Empty Data and the NBA's 2026 Summer Player Valuation Problem

In July 2026, in an office in Manhattan, the front office of an NBA team opened the tracking sheet of a 25-year-old guard. On it: 16 playoff games, a 42.3 percent three-point rate, and 21.4 points per game. A handwritten note in blue ink: “All-Star potential.” No one in the room asked a single question: does this 16-game sample carry any statistical meaning? I have spent most of my career asking that question. And the answer, in most cases, is no. I reopened my personal tracking file — more than 4,800 possessions hand-coded across three seasons. And I found what the official box score never shows: 68 percent of the successful threes in that 16-game run came from situations with no defender within 1.8 meters, compared to 39 percent during the regular season. That is not skill. That is context. Every number I touch carries a scar. And the scar here is a nine-figure contract built on an empty data shell. That summer was empty, but data never rests. The summer of 2026 is not the first time I have seen this. It is simply the summer when the gap between “the number” and “the meaning” became clearest. During the transfer window, the market runs on three currencies. The first is real money: cap space, release clauses, contract structure, seasonal payments. The second is phantom money: rumors, leaks, tweets posted at two in the morning. The third — and this is the most dangerous — is memory money: what a player once did in a moment that television replays over and over. The third currency never appears on a balance sheet, but it prices contracts. It has no verifiable source, no confidence interval, no sample size. It has only emotion, packaged as a number. I once wrote that basketball is never empty; only our way of looking at it is empty. But there is one exception: when we deliberately fill the void with a number that does not exist, we are no longer analyzing. We are fabricating. To understand why this repeats every summer, we need to return to a natural experiment the league accidentally created. In 2026, when the NBA built the “bubble” in Orlando, a variable that is permanently present in every model suddenly disappeared: the crowd. No stands, no roar, no home-court advantage in the traditional sense. The team listed as “home” was simply the higher seed. In many prior seasons, home-court advantage in the NBA was recorded at roughly a 60 percent win rate for the home team. In the bubble, that number nearly evaporated. I tracked every possession of the 2026 playoff run. What I learned was not that “crowds matter” — everyone knows that. What I learned was this: when one variable is removed, other variables suddenly surface. With no roar, referee decisions became more consistent. With no crowd pressure, young players shot with more confidence. And most importantly: the numbers we assumed were “skill” turned out to be largely “context.” That is the lesson I carried into the summer of 2026. When a team values a player based on 16 playoff games, it is doing the opposite of the bubble experiment. It is adding a variable that does not exist — “the moment” — into a model that requires stability. Let us do a simple calculation. Suppose a player’s true three-point success probability is 36 percent — the league average for a wing guard. Over 16 games, if he attempts an average of six threes per game, the total is 96 attempts. At a 36 percent rate, the expected number of makes is about 34.6. The standard deviation of the binomial distribution is the square root of (96 × 0.36 × 0.64), which is about 4.7. That means if this player is merely an average shooter, he still has roughly a 16 percent chance of reaching 39 or more makes — equivalent to 40.6 percent or higher. And there is about a 2.3 percent chance of reaching 43 or more, which is above 44.8 percent. In other words: a perfectly ordinary shooter, through luck alone, can look like a superstar for exactly 16 games. This is not a hypothesis. It is basic mathematics that anyone in that room could have performed in thirty seconds. But no one did. Because the number 42.3 percent is prettier than the correct answer. I remember a conversation with an NBA scout years ago. He told me something I have never forgotten: “We don’t pay for what a player has done. We pay for what we believe he will do.” That belief, if not anchored to data with a sufficiently large sample, is nothing more than organized fabrication. So what sample size is enough? For three-point rate, stability studies suggest you need roughly 750 attempts to distinguish a 36 percent shooter from a 40 percent shooter at conventional statistical significance. 750 attempts is about two to three full seasons. A 16-game playoff run provides less than a tenth of that. When someone says “this player proved his shooting in the playoffs,” they are talking about a sample nearly ten times smaller than required. But the problem does not stop at sample size. The deeper problem is the missing variable. In those 16 games, I coded every shot by four variables: distance to the nearest defender, time remaining on the 24-second clock, court location, and whether he was the shooter or the receiver after a screen. The result: 68 percent of his successful threes came from situations with no defender within 1.8 meters. In the regular season, that rate was 39 percent. The difference was not in the player. It was in the opponent’s defensive scheme, in how teammates created space, in the fact that he was playing next to a superstar who drew two defenders. In other words: that player did not shoot better. He was simply allowed to shoot easier. And the 148 million dollars the team was about to pay did not buy “the ability to shoot easier” — it bought a system and a superstar that the new team does not have. This is where I must be careful with myself. Because this argument can be pushed too far. If everything is context, then no skill exists. That is a false conclusion. Skill exists — but it only surfaces when we control for context. Stephen Curry shoots threes well in every system, against every defense, at every sample size. That is why he is the exception