An Esports Scouting Report That Returned Zero Entities: The Fault Sits in the Data Pipeline

Last month, in an office in Cau Giay, Hanoi, the leadership of an esports organisation opened a nine-page scouting report sent by a foreign partner. I sat in the fourth chair, to the right of the projector screen. Page one: game title — unidentified. Page three: target player list — empty. Page seven: three rows covering competitive, financial and personnel risk, all three carrying the same sentence: "insufficient information, no assessment performed".

Three measurable figures describe that report: zero information points, zero entities, zero game title. Three figures, and one signature.

The person closing the meeting said: "No red flags, let's sign." I did not object. I asked only that one line be added to the minutes: a report that performed no risk assessment is not a report confirming the absence of risk. The line was not added.

I kept a photocopy. It sits in the second drawer, next to the 2026 V-League xG report nobody agreed to publish. I was rejected in 2026 over a model. Seven years later, I am paid to write about it.

The esports transfer industry runs on three kinds of documents: contracts, payroll sheets, and scouting reports. The first two are bound by rules, have dispute mechanisms, have standard templates. The third has none. Nobody audits whether a scouting report is correct, because its result only surfaces eighteen months later, and by then the person who signed it works somewhere else.

A usable scouting report must answer five questions. Which game and which patch. Where the player sits on his career curve. How scarce that role is on the market. Whether that region is strong or hollow. And price, contract structure, remaining term. Remove the first question and the other four collapse simultaneously. The nine-page report removed exactly the first question.

The process producing it has two stages. Stage one reads the source article and extracts facts, entities, timestamps, sources. Stage two receives that output and analyses nine dimensions: patch, tournament format, roster, region, finance, governance, risk, narrative, industry transmission. I usually use a hospital analogy: stage one is the scanner, stage two is the radiologist reading the film. If the film is blank, the best radiologist alive cannot conclude the lungs are clean. He can only say: the film failed, shoot again.

That report did one thing correctly. It said the film failed. Nine pages, nine dimensions, every cell reading "insufficient information, no assessment performed". Technically, that is an honest answer. Commercially, it is a defective product delivered to a client with no warranty note.

What kept me in the room until the end was not the zero. It was the closing sentence of the signatory.

Esports is harder than football here: football has a shared frame of reference. A 105-metre pitch, eleven players, a 7.32-metre goal. Everyone measures with the same ruler. In esports, the game title is the law of physics. The title determines the unit of measurement, determines what is worth measuring, determines whether a given play is good or bad.

Ignore the game title and the entire downstream metric system becomes technically meaningless, not merely imprecise.

For CS2, the ruler is HLTV Rating, ADR, KAST, opening-duel win rate. For League of Legends, it is gold difference at 15, major-objective conversion, Oracle's Elixir indices. For DOTA2, people read GPM, XPM, net worth, buyback timing. For Valorant, the scale is ACS, first blood, VLR data. For Honor of Kings, the metric system is defined by Tencent and only partially public. These four systems do not convert into one another. A report that does not know which title it is describing cannot select a denominator, and without a denominator every ratio is decorative.

Patches follow the same logic. Riot ships roughly every two weeks, and a small coefficient change is enough to reorder the entire professional priority list. Valve moves on the major cadence, with far longer gaps between large updates, so a team can live inside one meta for a whole season. Tencent operates seasonally and by domestic market. Three different rhythms produce three different roster-planning doctrines. A report that cannot establish patch cadence cannot grade the magnitude of change: numerical tweak, mechanic adjustment, or full rework.

Those three categories are not equivalent. A numerical tweak leaves the old roster usable. A mechanic adjustment forces a playstyle change. A full rework forces a personnel change. Without that distinction, every transfer recommendation is a guess.

Then there is the structural detail I reread three times. In the "entities involved" field, the template read: "identify from the information points above". But the information-point list above was empty. The template asked the model to find entities inside a set with no elements.

