Nine Tiers and Blank Cells: How Basketball Analytics Is Fooling Itself

Miami, 2:47 in the morning. On the screen sits a nine-tier analytical table for basketball, one tier per section: tactics and technique, player data, operations and salary cap, league landscape, rules and governance, coaching staff and locker room, risk, media narrative, industry ripple. The first tier opens with a cold line: insufficient information. So does the second. The third, the fourth, all the way to the last — every cell blank.

The real story is elsewhere. This table is the final output of a deep-analysis pipeline, the kind sports newsrooms pay for to turn one article into a dozen pages of tactical notes. It ran clean. No errors, no crash, no red flags. It simply returned nothing, because the tier ahead of it — the extraction stage — returned nothing first: no title, no source, no article type, no information, not a single player's name.

At the bottom of that empty table, exactly one row is flagged as high risk. Its content: the danger of fabricating material to fill the blanks. That is what I want to talk about tonight — and it is the topic the basketball industry avoids very politely.

Context: an industry of maps nobody verifies

Basketball is the most instrumented sport on earth, and that is not an empty compliment. From the 2026-14 season, Second Spectrum became the NBA's motion-tracking provider, logging ball and player positions at roughly 25 frames per second. Multiply that across 48 minutes and more than a thousand games a season and you get a data mass nobody has ever fully read. That gap created an intermediate layer: people who do not watch the game, only the table.

That layer sustains an ecosystem of metrics. EPM comes from Dunks & Threes. LEBRON belongs to BBall Index. BPM lives on Basketball-Reference. DARKO is Kostya Medvedovsky's. RAPTOR was a FiveThirtyEight product until the site closed in 2026. Every metric prices value differently, and every metric carries a blind spot its casual users never see.

Running alongside is the industrialization of the template. A qualifying analysis needs nine sections, at least three conclusions each, two hidden insights, a comparison table and a glossary. Editors like templates because they are easy to approve. Writers like templates because they are easy to file. Neither side likes the most honest answer available: insufficient information.

Which is why today's case matters. Two pipeline stages were placed side by side. Stage one read the source, extracting information points, entities, author stance, time sensitivity, source quality. Stage two expanded that into nine analytical dimensions. Here, stage one extracted nothing.

The correct handling is what happened: stage two stopped, marked every position as unassessable, and stated why. No inference. No backfill. No invented teams, players or contracts to make the table look full.

I have covered basketball for years and have kept a personal non-traditional metrics board since 2026. It has never returned empty. But I have never once dared call it full.

What is actually missing from an empty basketball table

The tactics section needs a subject. It could be a pick-and-roll system with drop or switch coverage, a small-ball configuration, a chain of dribble hand-offs, a Spain pick-and-roll variant, or an isolation at the elbow. Here there is nothing — not one tactical concept extracted. Without a subject, every comparison dissolves: against the same team last season, against a peer group, against an ideal model. All three need a name to grip.

The player section is stricter still. A basketball article naming no players is close to impossible. No names means no points, no true shooting, no plus-minus, no usage rate. No birth date means no age curve, no prime-window judgment, no decline warning. Every path leads to one conclusion: to discuss players, you first need players.

On the salary side, this industry has its own language: the cap, the luxury tax, the second apron introduced in the 2026 collective bargaining agreement, Bird rights, the mid-level exception, future first-round picks as assets. All heavy analytical tools. But tools only mean something with an actual cap sheet. Without one, describing a team's financial flexibility or pricing a trade against the market becomes decorated guesswork. The most dangerous guesswork in this trade is the kind written in technical jargon.

The league-landscape section hits something coarser: the only surviving label is the lowercase word "basketball." That is enough to know the sport and not enough to know the league. The NBA differs from FIBA on three-point distance, ball-touch rules and game count. EuroLeague differs from the NBA in schedule structure. The Chinese league differs from both in import rules. The same tactical term carries different meanings across all three.

Rules and governance are empty in the same way. No disciplinary event, no contractual issue, no reform topic was extracted, so there is no basis to discuss rest rules, two-way contracts or a future draft age limit. Writing about rules without a rules case is writing about a courtroom with no file.

The locker-room section is the most sensitive, because it depends almost entirely on insider sourcing. A serious pipeline must tag anything below direct front-office reporting as low confidence by default. This section carries no tag simply because there is no person to tag.

