problem needs a concrete chain of evidence. I walk through six links.

Link one: what we measure. In a typical V.League match, most public data revolves around the ball. Where it is, who touches it, how often, where it goes. A player only enters the data at the instant he touches the ball. The consequence is that the behaviours that decide matches without touching the ball — movement that stretches the opponent's shape, runs that open space, dropping deep to cover when a teammate advances — remain entirely invisible. In other words, we analyse a football match using only its visible surface.

Link two: error compounds along the chain. Suppose each match captures only twenty per cent of the total actions that actually occurred. To calculate a pressing metric, I add up duels in the opponent's half and divide by the number of passes the opponent completed before being closed down. Both the numerator and the denominator come from an already filtered dataset. The error does not cancel out. It accumulates. And when I place Vietnam's pressing metric at the 2026 ASEAN Cup beside the pressing metric of a J1 League side, I am comparing two datasets collected to two different codes and assigning them the same meaning. That is a methodological fault, not a data fault.

Link three: the small-sample problem of a national team. Six qualifiers spread across four months against three stylistically different opponents. If I use six data points to conclude something about a tactical system, I am doing what no laboratory would accept. At club level a team plays thirty-eight matches a season and the error has a chance to flatten out. At Southeast Asian national-team level, that does not happen. Each match is an almost unrepeatable observation, shaped by squad availability, physical condition and flight schedules.

Link four: the transfer market and the trap of belief. "The contract is only the ending; the beginning sits in the spreadsheet." For years I have tracked the cases of Vietnamese players moving abroad. Nguyen Cong Phuong wore the shirt of Mito HollyHock in J2, then Sint-Truiden in Belgium, then Incheon United in the K League. Nguyen Quang Hai played for Pau FC in Ligue 2 across 2026-2026. Doan Van Hau joined SC Heerenveen in 2026 but never made a competitive Eredivisie appearance. Nguyen Tuan Anh spent time at Yokohama FC in J2.

Looking back at all four, the central question still has no answer: which V.League metric predicts the ability to adapt abroad? In 2026 I tried to answer it with a small model. I gathered data on more than two hundred players moving from Southeast Asian leagues to Europe and looked for a correlation between distance covered per match and the share of minutes played in the new league. The correlation coefficient I obtained was 0.67. It sounded convincing. But when I audited the distance data, I found that most figures came from three different providers using three different methods, including matches estimated purely from camera position. I had to downgrade the model's confidence severely.

Link five: natural experiments left unused. The 2026 ASEAN Cup title created a rare natural experiment. Nguyen Xuan Son joined the national team, scored repeatedly, then suffered a serious injury in the second leg of the final. The sequence of matches before and after that moment is an almost perfect control pair for measuring how dependent the team is on one individual, and how the attacking structure changes in his absence. I tried to reconstruct that comparison. The problem is that too few matches qualify, and the sample is muddied by different opponents in each round. The result lacked the statistical power to conclude anything. I logged it as an open hypothesis, not a finding.

Link six: the same disease on a different track. In athletics, the data looks more complete because time is machine-measured. But a 100-metre performance only means something alongside wind speed, temperature, altitude and track condition. I once cross-checked the results of several domestic meets and found that a good number of published result sheets were missing the wind-speed column entirely. Without that column, a 10.5-second run might be genuine progress or might simply be a tailwind. One gap, two opposite readings.

"I collect mistakes, classify them, and then I know where a team is heading." The way I work now is the inverse of the way I worked at seventeen. I do not begin by hunting for the number that supports my hypothesis. I begin by listing the data fields that are still empty.

The familiar conclusion at this point would be: Vietnamese football lacks data and needs more investment. I want to set a counter-argument beside it, because that conclusion is right on the surface and wrong at depth.

Buying more cameras, hiring an international provider, installing tracking systems — all of it helps. But if the recording process does not change, we will only produce more data carrying exactly the same defect. The core issue is that we are confusing a zero with a null.

A zero is a measurement: the system observed and confirmed that nothing happened. A null is a silence: the system did not observe, or observed without recording. In a spreadsheet, the two are often represented by the same character. In a reader's mind, they become entirely identical.

When a player has no duel metric anywhere in the dataset, very few people ask whether he duelled without it being recorded, or genuinely never duelled at all. Most will assume the latter. From there a chain of reasoning is built: the player avoids contact, lacks fighting spirit, does not suit a high-tempo game. That chain sounds entirely reasonable, and it rests wholly on a blank that was never confirmed.

Most tactical conclusions about Vietnamese football are built on empty fields misread as zeros. This does not happen because someone deliberately lies. It happens because nobody is taught to record their own ignorance.

It took me a year to learn to write "not measured" instead of writing "zero." In 2026, when the pandemic kept me away from Yodoko Sakura stadium, I rebuilt Cerezo Osaka's 2026 pressing dataset from video, logging 1,240 situations. I predicted Cerezo would decline once the league resumed, having lost their home advantage. They finished fourth, while my model predicted second. I was wrong, and I logged that error as a field of its own: the "crowd effect" variable, not yet measured.

There is a defence of the opposite position that I still remind myself of whenever I write. If we waited for perfect data before analysing anything, no analysis would ever be written, and fans would still have to make judgements with nothing in hand. Football runs on a calendar; it does not wait for infrastructure. Being forced to conclude from incomplete data is the default working condition, not the exception. What separates a serious analyst from a loud one is not whether the data is complete, but whether they state their level of certainty.

"Data does not create stories; it strips the stories of others bare." An honest model is one that says clearly what it does not know.

The signal I will watch through the coming season is not the number of goals but whether a Vietnamese league publishes an event-data repository to an open standard. If it does, independent analysts will begin to reproduce each other's results, and the argument will shift from who speaks better to which method is right.

If it does not, we will keep watching football, and we will keep misreading it with great confidence.