The Twenty-Seventh Miss and the Empty Nine-Box Table

The Twenty-Seventh Miss and the Empty Nine-Box Table

On the night of 28 May 2026, at Toyota Center in Houston, I sat in row eleven behind the home bench, next to the tunnel entrance.

Third quarter. Trevor Ariza caught the ball in the right corner, completely open, and shot. The ball hit the front rim and bounced out. That attempt was the seventeenth consecutive Houston miss from beyond the arc in Game 7 of the Western Conference Finals.

Nobody in the stands was shouting anymore. Eighteen thousand people inside Toyota Center sat as silent as an empty gym. What I remember most is the sound: the ball bouncing on the hardwood, steady, like a countdown clock nobody wanted to hear.

The final horn. The Golden State Warriors won 101-92. Houston finished the game with seven made three-pointers on forty-four attempts. That streak of twenty-seven straight misses remains the ugliest scar on a team that had won sixty-five regular-season games.

What nobody has ever explained to me, to this day: how a roster built on the league's most sophisticated shot-quality model ended the most important game of its season with forty-four attempts from deep and only seven makes.

When I left Toyota Center that night, I had a sheet of paper in my pocket, printed full of numbers. Every one of them was correct. Every one of them was useless.

A Team Designed by a Spreadsheet

To understand why those forty-four attempts were a systemic wound, you have to go back to the summer of 2026, when Daryl Morey took the general manager's chair in Houston. He brought with him a near-religious conviction: the mid-range jumper is a mathematical error, no matter who is shooting it.

By the 2026-18 season, that conviction had become a machine. Mike D'Antoni was head coach, James Harden won MVP, Chris Paul arrived from the Clippers, and Houston went 65-17. They led the league in three-point attempts. They led the league in offensive efficiency. They were the team that made every analytical model bow.

On paper, that Houston team was almost perfect. They did not take bad shots. They generated corner threes, threes off screens, threes that every shot-quality model grades highly. And through the regular season, the ball went in.

Then Chris Paul tore a hamstring in Game 5. He sat out Games 6 and 7 in a grey suit on the bench. An offense designed to split playmaking between two men now had one man, and that man had to carry forty-four attempts in a state of isolation.

The model did not account for a hamstring. The model also did not account for a shooter who knows he is inside a streak that will not end.

The Empty Nine-Box Table

I tell the Houston story to get to something else, something that happened a few weeks ago on a winter morning in Chicago.

A friend who works for a sports data aggregator sent me a report. It had a formal title page, nine sections, a risk matrix, a confidence scale, a glossary of technical terms. Formally, it looked better than anything I have ever received from an NBA scouting department.

I opened each cell. Cell one, empty. Cell two, empty. Cell three, empty. By the ninth cell, still empty. All the report would tell me, over and over, was one sentence: insufficient information, cannot assess.

My friend apologised profusely. He said the upstream data pipeline had broken, that no source article had been ingested, so the analysis layer had nothing to dissect. He called it a technical fault.

I read that report three times, and I found it better than a great many number-stuffed reports I have read.

It did not invent. It did not guess. It did not fill the void with "according to a source close to the situation." When there was no data, it said plainly: I do not know.

In twenty-two years calling NBA Finals on air and forty-four years in this trade, I have heard thousands of models. The number of models willing to say "I do not know" fits on one hand.

A Model That Does Not Know It Does Not Know

Watching NBA basketball across two decades, I have seen data enter the locker room in three distinct waves.

The first wave was box-score arithmetic: points, rebounds, assists, and a stack of percentages. The second wave was advanced metrics, where Dean Oliver laid the foundation with his Four Factors, Daniel Myers built Box Plus-Minus, and an entire generation of indices followed: Estimated Plus-Minus, LEBRON, DARKO, and FiveThirtyEight's RAPTOR before it was retired along with its parent site in 2026.

The third wave was tracking data. From the 2026-14 season, STATS camera systems covered all thirty NBA arenas. From 2026-18, Second Spectrum took over as the league's official provider. Every second, the system logged the position of every player and the ball. Every shot was assigned a probability. Every pass was measured for force, angle and spin.

What did those three waves deliver? A great deal. We know more about spacing, about efficiency, about the value of a corner three versus a mid-range jumper. We discovered that a centre who can pass from the top of the arc is worth more than a centre who can only score.

But those three waves produced something else, and that something is the subject of this piece: borrowed authority — power granted to a conclusion simply because it arrives with a table attached, not because the table has traceable provenance.

