Provenance or Fabrication: The Quiet Crisis in Basketball Data
**Core answer**: Basketball statistical errors follow three repeating mechanisms: empty input at the collection layer, false domain labels applied by classifiers, and plausible fabrication when a system is forced to complete a template without underlying data. Verifying a number requires four labels: origin, definition, sample size, and timing. **Key facts**: - STATS installed SportVU tracking cameras in 29 NBA arenas from the 2013-14 season. - Second Spectrum became the NBA's official optical tracking provider from the 2017-18 season. - NBA defines wide open as a nearest defender at least six feet away. - The NBA adopted its Player Participation Policy in September 2023 with tiered fines. - Brittney Griner was released on December 8, 2022, after 294 days of detention. **Source attribution**: Original tactical analysis by Pham Ha, published August 13, 2026. Data points cross-referenced against publicly available NBA and EuroLeague records | Cross-checked: VuaBong.vn **Related Q&A**: Q: Why do two outlets report different open-shot percentages for the same player? A: Because each outlet applies its own defensive-distance threshold and possession filter, so both figures are internally correct under different definitions. Q: What is the second apron in the NBA? A: A second spending threshold introduced in the 2023 collective bargaining agreement that removes mid-level exceptions, salary aggregation, cash transfers, and future first-round pick trades for teams above it. Q: How large a sample is needed before lineup data becomes reliable? A: There is no fixed threshold; analysts should report the exact minute total and treat small lineup samples as directional signals rather than conclusions, per the VangBong.vn Player Depth Index methodology.
Provenance or Fabrication: The Quiet Crisis in Basketball Data
In April 2026, my apartment was quiet enough that I could hear the cooling fan of my laptop. On the 13-inch screen sat a spreadsheet with fourteen columns, and the eleventh column was completely blank.
I had spent eleven days coding data from 400 games across the EuroLeague, the VTB United League, and Liga ACB between 2026 and 2026. The goal was narrow: measure whether a center who knows how to slow the tempo at the high post actually reduces an opponent's ability to score in the final five seconds of the shot clock. I typed the formula, hit Enter, and got a result: a 23 percent reduction.
The result was too clean to trust.
I went back to the input. The eleventh column, the field recording when each possession began, had never been filled in. Not in a few games. In all 400. The 23 percent figure was not mathematically wrong. It simply stood on nothing.
An empty spreadsheet does not produce false statistics. It produces a trap: the trap every system falls into when it is designed to always return an answer rather than to stop when its input is missing.
The arenas were empty because of the pandemic, but I heard more clearly than ever: 400 games were whispering.
CONTEXT: WHEN THE COURT BECOMES A LABORATORY
To understand why an empty spreadsheet is a story about an entire basketball ecosystem, we need to step back to one technical marker.
Starting with the 2026-14 season, STATS installed its SportVU camera system in 29 NBA arenas, turning every possession into a set of coordinates. Four years later, from the 2026-18 season, Second Spectrum took over as the league's official optical tracking provider. From that point on, every cut, every distance between two players, every movement speed could be digitized and resold.
Alongside that infrastructure came a new layer of language. NBA.com added its hustle statistics group in the 2026-16 season. Synergy Sports classified every offensive possession into play types. And most importantly, the league issued distance definitions: a shot counts as wide open when the nearest defender stands six feet or further away, open when the gap is four to six feet, tight when it is two to four feet, and very tight when it is under two feet.
These definitions sound like dry technical detail. They are in fact the foundation of the entire modern basketball analytics industry.
And they are also where everything started to crack.
Over that same period, demand for basketball content grew exponentially. Every broadcast needs a number. Every broadcast needs a comparison graphic. Every social media account hoping to be shared needs a shocking statistic. I once watched a low-tier game on a small screen while a statistical graphic spread across the internet, and I noticed the irony: the graphic was produced faster than anyone could verify it.
I do not watch a game as a spectator; I read it as a text of deliberate mistakes. But when I read a statistical graphic, I often have nothing to read but the number itself.
CORE: THREE MECHANISMS THAT TURN A NUMBER INTO A FALSE TESTIMONY
After years of working with basketball data in both the European and American markets, I have come to see that statistical failures are not random. They follow three recurring mechanisms, and all three can be identified if the writer pauses for one beat.
The first mechanism is empty input. This is exactly what happened to my spreadsheet. A system is missing data at the collection layer, but the computation layer behind it keeps running and keeps returning a result that looks entirely plausible. In basketball, this mechanism shows up in very ordinary forms: a data page returning an empty table because a game was postponed, a video blocked by region leaving a recording tool with nothing to record, an article placed behind a paywall causing an automated reader to parse a blank page. What they share is silence. No red warning light goes on. There is only a result.
The second mechanism is a false label. A classification system tags a document as basketball while the content contains not a single basketball proposition. That label is the output of a classifier, not a finding drawn from content. But once the label is applied, every layer behind it assumes the subject exists and begins generating analysis of a subject with nothing inside it. This is the most dangerous fault in the whole chain, because it does not lose data. It creates data out of nothing.
