When the Football Algorithm Loses to an Empty Spreadsheet: Lessons from a Broken Esports Analysis Pipeline

There is a moment in this profession I call "the thirtieth minute": thirty minutes before kickoff, the data system collapses, and you must choose between talking around it or talking straight. I once chose to print three pages of outdated figures, marked clearly, and walk into the studio with a Plan B in hand. But this time, what I received was not a system failing mid-run — it was an esports analysis pipeline returning exactly one outcome: a complete nine-dimension analytical framework with every content field filled by the letters N/A.

That is where the real test began.

The context here is not a match. It is a two-stage pipeline: Stage 1 is responsible for deconstructing the source article — extracting the title, source, article type, one-sentence summary, author stance, article purpose, list of information points, entities involved, time sensitivity, and source quality. Stage 2 uses that output to conduct deep analysis across nine dimensions: patch and meta, tournament system, teams and players, regional landscape, club finance, rules and governance, risk profile, public narrative, and industry transmission.

The Stage 1 output returned: title N/A, source N/A, article type "unclassified", empty summary, author stance N/A, purpose N/A, a completely empty information-points list, an entities field containing exactly one instruction — "identify from the information points above" — and time sensitivity marked "not assessed in Stage 1". Only one field was correctly populated: domain label, esports.

One correct field out of eleven. If that were an operating metric for a team, it would be a signal to fire the coach within the week.

But what made me stop was not the 1/11 figure. What made me stop was how the Stage 2 system reacted. It did not fabricate. It did not guess which game, which league, which team. It output the entire nine-dimension framework labelled "N/A — insufficient information" in every substantive position, along with the most important note in the entire document: in this text, N/A means "unknown, not assessable", and must never be read as "no issue found".

This is the point I want esports fans to remember, because it applies equally to football: a report with no risk flags is not a clean report, it is simply a report yet to be checked. The difference between "no evidence of risk" and "evidence of no risk" is the entire foundation of this profession. Conflating the two is the fastest way for a club to lose money, a league to lose credibility, and a player to lose a career.

Technically, the failure signature here is clear. The domain label was correctly populated, but the information-points list was empty, title and source were both N/A, and the entities field still contained the template's instruction text. Those four signals together do not indicate that the source article was empty. They indicate that the extraction step never ran — or exited early immediately after domain classification, before reaching information extraction. In other words: this is a process failure, not a content failure. The source article is very likely intact and re-extractable.

I have seen exactly this pattern on a different pitch. In 2026, during the European Championship, I handled the data for the Italy versus England final. I put a judgement into the bulletin based on xG differential — Italy 2.1 against England's 0.9, despite only 42% possession. The director called me rigid. When Italy won on penalties, he apologised and handed me the data lead role for the AFC U23 semi-final in Shenzhen. The lesson I took was not "I was right". The lesson was: when you have a standardised process, you can hold your position under challenge. When you have no process, you have only a feeling — and feelings do not survive studio pressure.

How does that standardised process actually operate? It has three layers. Layer one: data in. If the input data is empty, the process must stop; it may not run on. Layer two: analysis. Only once layer one is verified may layer two issue judgements. Layer three: reporting. Every judgement at layer three must trace back to a specific data point at layer one.

The system in that document did layer one and layer three correctly, but skipped layer two in any meaningful form. It stopped at the right moment — it did not invent a story about some patch, did not assign some team, did not construct a public narrative out of nothing. That is correct behaviour. But it also reveals a larger gap: if this empty result is fed into an automated chain without a gate, the system will generate esports analysis at scale that is fluent, syntactically plausible, and entirely fabricated in substance.

To me, this is the most serious risk in the whole story — more serious than an abandoned analysis.

Look at the thresholds required for any esports analysis to carry value. Priority zero: the game title. Without it, not one of the nine dimensions is valid — LOL, DOTA2, CS2, Valorant, Honor of Kings, and Peace Elite have patch cadences, meta lifecycles, and tournament structures so different that cross-comparison is methodologically meaningless. Priority zero, second: at least one named entity — team, player, tournament, or publisher. Priority one: the patch identifier if the article concerns gameplay; tournament name, tier, format, series length; source name and publication channel. Priority two: publication date, regions involved, any monetary figure. Priority three: the alleged rule violation, if any.

That list is not long. But it is the boundary between analysis and speculation.