I do not doubt the competence of data analysts. Many of them are smarter than I am. But I hold one specific suspicion, and it has grown every year: analysts are moving deeper into the locker room, and their conclusions are drifting further from the actual rhythm of the game.

What is that rhythm? It is the second starter walking into a game with a sore ankle. It is a star shooter missing two in a row and beginning to release half a second slower. It is a head coach who knows that if he calls a set his bench cannot run, he loses the locker room for three weeks.

A spreadsheet cannot record those things. And because it cannot record them, it usually assumes they do not exist.

Back to Game 7 in Houston. The model knew that Trevor Ariza was shooting a right-corner three from a spot where the league-average success rate is roughly forty percent. The model did not know that Ariza had just missed four in a row, that he had played forty minutes, that the whole team was in the state American players call icing up, and that in that state an open forty-percent shot becomes a thirty-percent shot.

The difference between forty and thirty percent, multiplied by seventeen attempts, was the season.

Kazan, June 2026

Three weeks after Game 7 in Houston, I flew to Kazan, Russia, to cover Germany's final group-stage match against South Korea.

On 27 June 2026, the reigning world champions lost 0-2 and were eliminated in the group stage. The world was shocked. Broadcasters called it the biggest upset of the tournament.

I was not shocked. I sat in a beer hall near Kazan Arena, bought a few rounds for Korean reporters, and told them something that got me hammered by German fans for weeks afterwards: Germany had probably lost before the first ball was kicked — people simply did not have sharp enough eyes to see it.

The basis for that claim was a comparison I ran myself from the tournament's own tracking data. Compared with 2026, Germany in the 2026 group stage completed roughly twelve percent fewer wide vertical passes — the exact pass type that chance-quality models grade most highly, because it breaks a defensive block horizontally before penetrating vertically.

Twelve percent fewer. It sounds small. But in a system whose entire strength lies in stretching opponents, losing twelve percent of your wide vertical passes means losing the bridge between idea and goal.

Nobody in a beer hall in Kazan at eleven at night wanted to hear that. But I wrote a fifteen-hundred-word piece saying one thing: Germany died of arrogance, not weakness. And arrogance, in football as in basketball, always leaves a trace in the data before it leaves a trace on the scoreboard.

That piece drew three thousand angry comments. It was also shared across Europe. To me, both outcomes were success.

The Reversal at Anfield

But if I only told the stories where I was right, this piece would be self-congratulation, and I do not have time for that kind of writing.

On 14 January 2026, I sat in a small studio in Chicago watching Liverpool host Manchester City at Anfield. The match ended 4-3 to the home side. While the whole broadcast was praising Kevin De Bruyne, I shouted a line that got mocked across English football forums: Mohamed Salah will break the Premier League scoring record.

At the time, Salah had eleven goals in eighteen matches, and the record to beat was thirty-one goals in a thirty-eight-game season.

I staked my reputation on a player the analytics community then ranked in the second tier. By season's end, Salah had thirty-two goals, breaking the shared record of Alan Shearer, Cristiano Ronaldo and Luis Suarez.

Why did I say it? Because I did not only read the table. I watched every phase, and I saw something the table had not yet registered fast enough: Salah's dribbling speed in the opposition half, and his rate of touches inside the box per ninety minutes. His expected-goals figure was still low, but the chances he created himself had already outrun what that figure reflected.

The lesson here is clear, and it is a lesson neither camp wants to hear: data is only as correct as the reader's willingness to place it beside their own eyes.

The same tracking table looks like a second-tier player to someone reading the summary. It looks like a machine accelerating to someone rewatching the tape half by half.

Salt Lake City, June 2026

On 18 June 2026, the Utah Jazz led the Los Angeles Clippers two games to none in the Western Conference semifinals. Utah were the West's top seed, with fifty-two regular-season wins. They had Rudy Gobert, who had just won his third Defensive Player of the Year award.

That night, the Clippers won 131-119 and closed the series in six. Terance Mann, a second-year player, scored thirty-nine points.

For a week afterwards I read the data breakdowns of Utah's collapse. Nearly all of them concluded that this was a shooting-variance accident, that the Jazz were still the best defensive team in the league by composite defensive rating, that the model was not wrong and only the results had gone against them.

I rewatched all six games over three days, and I did not believe that conclusion.

Utah's problem was not in the numbers. Their problem was that their entire defensive system was built around Gobert in deep drop coverage, and when the Clippers pulled five shooters outside the arc, deep drop became a hole in the middle of the floor. Utah's composite defensive rating still looked good, because it was averaged across a whole season. But Utah's regular season did not include six straight games against a small Clippers lineup and a coach patient to the point of cruelty.

