Five Platforms, Five xG Values: V.League 2031 and the Empty-Data Crisis

On the night of July 12, 2031, I stayed behind at my Da Nang office after the V.League Round 18 match between Hanoi FC and SHB Da Nang. The score was 2-1 to the visitors. What made me stop was not the result, but what appeared in the two hours after the final whistle. Five leading regional sports data platforms published expected goals (xG) figures for Hanoi FC in turn: 1.82, 2.41, 0.94, 3.10 and 1.63. Five numbers for one match, one goalmouth, twenty-two players. The gap between the highest and lowest value is 3.3 times. The same half of football could be told as 'total domination' or 'a lucky win', depending on which table the reader opens first.

I have spent fifteen years cross-checking player tracking data after every round, and I had never seen a spread this wide in a league I follow directly.

In 2026, when I was the only female reporter in the post-match press room after SHB Da Nang vs Hanoi FC, I asked the coach about his team's xG of 0.4 despite a 1-0 win. A male reporter cut in: what does a woman know about football, she just makes up numbers. I did not argue. I recorded the full tracking data of all 22 players and published a 3,000-word analysis that night, proving the win came from luck, not from control. When the press room laughs at xG, I know I am reading exactly the book they have not opened.

Fourteen years later, that book has been opened. But it has been torn into five different versions, and nobody knows where the original is anymore.

The V.League 2031 context is very different from when I started. Every match is captured by multi-point camera systems, each stadium has at least twelve angles, and player position data is sampled twenty-five times per second. The raw material is abundant, but the number of organisations publishing xG for the league has risen from three to more than twenty. Of those, only seven disclose how many matches their model was trained on, what data it uses, and how it is calibrated per competition.

I kept an internal tracking table through the 2031 season. For each match, I recorded the xG each platform published, then compared it with the raw data I gathered myself from the stadium tracking system. The result: the average gap between the highest and lowest platform was 0.71 xG per match — nearly one goal per match described incorrectly. For PPDA, the pressing-intensity metric (passes allowed per defensive action, lower being more aggressive), the error between platforms reached 4.8 units in a single match.

The worrying part is not that different models give different results — that is inevitable. The worrying part is that most platforms have lost their ability to trace the source, turning a verifiable number into a number that can only be believed.

Previously, when a metric was created, people always knew where it came from: a specific model, a specific dataset, a specific version. In 2031, most metric suppliers no longer publish model information. They publish a bare number with a brand label. One domestic platform I checked published xG for 380 matches in the season, but when I asked about its training set, the answer was 'intellectual property'. Another updated a team's average xGA just eleven minutes after the final whistle — too fast to run any trained model.

I recall a line I once wrote: a single number can lie, but a model verified across 10,000 matches has no reason to pretend. That remains true in 2031, except now we need one more condition: the model must let outsiders see its insides.

Over the past three months, I ran a small experiment. I randomly picked ten V.League matches and compared five platforms against the raw tracking data I processed myself. Only two of the five had a deviation of under 0.2 xG from my hand calculation. The other three differed by over 0.5 xG in at least four of ten matches. One platform had an error of 1.4 xG in a single match — more than the actual xG value of an entire team.

The difference between the trustworthy group and the rest is not technology. All five use stadium cameras and machine-learning models. The difference is that the trustworthy group still lets outsiders review the definition of a chance — what counts as a meaningful shot, which situations are excluded, how own goals are handled. The rest only publish the final total.

The counterintuitive angle here is this: the cause of the 2031 data crisis is not the growing number of automated models, nor the AI content-generation tools. Both are neutral. The real problem is the disappearance of the audit trail — the ability to trace a number back to its origin, which the data-journalism industry once treated as the minimum condition for publication.

When nobody has to prove where their number comes from, the incentive to produce the best number is replaced by the incentive to produce the fastest, cleanest, most shareable number. And in that race, the xG of 3.10 for Hanoi FC travels further than the 0.94 — not because it is more correct, but because it is more dramatic.

That is also why I keep an old habit: before citing any metric, I cross-check at least two independent sources, read the method carefully, and recompute a small sample by hand. I do not do this because I doubt data. I do it because a sports outlet that misquotes an xG may be right commercially, but wrong on the truth.

The solution is not to ban new tools. It lies in a minimum standard: any platform wanting to publish metrics for a professional league must disclose its model version, training-sample size, and how it handles edge cases. Those that do this are not slower. They are just more patient.

The V.League 2031 season is not over. The next round will have at least four live matches, and at least twenty platforms will publish metrics. The question I keep is not which platform is right. The question I keep is whether, by the end of the season, enough numbers will still be brave enough to let the reader see their path — from the goalmouth, through the camera, through the model, into the writer's hand.