Risk is the one section with a clear result, and that result sits outside the sport. Six professional risk categories — injury, load management, roster construction, tactical decodability, long-term contracts, schedule pressure — cannot be rated. But one risk rates high: analytical-integrity risk. A report that looks authoritative while anchored to no facts does more damage than an openly empty one.

The media and ripple sections follow the same logic. No rumor means no rumor cycle to measure. No transaction means no sneaker market, no broadcast-rights package, no Asian market variable, no Olympic workload to trace. Here I will say plainly what I still believe: a transfer is never real until I write it real. The writer creates transaction reality by naming it.

Two findings beyond the analysis

One finding is a systems-design problem, and it is more serious than it looks. Two extraction-stage output fields are defined circularly. "Entities involved" is described as identifying from the information points above, while the information points are empty. "Source quality" is described as judging from the source fields, while the source fields do not exist. Two fields point into the same hole.

That means the defect is in the specification, not the run. Fixing the model will not fix this. You must fix the field definitions, make title, source, type and information points non-nullable, and attach an explicit failure status to any run that cannot populate them.

The second finding is ontological, and it is the part I want to keep. The heat map has become modern basketball's new astrology. It is beautiful, it is colored, it convinces viewers they are seeing something. But a heat map shows where a player once stood, not what his job was in the system: who ordered the movement, who drew two defenders, who stood still to create space. The real role lives in the system software, and that software never makes it onto the map.

By the same logic, a complete nine-tier analytical table is a ritual of completeness. Nine cells, each with a conclusion, each conclusion with a number. The reader nods. Nobody asks whether the fourth cell was necessary, and nobody asks where the number came from.

One more note on development systems, because they sit inside the same story. Big-club academies are praised as talent factories while functioning as stockpiles. Tracking data at academy level is thin, unpublished and unverified, and the result is that fewer than one in ten prospects in major academies ever gets a genuine path to the first team. The rest serve as training decoration and are resold as a line of assets.

The counter-argument: fix the audience before the machine

The safest prediction from this story is a process fix: mandatory fields, hard gates, warnings. I want to move the other way. That pipeline did not appear on its own. It exists because someone pays for a full table.

Picture an editor receiving two drafts. The first has nine sections, three conclusions each, a third of it speculation written in a confident voice. The second is one page reading: not enough data to conclude, source needed. Which one makes the front page? I have been in this trade long enough to know the answer, and the answer is not in the writer's hands.

If a newsroom rewards filling blanks, then every pipeline, however strictly designed, will soon learn to fill them. This is the point most technology commentary misses: the problem of modern sports content is mostly a demand problem, not a tooling problem.

Basketball has a perfect counter-example. The 2026 NBA Bubble in Orlando, from July 30 to October 11, 2026, produced the cleanest data the league has ever had. No crowd, no outside noise, no on-site media pressure. Damian Lillard averaged 37.6 points across eight seeding games and dragged Portland into a play-in against Memphis. Based on my own experience watching games in that period, I wrote that Lillard would be king of the crowdless environment — and I wrote it before the results confirmed it.

The 2026 NBA Bubble had no spectators. I had no choice but to listen to myself.

Yet the same environment, with perfect data, still produced hundreds of unfounded narratives. A flawless data environment still generated a distorted sea of interpretation. That tells me the problem is not input quality. It is what people want the data to say.

Sports culture is an endless argument after the final whistle. We debate what cannot be verified and reward whoever is loudest. I forge hot takes, but the truth is what I have been forging longest.

Where could I be wrong? In assuming the source was a real article and the extraction failed. If the source was a bare headline, a paywalled stub or a captionless video, the correct conclusion is still to stop — but the fault lies in source selection, not processing. Then expanding into nine sections is not a technical accident but a wrong editorial decision made at the start. I once mispronounced a player's name on a live stream and got laughed at all night. The lesson was not pronunciation. It was never to be loud about something you have not checked.

What I believe after tonight

I believe that within twelve months, at least one major sports platform will publish an explicit gate: if the input data is empty, the analysis does not go live, however beautifully written. I also believe at least one basketball analysis will be pulled or corrected for the most boring reason on earth — a writer filled a blank with a guess, and readers took the guess as fact.

I do not write to be right. I write to open an angle nobody has looked at. Euro 2026 taught me that a hot take need not be correct, only timely. After tonight I want to add a clause. A hot take may not need to be right; data does. And the first thing data needs is permission to say: I am empty.

The question I leave with readers, with newsrooms, and with myself: if an honest empty table beats a dishonest full one, why do we keep printing the full one?