The Empty Record: When Basketball Data Disappears and Nobody Checks

A VBA stat sheet sat on my desk with one line consisting of four zeros. A bench player logged 9 minutes, no points, no rebounds, no assists, no fouls. Four zeros lined up neatly, looking like a completed report. I opened the raw file from the arena tracking system to cross-check, and that line held no data at all. Not zero. Empty.

The system lost its connection for six minutes in the third quarter. The stat crew that night typed 0 into a cell that procedure said should stay blank, because a white cell in a spreadsheet looks more like a formatting error than an event. The next morning, the whole meeting room read those four zeros as a verdict on a player's ability. Nobody asked what the system had lost.

Since that morning, I have understood something about my trade: the most dangerous error in basketball analysis is not a value that was miscalculated. It is a value that never existed but was printed anyway with a credible face.

Context: more data, less checking

Basketball has moved from handwritten ledgers to coordinate-tagged tracking systems. The NBA installed SportVU in every arena starting with the 2026-14 season, then shifted to Second Spectrum. Every possession is logged with player positions; every shot carries an angle, a distance, a release speed. The VBA also runs live statistical software with staff keying in each play. The static structure of a game, its pace, its gaps, its movement habits, all of that only becomes visible when you drop the crowd noise and look at how the data was collected.

The common belief is that more data means fewer mistakes. Twenty-two years of watching this industry show me the opposite in one very specific layer. As collection density rises, error does not disappear. It changes floors.

A basketball statistics workflow has three floors. The bottom floor is raw data. The middle floor is the collection note: which game, which arena, with or without a crowd, whether the system ran in full. The top floor is the summary sheet handed to the coaching staff. Almost all resources go to the top floor, where people present. Almost nobody spends time on the middle floor, where people record the circumstances under which the data was taken.

The result is reports that look solid. A sheet can list twelve games, all twelve flagged as processed, and three of them missing one quarter of tracking. Technically, coverage is 100%. Informationaly, those three quarters never existed.

Core analysis: zero and blank are two different truths

In a basketball box score, the two lines below can look identical.

Player A: 9 minutes, 0 points, 0 rebounds, 0 assists. Player B: 0 minutes, 0 points, 0 rebounds, 0 assists.

Player A's line is a fact. It says a person stood on the floor for nine minutes and produced no measurable impact, and that carries tactical meaning: perhaps he was placed in the wrong situation, perhaps he was only there to set screens, perhaps the opposing defence had sealed every passing lane. Player B's line says nothing. It only says someone did not play, and most systems append a DNP tag with a reason. That tag is the information.

Strip the tag and keep the four zeros, and the two lines merge into one. From there, every average carries a small error, and that small error does not report itself. It sits quietly inside the total until the end of the season, when it becomes a wrong conclusion about a human being.

My handling of this is almost mechanical. When I receive a player efficiency figure, the first question is not whether it is high or low. The first question is how many minutes the sample covers, at what stage of the game, and whether the logging system was complete. For team data, the question is which games had tracking, which were keyed by hand, and whether those two sources were poured into the same chart.

Take an example one floor up. In the 2026-16 season, Stephen Curry made 402 three-pointers, a league record. Place that value beside a 45.4% success rate on roughly 11 attempts per game, and place it inside Golden State's offensive system, where he moved off the ball to stretch the defence before catching, and it becomes a fact with meaning. Remove all context and keep only the number 402, and you have an empty legend: textually correct, tactically useless.

The same principle applies on a smaller floor. When the arena is empty, I start hearing the sound of the game. In 2026, the NBA returned in Orlando without crowds, and home court was little more than a label on the scoreboard. A season without spectators is also a season with its own data. Anyone who compares that period directly with a crowd-filled season without tagging the collection conditions is mixing two different samples onto one vertical axis.

Also in 2026, with leagues suspended, I spent eight months re-watching VBA 2026-2026 games and building a small dataset on free-throw performance at home and away. The only anomaly I found: under a hypothetical no-crowd condition, free-throw rates for one group of players under 23 rose by 7 to 9%. I wrote a long report, self-published it on a personal blog, and sent it to four VBA head coaches. Nobody replied. Three months later, one called to ask about my method for calculating a psychological stability index.

What I emphasised in that report was not the size of the increase. It was three lines at the top of the document: unverified assumption, small sample, margin of error. Those eight months taught me that the hardest part of basketball analysis is not finding a pattern. It is writing down the conditions under which the pattern was taken.

In 2026, on my first night in the tactical commentary seat at the Quan Khu 5 arena, I pointed out that Danang Dragons were defending the pick-and-roll against principle, letting Saigon Heat score 11 straight points in the second quarter. A spectator messaged in, essentially asking what a woman knows about zone defence. I did not argue. I rewound the tape, counted four times Heat repeated the exact same attack from the right wing, and built the movement chart for every player. Nobody asks me whether I understand basketball anymore, because data has no gender.

But if those four possessions had fallen inside the six minutes the tracking system lost, I would have had nothing to count. That is why I check the source before I check the conclusion.

There is a class of error I call cascade failure. It begins in data collection, spreads to tagging, then spreads to conclusion. Symptoms show up in three places but the cause is one. When the summary sheet reports that every game has data, nobody on the coaching staff walks back to check the arena cameras. They use the result. And the result was already clean before it was ever correct.

What stands out is that cascade failure does not produce suspicious-looking reports. It produces normal-looking ones. A record labelled as a game but holding no content still counts as covered in every summary. It only contributes zero to every sum, and nobody chases a zero.

The counterintuitive angle: filling a blank is the politest way to lie

The natural instinct on seeing a white cell is to fill it. In statistics this is called imputation, and in many industries it is a legitimate technique. In basketball, filling a missing game with a season average is lying with a straight face. You repaint the picture in exactly the colour you want, then hand it to a coach as a photograph of the scene.

First counterintuitive point: adding data does not reduce error. It pushes error from the visible floor to the hidden one. A team with tracking cameras in every game will feel more confident than a team with only handwritten ledgers, even when their underlying error rates are identical. The confidence rises not because understanding rose, but because the cells look fuller.

Second counterintuitive point: the most certain-sounding reports are usually built from the emptiest inputs. A record with no content can still generate three pages of conclusions, if the writer is willing to lean on general basketball memory instead of the game in front of them. Readers do not see the missing part. They only see the fluent part.

Emotion is the reporter, data is the referee. But referees also miss games. And when the referee is absent, the score is still published, still discussed, still filed away.

There is a gap I always have to remind myself about: accurate data and an accurate game are two different things. A value can be precise to the decimal and still lead to a wrong conclusion, if it was collected under conditions other than the ones you assume. A free-throw rate measured in an empty arena cannot sit on the same chart as a free-throw rate measured in a packed one, unless you label it. Remove the label, and the sheet still looks clean, while the conclusion has long since drifted off the ground.

What to watch in the next game

Analysis is not about proving I am right, it is about letting the game speak. For the game to speak, the microphone must first not cut out midway.

So now, whenever I receive a statistical sheet, the first thing I do is not read the numbers. I ask the stat crew three questions: how many minutes the system lost that night, which cells should have stayed blank, and which games in this sheet were keyed by hand. Those three answers decide how far I am allowed to conclude.

In basketball, the final shot is decided forty minutes earlier. And the final test of a dataset is not the part it displays, but the part it dares to admit it lost.