A Birth Certificate for the Schema: The 'Tennis' Dataset That Contained Crude Oil

One

A record landed on my desk last week. One label on top: tennis. I opened it. No player inside, no court, no first-serve percentage, no break-point conversion. What was there instead: Brent crude at $105.52, WTI at $92.93, a $12.83 spread between the two benchmarks, US diesel at $6.528 a gallon, and 33.7 million barrels of crude moving through the Strait of Hormuz every week.

Zero on the scoreboard. Zero is the loudest number in the file. In 2026, during the Russia World Cup, I built a spreadsheet tracking xG and PPDA across all 64 matches. That experiment started with one question: what did the scoreboard hide? This time the question has flipped. The scoreboard shows nothing, because the scoreboard has been placed on the wrong pitch.

Someone called this record tennis. It does not contain a single molecule of it.

Two

Context matters here, because the lesson hides in the pipeline.

How does a sports data pipeline work? A match happens. A scorer collects data. It enters a database. A domain label sits at the top: tennis, football, cricket. Then it routes to analysts. The label is the ticket that decides which analytical engine runs.

Wrong label, wrong engine. Two kinds of damage follow. A crude-oil report contaminates a tennis database; and a potentially important energy-market analysis sits stuck in the wrong department.

In 2026 a rotator cuff injury ended my junior career at the Barishal divisional training centre. From a bed of pain I made a decision: I would not quit the sport, I would count it. That year at the Ramna complex I hand-logged serve percentage, unforced errors and break-point conversion for all 32 matches of the National Tennis Championship, and named the Facebook page Data Court. My post showed the champion won only 54 percent of baseline rallies but 78 percent of net approaches. It circulated through Dhaka's tennis clubs; BTF officials started reading.

Injury taught me that pain is just unstructured data waiting for a schema.

The baseline year is 2026, when the National Championship launched; the high-water mark is 2026, the Davis Cup Asia/Oceania semi-final. Then three dormant decades. I read those decades as missing observations, not missing ability. The schema was broken, not the players.

I built my first database because memory alone could not carry the weight of a season. In 2026, when sport stopped, I assembled a database of more than 500 matches played behind closed doors. Home advantage in football fell 32 percent without crowds; tennis serve percentages stayed flat. That same hiatus I learned Python and SQL and built my own scraping tools.

So the label question is not theory for me. It is method.

Three

Now the dissection.

Across the nine dimensions I use for tennis analysis—technical and tactical, data and form, tournament system and schedule, tour positioning, rules and governance, team and player management, risk, media narrative, industry transmission—every single one returned the same answer: insufficient information, off-domain.

That is the first real finding. A fully null nine-dimensional matrix is not a failure; it is the only legitimate analytical yield this record can produce. Forcing a mapping—treating oil supply as a serve, Hormuz flows as return points won—would not be analysis. It would be fabrication.

I refused. That refusal is the story.

The null is not passive. In a healthy pipeline, a wrong-domain record should be stopped long before the analyst layer. Here it reached me, and the analyst layer caught it alone. The defence worked, but only at the last gate.

Then there are three internal Stage-1 defects.

First, the domain label itself. The entire article concerns oil markets and Middle East geopolitics.

Second, the Entities Involved field was never populated. A placeholder sits in its place, instructing the reader to identify entities from the information points above.

Third, Time Sensitivity was explicitly left unassessed.

Read together, these three suggest something: probably not isolated carelessness, but a process bug. Placeholder text in one field and a skipped assessment in another rarely appear together by random error. That pattern looks like a specific code path failing.

Now the uncomfortable part.

The scenario described—a US–Iran war running since the end of February, a naval blockade, a Hormuz closure, record US diesel prices—does not match mainstream reporting. There is a LONDON dateline and no named outlet.

Let me be precise. This is not proof the material is false. It is an indicator that it is unverified. The difference is enormous. The pipeline question becomes: where did this record come from? Who compiled it? Which wire carried it?

This is where blockchain enters, and it enters exactly on point.

Provenance is now central in sports data. It means a birth certificate for information: who collected it, when, from where, by what method, and whether anyone altered it later.

Blockchain answers part of that brilliantly: change detection. Once a record is hashed and anchored, no one can quietly edit it. A forensic audit trail exists.

This matters acutely for Bangladeshi tennis, because our verifiable name pool is roughly six: Khaled Salahuddin, Sree-Amol Roy, Shibu Lal, Ranjan Ram, Jonathan Mridha and Zarif Abrar. Mridha's career high was around 508. Zarif Abrar's 2026 ITF junior title—the first by a Bangladeshi—is a trend line, not a trophy.

What does one bad record do to a six-name dataset? The arithmetic is brutal. In a large dataset one bad record is noise; in a small dataset that same record is a percentage-determining event. When n is small, one false value carries enormous weight.

Our domestic scoreboard hides other things too. A home Davis Cup tie in Dhaka has moved local tennis more than any talent hunt. This is a club-based, elite-adjacent sport bound to Ramna, Gulshan and the Officers Club, not a stadium sport. Yet sponsors follow TV, and TV ignores the sport.

Now apply that same economics to data. Pipelines chase volume, not verification. Labels are applied fast to grow volume; verification happens slowly, if at all.

Suppose every Bangladeshi tennis record were hashed on-chain. Who, when, and where would become immutable. Excellent. But a quiet trap remains.

Blockchain can prove a record has not changed since it was written. It cannot prove the record was correct when written.

Blockchain is a notary, not an editor. It seals; it does not judge. And if the label itself is wrong, immutability makes the error permanent. A wrong label that can be corrected is far safer than a wrong label carved in stone.

When I wrote the first data-driven profile of 14-year-old Zarif Abrar in 2026, it worked because every layer was questioned: where did this number come from, who saw it, how many times. The right to ask was doing the work of provenance.

Every transfer rumour is a missing value. So is a mislabeled dataset.

Four

Now turn the mirror, because this is where I learned to distrust my own first reaction.

The reflex: automation broke, AI is dangerous. Comfortable, and probably wrong.

State the null hypothesis plainly. Assume the label was a human error, or an automated router mis-keying on a keyword. Neither is proven.

What does the null say? If only one record is mislabeled, we cannot claim systemic rot. n equals one. One case supports no conclusion.

Yet two things appear together: a wrong label and incomplete fields. Different failures, same batch. That combination is the real signal.

The second counter-intuitive observation: the most valuable output here is not the error. It is that the framework refused to fabricate. Nine dimensions returned null rather than dress themselves up.

Expected goals are not prophecy; they are a lantern held against a dark stadium.

And the real blockchain story is this: the industry talks endlessly about currency security and rarely about data authenticity. We built technology for trustless environments but left the human layer of pre-verification empty.

Five

What should we watch? Six signals sit in my notebook.

Domain-label accuracy. Compare label against keyword classification per batch. If another record with no player or tournament name arrives labelled tennis, routing is systematically failing.

Source provenance. Cross-check claims against independent records. Absence of confirmation is not confirmation.

Field completeness. Measure population rates for Entities Involved and Time Sensitivity separately. Repeated placeholders mean an extraction bug.

A label-confidence gate. Add a keyword-consistency check before the domain label is committed.

And a condition for ourselves: we need our own schema before we can catch anyone else's bad labels.

Who asked for a summary? I am leaving numbers instead: one wrong label, three incomplete fields, six verifiable names, and a null matrix that refused to break its own honesty.

The next-round signal is this. Do not stop asking what the data says. Start asking who said it.