That is the clearest signal about the nature of the fault. An article with genuinely empty content would return an empty list, but the structural fields would still be populated: domain label "esports", article type "unclassified". The model received the request, ran, filled the scaffolding, and had nothing to fill the body with. Two possibilities: the source article truly contained nothing extractable, or the text was truncated before reaching stage one.

I lean toward the second, because the probability that a real esports article contains exactly zero entities is very low. Even the shortest transfer item contains at least one person's name and one team's name.

The cheap, fast check: count the characters of the input text. If it is zero or near zero, the fault sits in the ingestion layer. Re-run stage one with verified non-empty text. If it still returns zero information points, the fault sits in the extraction model. If it is still empty after the second pass, classify the article as non-analysable: an image, an embedded video, a paywall stub, or pure advertising. Three outcomes, three different remedies. The one thing that must not happen is shipping it to the client as-is.

I have been on the other side of this. In 2026, when global football stopped, my firm took a consulting contract with a V-League club. I pulled the distance-covered data of eleven key players from the 2026 season, calculated an average 15% fitness decay after three months of no-contact training, and proposed a 20% wage-bill cut on long-term contracts. The head coach objected on the grounds that "these players have brand value". Brand value is not a variable in the model. When football returned, that group averaged 8.5 km per match, 1.2 km below their pre-pandemic level.

When I sent the wage-cut memo, they looked at me like a man without feeling. I was delivering data, not emotion.

The lesson was not in the 15%. It was this: missing injury data does not mean the absence of injury risk. Missing fitness data does not mean the player is fit. The absence of evidence and the evidence of absence are entirely different things, and in any decision process with money in the middle, the two get read interchangeably every single day.

A cell reading "no assessment performed" is a gap in the file. It is not a tick confirming the file is clean.

This is where I think the trade is breaking. The buyer pays for certainty. The seller delivers a document with the appearance of certainty: nine pages, ten tables, numbered sections, colour printing. Inside, empty cells are formatted identically to populated cells. Nothing on the page tells the reader that this cell is empty because no data exists while that one is full because data exists. Same typeface. Same point size.

In an organisation where the decision was made before the report was commissioned, the empty cell is the perfect instrument. "No red flags" is the easiest sentence to say in any meeting and the hardest to falsify. Nobody can argue against a negative claim when there is no source data to check it against.

I built an internal valuation model to quantify the damage. The formula is not complicated: transfer value is a function of actual minutes played, the age-curve slope for the role, role scarcity in the regional market, fit with the current patch, and months remaining on contract.

With complete data, a 21-year-old with 1,200 minutes last season, eight months left on contract, in a scarce role, produces a valuation band roughly 12% wide around the central value. Twelve percent is negotiable. Remove the "patch fit" field because the game title is unknown, and the band widens to roughly 34%. Remove the minutes field because the player list is empty, and it widens to roughly 60%. At 60%, the two sides are no longer negotiating. One side is buying a player. The other is selling a lottery ticket.

Band width is the only measurable damage from an empty report. It does not appear on the balance sheet. It appears here: the club pays 30% above fair value, nobody knows why, because the report said nothing at all.

I track regional matches by logging every play, not by rewatching highlights. That habit dates to 2026, when I was still competing and kept opponent notes by hand. One match is a story. Fifty matches are the truth.

And here is the Vietnamese market layer. Domestic esports organisations increasingly buy reports from foreign vendors, on the belief that data abroad is better. Technically that is correct; their data infrastructure is denser. But one factor rarely enters the calculation: a foreign vendor has no incentive to track Vietnamese leagues. A report on a VCS player, produced by a company that does not read Vietnamese comments and does not watch Vietnamese streams, will carry a far higher empty-cell rate than a report on an LCK player. That empty-cell rate is a procurement variable, and it is not written into any contract.

There is a direct financial consequence. A scouting report costs a few thousand US dollars. A bad contract costs a season, plus the opportunity cost of an occupied import slot, plus the cost of a mid-season replacement after the market has closed. The ratio between those two numbers is in the hundreds. Which means: an organisation that saves money by not auditing report quality is accepting a risk hundreds of times larger for every dollar saved. I have never seen that arithmetic presented in a budget meeting.