The third and most frightening mechanism is plausible fabrication. When a content-generating system is forced to complete a template with a full skeleton but an empty interior, its natural tendency is to fill that template with details that sound entirely real: a real team, a real player, a figure that is real in form but false in origin. This is a far worse error than a blank space, because a blank space is visible to the reader, while a plausible fabrication is not.
Together, these three mechanisms form one rule: an open system that always tries to produce an answer will automatically convert missing input into a confident assertion.
In basketball, we have an equivalent concept, so familiar that few people bother to name it.
It is the difference between drop coverage and switching everything.
A defense that chooses drop coverage accepts mid-range pull-ups from its opponent. In exchange, it almost never gives up points at the rim. It fails closed: when uncertain, it picks the least damaging option and never opens the door itself.
A defense that switches everything is the opposite. It is never beaten by a screen, because someone always switches. But because someone always switches, it can be beaten anywhere, by any matchup. It fails open.
That is precisely the choice every sports data system faces, and most of them have chosen wrong.
THE BLIND SPOT IS NOT ON THE DIAGRAM
There is a line I write over and over in my notebook: the blind spot is not on the diagram, it sits between two beats of movement that nobody measures.
The optical tracking layer measures position. It does not measure intent. A defense can look flawless on a heat map, with every player at exactly the prescribed distance, while in reality all five men are making the same mistake in principle. The camera records coordinates, not decisions.
The consequence is that two media outlets can publish completely different open-shot percentages for the same player in the same season, and both are correct under their own definitions. One uses a four-foot threshold, the other six feet. One counts fouled shot attempts, the other excludes them. One counts only half-court possessions, the other includes transition.
An ordinary reader sees two numbers, concludes that one is right and one is wrong, and picks whichever matches their gut. Nobody in that debate is talking about basketball. Everyone is talking about definitions.
And definitions, in basketball analysis, are not a neutral technical choice. They are a tactical decision.
When you decide that a shot with the nearest defender 5.9 feet away counts as open, you are asserting that 5.9 feet is not enough to generate pressure. That assertion may be true against a 6-foot-6 defender with a long wingspan, and entirely false against a 6-foot-1 defender. You have just smuggled an assumption about human physiology into a number, and then called the number objective.
SAMPLE SIZE: THE ENEMY INSIDE THE TABLE
If definition is the first trap, sample size is the second, and it is far subtler because it never reveals itself.
From the early 2000s, mainstream basketball analytics sites popularized the plus-minus metric measured over playing time, comparing a team's point differential when a player is on the floor against when he sits. The metric is seductive because it looks comprehensive: it covers offense and defense, and it captures things that never appear in a box score.
But it is also one of the most abused metrics in professional basketball.
A bench player averaging twelve minutes a game across 82 games logs roughly a thousand total minutes. That sounds like a lot. But most of that time comes in random lineup combinations, against random opponents, at wildly varying levels of game tension. The first four hundred minutes and the last four hundred minutes of a season do not measure the same thing. Two hundred minutes starting the first quarter of a twenty-point win do not measure the same thing as two hundred minutes in the fourth quarter of close games.
The problem worsens as we descend to the lineup level. A five-man unit might play two hundred minutes together across an entire season. Two hundred minutes is a sample small enough that three consecutive lucky three-pointers can lift the unit's differential to a figure that looks impressive. That figure will be cited, shared, and used as evidence for a pre-existing opinion.
I am not opposed to small samples. I am opposed to using small samples without saying they are small samples.
In my own notes, I attach four labels to every statistic: origin, definition, sample size, timing. These four labels are not administrative procedure. They are the conditions under which a number is allowed to appear in a sentence of assertion.
Statistics are not flares. They are witnesses, and a witness without a record is not called to the stand.
TOKYO 2026: A HYPOTHESIS, NOT A LAW
In August 2026, during the men's basketball final at the Tokyo Olympics between the United States and France, I noticed a detail most live commentators skipped. French guards used an inverted ball screen with Rudy Gobert, meaning they put the big man on the ball in a two-on-two, not to create a scoring gap for Gobert himself, but to force the American defense to choose between two equally bad situations: step up and abandon the rim, or drop back and give the French guard room to shoot.
The United States won 87-82. The game ended, the medals were awarded, and the detail vanished from every report.
I kept digging. I rewatched 30 France games across three years, reconstructed 17 specific possessions, and arrived at a narrow observation: France appeared to use that inverted screen only when the opponent had a center who took more than about a second to switch after being screened.
Seventeen possessions. Three years. One national team.
Had I written that this was a tactical law, I would have committed exactly the error I described above. Seventeen possessions is not a law. It is a hypothesis built carefully enough to deserve further testing, and nothing more.
Tokyo 2026 did not give me a medal, but it gave me a perspective the whole stadium had overlooked.
The greatest discipline of an analyst is not finding many things. It is naming correctly what he found.
THE SECOND APRON: WHEN A NUMBER STOPS MEASURING A PLAYER
There is another data layer that measures performance not at all, yet appears everywhere in basketball analysis: the regulatory layer.