that proves the rule, not the rule. I have been criticized for trusting percentile data too much. People say I turn players into soulless numbers. They are partly right. When I place a player in the 92nd percentile for three-point rate, I can forget that behind that number is a human being who practiced 500 shots a day for ten years. But the opposite error is worse: trusting the moment and calling it essence. I learned this from a failure of my own. Years ago, I analyzed a team’s 12-game winless streak and pointed out that the problem was not the defense — where everyone placed blame — but a central midfielder who touched the ball nearly 40 percent less than at the start of the season. When I published that finding, a colleague asked me: “What if he touched the ball less because his teammates stopped passing to him?” I had no answer. That was the first time I understood that correlation is not causation, and that every variable may be the result of another variable I have not yet seen. In the 2026 transfer window, the biggest missing variable is not some advanced metric. It is a question: what system will this player play in? A 40 percent shooter in Team A’s system can become a 34 percent shooter in Team B’s system, where he must create his own shot. A highly rated defensive guard, praised for playing next to a rim-protecting center, can collapse when that center is gone. But no one codes “future system” into the tracking sheet. They code “past performance,” then pay for the future. That is why I tell my editors that the transfer window is the time when data is most easily abused of the entire year. Not because there is little data, but because too much money is staked on misreading it. Look at how the market reacts to a young player with one good playoff run. His value spikes not because the data sample has become more reliable — its size has not changed at all — but because the story has become more compelling. This is the point I call the “visibility paradox”: the less data there is, the easier it is to tell a story, and the easier it is to tell a story, the easier it is to misprice. Conversely, a player who performs consistently over eight seasons but has no memorable moment is often undervalued. His data is greater, but his story is duller. The market, by nature, does not reward stability. It rewards volatility retold in a compelling way. This is the counterintuitive angle I want to put on the table: the NBA’s problem is not a lack of data. This league has more data than any other sport on the planet — motion tracking, ball tracking, pose data, load data. The problem is that data is used to answer the wrong questions. We measure what is easy to measure — points, shooting rates, touches — and ignore what is hard to measure — the defensive attention a player draws, the space he creates for teammates, the value of not turning the ball over. The hard-to-measure things are often the ones that decide wins. It took me years to understand that the job of a data journalist is not to find the prettiest number. It is to find the truest number, and sometimes the truest number is a gap. It is an empty data cell that we must be honest about rather than fill with a story. In this analysis, I deliberately did not name the team or the specific player in the opening example. Not because I lack data. Because I want you to focus on the method, not on a name. If I named him, you would argue about the player. If I give you the method, you can apply it to anyone. That is the difference between a rumor and an analysis. A rumor needs a name. An analysis needs a method. So what will change systematically in the summer of 2026? First, I am watching how teams handle the “two-season problem.” More and more front offices recognize that a breakout season can be noise, and they are beginning to require a minimum sample size before negotiating large contracts. This is a positive signal, but it also creates a new risk: teams may miss genuinely explosive young players because they wait for evidence that only arrives after the contract has been signed. Second, I am watching the rise of position-based defensive metrics. For years, defense has been the black box of basketball data, because it depends on what teammates do. New models attempt to separate individual contribution from system, but they still require far larger samples than a single playoff run provides. If you see a team pay big for a defensive guard based on 20 games, be wary. Third, I am watching how the market prices “availability.” A player who plays 75 games a season at steady efficiency has a higher cumulative value than a player who plays 50 games with dazzling peaks. But the market often pays more for the second. This is an asymmetry I believe will be corrected in the coming years, as teams begin to model expected value across an entire contract rather than peak value. Before you watch the game, watch how the data breathes. And before you believe a number during the transfer window, ask three questions: What is the sample size? Which variable is missing? If this number were random, what would it look like? If you cannot answer all three, you do not have data. You have an empty shell. I write this in July, when the market is at its most feverish. Every day brings dozens of new rumors, hundreds of millions of dollars committed, thousands of analytical tweets. Amid that chaos, I remind myself of one thing: the chaos on the basketball court always has a hidden order, and my job is to find it, not to create it. I do not know what the team in the opening example will do. Perhaps they will sign that contract, and that player will become a genuine star. Perhaps he will fail, and everyone will say “we knew it all along.” Both outcomes are possible with probabilities I can estimate — and that is precisely the point. I do not need to know the future. I only need to know that the decision, right or wrong, was not built on a number that does not exist. If the summer of 2026 teaches the league one thing, I hope it is this: when data goes silent, the most honest answer is “I do not know yet.” And sometimes, that is the most valuable insight an analyst can offer.