There is one more layer, the cultural one. Vietnamese esports fans read fast, react loudly, and are often right on the eye test. But the eye test is a sample of size one. It is not wrong. It is simply not enough to price anything. Between the transfer board and the pitch, I choose to stand in the middle, measuring both sides.

Which brings me back to the closing sentence in that meeting, because it is the centre of the whole problem.

The conventional reading: an empty report is the vendor's fault. The second reading: an empty report is the procurement process's fault. The third reading, and the one I believe: the empty report was one of the most honest documents in that room.

That sounds counterintuitive, so it needs numbers. Across the ten scouting reports I have read with every field populated, the number whose source data actually supported each individual field is very low. Most fields are filled by transitive inference: one match implies a season's form, one highlight implies ability, one rumour implies price. Those reports look complete and carry lower reliability than a report that states plainly: no assessment performed.

Put differently, two distinct failure modes exist and they get read as one. A technical pipeline failure is real, and it produced the nine-page report. An honesty failure is not what that report committed. Its error was at delivery: no warning, no note, no recommendation to re-run.

And here is where correlation misleads. One club signed after an empty report and reached the playoff stage that season. People will use that outcome to argue the process is fine. One sample, one positive result, one false conclusion. If I sign ten contracts off ten empty reports and four work out, that 40% looks acceptable until it is compared with the 55% of a process that audits its data. Nobody compares, because nobody runs two processes in parallel.

One more point, stated plainly even though it is uncomfortable in an article about a specific organisation. A vendor returning zero entities on a real article is a procurement signal. It measures the quality of that vendor's ingestion layer. A vendor that fails to notice its own input was empty will also fail to notice subtler faults: a player with the wrong role, a match counted twice, a patch assigned the wrong date. Those faults do not produce empty cells. They produce wrong numbers, and wrong numbers are more dangerous than empty cells because they look like truth.

Empty cells are visible to the reader. Wrong numbers are not. In practice, an empty report is therefore safer than a full report with no sourcing.

I do not trust intuition. I trust the version of intuition that has been verified across seven seasons. What seven seasons verified is this: bad decisions rarely come from missing data. They come from a document that looks complete, where nobody asked for the source of each cell.

So what should be done with the nine-page report. Three things, all cheap.

First, every information point carries a timestamp and a source tag at the moment of extraction. Without both tags, the point does not enter the model. I have applied that rule since 2026 and it has saved me more often than any model.

Second, every information point carries a source-reliability score. Official tournament-organiser data sits at one level. Community statistics platforms sit at another. Transfer rumours sit at the bottom and must be explicitly flagged as rumours. Blending three levels into one table without labels is the fastest way to turn a good model into a well-formatted random-number generator.

Third, and most important: a mandatory line in every report, bold, on page one. That line must separate two states. "No assessment performed" means the assessment has not been done. "No risk present" means the assessment was done and found nothing. Those sentences differ in substance and differ in liability. Merging them is the reader's error, but creating the conditions for that merge is the report writer's error.

For the organisation in Cau Giay, I proposed one thing: record this vendor's empty-cell rate across the last three reports. Above 30%, change vendor. Below 10%, keep them and request a re-run of this report. Empty-cell rate is a procurement metric, and it is free to measure.

I do not know whether they did it. The photocopy is still in the second drawer.

What I do know is that the next tracking cycle sits elsewhere. When the regional esports transfer market reopens, the question will no longer be which organisation signs the best player. It will be which organisation publishes its empty-cell rate. The first vendor willing to publish that number re-prices the entire segment, because every other vendor will have to follow or be read as hiding something.

And deeper still: when an organisation dares to write into its minutes that "we signed on incomplete data, and this is the risk level we accepted", it moves from being led by data to steering it. The distance between those two states is not a better report. It is a decision about who owns the gap.

Even a billion-dollar contract starts with a small note about minutes played. The problem is that very few people read the note before they read the signature.