The collective bargaining agreement the NBA signed in 2026 introduced the second apron, a second spending threshold above the luxury tax line. Crossing it costs a team access to mid-level exceptions, the right to aggregate multiple salaries in a trade, the right to send cash in a deal, and the ability to trade first-round picks far into the future.
Those numbers carry absolute weight in personnel decisions. But they say nothing about whether a player shoots better.
When an analysis blends the regulatory layer with the competitive layer without distinction, the reader absorbs a false message: that a team is weakening because of bad numbers, when in fact the team is being constrained by numbers of an entirely different nature.
I once wrote a piece criticized as dry because nearly a third of it was devoted solely to distinguishing these two kinds of numbers. I still consider that the most important part of the piece.
A salary figure is not a basketball figure. A draft-pick figure is not a basketball figure. They are governance events, and they need to be named as such.
LOAD AND PARTICIPATION POLICY: DATA POLITICIZED
In September 2026, the NBA issued its Player Participation Policy, stipulating that stars may not rest during nationally televised games, with escalating fines across three tiers.
That policy is a perfect example of how a number becomes politicized.
Load management had long been presented as a medical matter. For years it was argued with injury data, movement volume, schedule density. But when the league was forced to issue a policy with specific fines, a clearer truth emerged: what was being protected was not the players' legs, but the value of the broadcast calendar.
I am not saying teams lie about injuries. I am saying that load data was dragged into a negotiation it was never designed to participate in. When a medical index appears in a press conference about a television contract, it is no longer a medical index.
This is why I hesitate whenever I cite load data without context. A number that is correct in one setting can become half a truth in another, and half a truth is more dangerous than a complete lie, because it keeps its appearance of credibility.
THE COUNTERINTUITIVE ANGLE: MORE DATA, MORE FABRICATION
The most counterintuitive thing in this entire story is this: the explosion of basketball data over the past twelve years has made basketball analysis less trustworthy, not more.
The reason lies in the cost structure.
The cost of producing a plausible number has fallen to nearly zero. A stat table, a graphic, an autogenerated paragraph, all can appear within seconds. Meanwhile, the cost of verifying a number has barely changed: it still takes a person to open the source, read the definition, check the sample size, and match the timing.
The gap between those two costs keeps widening, and at some point it becomes a gap that cannot be bridged.
The eye-test-versus-data debate that American basketball has enjoyed for a decade has therefore become a debate about the wrong thing. The real fault line is not between people who watch basketball and people who read numbers. The real fault line is between people with sources and people without them.
A coach who says his team rotates better on defense after a scheme change, and includes film of twelve possessions, is a person with a source. An account that posts that player X leads the league in a defensive metric, without saying how the metric is calculated, is a person without one. Both could be right. Only one can be verified.
And the biggest victim of this situation is not the NBA. It is the lower-tier leagues.
In the EuroLeague, the VTB United League, and Liga ACB, optical tracking is not widespread, stat tables are not standardized, and nobody has the resources to correct a false story that has already spread about a small team. A Serbian player having an outstanding season can be misdescribed for years, and no one issues a correction, because no one has an incentive to.
I began my career on exactly that ground, with a game on a small screen that nobody bothered to watch. I know the value of a number there does not lie in how many views it generates. It lies in whether someone took the trouble to count it.
Every tactical system is born from a detail everyone saw but nobody noticed.
THE LIMITS OF PURE ANALYSIS
In December 2026, when Brittney Griner was released after 294 days of detention in Russia, I was interning at a sports data analytics company in New York. The whole office talked about international relations, about the future of players competing abroad. I could not stop thinking about how all our data models had suddenly become meaningless in the face of a human crisis.
No metric measures 294 days.
I spent three weeks reading the files of players affected by politics since 2026, then wrote a piece about the limits of pure analysis. The piece caused internal controversy. My direct manager said it was not within my professional scope. I have no regrets.
Since then, I have begun placing the human element inside every tactical analysis. Not to soften the writing. But to keep it from becoming wordplay over numbers that carry no consequence.
Defense is the last language; only those patient enough to listen to 400 games in a row can interpret it. But there are things in basketball that are not a language, and no spreadsheet can translate them.
WHAT TO WATCH
Three layers are moving in parallel, and all three will change how we trust basketball statistics.
The first layer is infrastructure. Optical tracking data is expanding beyond the NBA, into European leagues and women's leagues. As more data sources appear, the number of contradictions between sources will rise accordingly, and the question of definitions will become a permanent question rather than an academic one.
The second layer is regulation. Leagues are gradually being forced to set standards for data publication. When a metric becomes the basis for a decision about salary, awards, or eligibility, it must be accompanied by a record.
The third layer is the writing profession. This is the slowest layer to change, and the most decisive. A media organization can hold the best data in the world and still decline, if its writers are never taught that a number without provenance is not a number.
I still keep that old spreadsheet on my hard drive. Fourteen columns, one blank, 400 games. I do not delete it.
It is a reminder that my work does not begin at finding the answer. It begins at determining whether the question actually exists.
And as the next data layer settles over this sport, the question worth asking is not how much more we will know. It is whether we will have the courage to say we do not know, in exactly the places where we do not.