This is the largest blind spot in every basketball data model: they measure averages, while the playoffs measure matchups.

The model says Gobert is the most valuable defensive player in the league. That is true across eighty-two games against thirty different opponents. It is not true across six games against one team that has decided it will not shoot where Gobert stands.

And here is a moment that grows rarer every year yet still exists: a man in row eleven, with no access to the tracking database, can see what an entire analytics department missed — simply because he is looking at the right spot while they are looking at the whole map.

The Serbian and the Reverse Test

If this piece were only an indictment of data, it would be a bad and dishonest piece.

There are times when the data is right first and the eye must bow. Nikola Jokic is the clearest case I have witnessed.

In 2026, the Denver Nuggets drafted Jokic forty-first overall. At the time, most European basketball watchers, myself included, saw a slow centre who could not jump and would be nearly unplayable on defence in the NBA. I said exactly that on air, and I wish I could forget it.

The data models did the opposite. From his earliest seasons in Denver, when Jokic played only twenty minutes a night, impact-based metrics already placed him among the most valuable players in the league. Then came 2026-21 and his first MVP. Then two more.

The data did not invent a star. The data saw something my eyes could not, because my eyes were searching for the qualities this sport taught me to search for forty years ago.

That is the confession I have to write down: in most cases, the model is right and my memory is wrong. But in the cases that decide seasons — forty-four threes in Houston, twelve percent of wide vertical passes in Kazan, six games of drop coverage in Utah — the model went silent exactly at the moment it needed to speak.

Where I Could Be Wrong

I do not want to close this section with a hollow moral appeal. I want to name exactly where I could be wrong.

First, I may be confusing two different things: a bad model, and a good model used badly. If Houston lost because Chris Paul was injured, that is not the model's fault. If Utah lost because the Clippers made twenty percent more of their hardest shots than anyone predicted, that is not the model's fault either. There is a strong chance I am using a run of bad outcomes to indict an entire analytical field that is doing its job correctly.

Second, my suspicion may be nostalgia in disguise. I started this trade at sixteen, when all you had was one sheet of paper and a pencil, and I have an incentive to believe my era was the best. I have watched contemporaries reject the three-pointer not out of science but because they did not want to relearn the sport. I do not want to become that man.

Third, and this kept me awake longest: perhaps my reliance on feel is the bigger problem. A data analyst who errs errs verifiably, correctably, with logs. A writer like me errs in silence, in a column nobody cross-checks, in a broadcast line nobody records.

Forty-four years of watching this sport taught me one certain thing: I have been wrong many times. My hits outnumber my misses only because I talk more. And that empty nine-box table my friend sent me has one quality I have never had: it knows when to stop.

I once ran into a case that made me think for a long time. An Eastern Conference team, for years, ranked very high in offensive rating yet kept exiting in the second round. A whole system of metrics explained that they were unlucky. I believed there was a structural problem: that team had no second ball handler for when the primary option was smothered. I still cannot prove I was right, and I leave it there as an unpaid debt.

The sleeping giant is not a spell I cast on every favourite that loses. I know that trap. A brand cannot live on one sentence.

Three Falsifiable Predictions

I do not want to close with a summary. I want to leave behind three things that can be right or wrong, and I will answer for them when they are wrong.

One. Within twenty-four months, at least one NBA team will hire into a role whose title is not data analyst but data-quality auditor — someone whose job is to trace the provenance of every metric before it reaches the personnel meeting.

Two. Within the next three seasons there will be at least one major trade in which the buying team overpays for a player with strong metrics whose production came largely in low-pressure games; the selling team knows it and nods.

Three, and this is the one I genuinely want tested: within five years, a model will appear named after the very principle that empty table obeyed — a model that knows how to decline to answer when the input is insufficient. Whoever builds it first will hold a significant share of the credibility of American basketball analytics.

Every giant's failure is a slap at those who collect names instead of people. But in this era the slap stings more: it lands on those who collect numbers and forget to collect provenance.

The twenty-seventh miss on 28 May 2026 did not say Houston was wrong. It said a team and a model can both be technically correct and both stand outside the Finals door — purely because between them there was no one willing to look into the gap the spreadsheet refused to fill.

And that empty nine-box table from a winter morning in Chicago is still on my desk. I did not print it out to mock it. I keep it because it is the most honest document the sports analytics industry has sent me in ten years.