I want to be direct about how we read numbers in this industry. In football, we have grown used to 60% possession being hailed as proof of dominance. But the Liverpool versus West Ham 4-1 match I wrote about at thirteen had only 38% possession, with 19 shots and 7 on target. That 38% figure did not lie. It simply said something different from what people wanted to hear. And that is precisely why an empty "information points" field is more dangerous than a bad number: a bad number is still information; an empty field is the absence of information, and that absence is routinely read as permission.

The same logic applies to esports. A team with no red financial flags in a report is not necessarily healthy. A player absent from an injury list is not necessarily fit. A tournament with no recorded integrity violations has not had its integrity verified. Those three sentences sound obvious. Yet in operational reality, I have seen million-dollar transfer decisions made on exactly this kind of silence.

The second notable thing in that document is its structure. It does not merely say "I don't know". It says "here is exactly what I need in order to know, in priority order, and here is which analytical dimension each data field will unlock". That is the difference between a person admitting a lack of information and an operator admitting a lack of information.

The first says: "I have no data." The second says: "I have no data, and here is the data-recovery specification — send me the game title and I can open four analytical dimensions immediately."

In my profession, the second person always has work.

Look at how that system handled each dimension. On patch and meta, it offered a conditional conclusion: if the article concerns a stat-adjustment patch within roughly plus or minus five percent, competitive disruption is typically low to moderate with a short adaptation window. If it concerns a mechanic rework or map rotation, disruption is typically high and can invert a regional hierarchy. This is not a judgement about a specific event. It is a rule of thumb, explicitly labelled as a rule of thumb, with low confidence attached.

That is how a good analyst says "I don't know" while retaining professional value.

On tournament systems, it made a claim I consider the single most important in the entire framework: format variance is the largest lever on upset probability. A BO1-heavy or single-elimination structure systematically inflates the apparent strength of weaker teams; a BO5 or double-elimination structure rewards adaptation and roster depth. No format was supplied, so no variance estimate is possible.

I have seen this in football: a team wins one knockout match and is hailed as title contenders, when across the whole season they finished eighth. One match is not a season. One BO1 is not a tournament. And one long-range strike into the top corner is not a tactical system.

On teams and players, it set out a very practical classification threshold: replacing three or more starters is conventionally treated as a rebuild rather than a reinforcement, carrying higher coordination costs and a longer adaptation window. That threshold applies to both esports and football. Barcelona replacing half a squad in one transfer window is not reinforcement — it is reconstruction. And reconstruction takes time, which the league table does not provide.

It also set out a role-specific age sensitivity threshold: reaction-dependent roles such as entry or rifler in FPS, mid or AD carry in MOBA, decline earlier than shot-calling and support roles. I want to extend that claim to traditional sport with a caveat: an attacking full-back at 28 and a holding midfielder at 33 do not sit on the same curve. Anyone saying "a 30-year-old player is finished" is using a single curve to describe twelve different roles. Numbers never lie; only impatient readers do.

On club finance, the document offered a figure I want you to note: esports clubs commonly run salary-to-revenue ratios above 80%, making them structurally loss-making. That is an operating reality, not an opinion. And it explains why "club fails to pay wages" stories return as regularly as a rainy season. Sponsor concentration risk — a single sponsor exceeding half of revenue — is a standard red flag. Parent-company contagion risk, particularly where the backing capital is real estate or a streaming platform, is a recognised systemic risk vector in this industry.

But every one of those judgements is conditional. No event was described. No club was named. And with no source, it is impossible to distinguish between a figure from a club press release, a journalist's estimate, and a leaked contract value. Those three sources carry entirely different weight. Process is the only thing that holds when pressure rises.

On risk, the document offered a rating I want to quote verbatim for its precision: the overall risk rating is N/A, cannot be rated. The basis: a risk rating requires at least one identified risk item with an identified subject. The input contains zero identified subjects. The only assessable risk in the entire document is a risk about the analytical process itself — the risk that downstream decisions are made on the basis of an analysis whose inputs were empty. And that risk is rated High.

That is a conclusion I would frame and hang on the wall of every club's analytics room.

On public narrative, the document set out the mechanism I consider the most accurate description of an esports "phenomenon" lifecycle: a heat cycle of four stages — budding, accelerating, climax, backlash. Overhyping at the climax stage typically plants the seeds of a later backlash. And the primary test for whether a narrative is organic or manufactured is cross-channel consistency — official media, vertical esports media, and short-video or community channels.

I want to add one observational variable from my own tracking experience: when a player appears simultaneously across all three channel types within the same week with the same talking points, the probability is high that this is a deliberate communications campaign, not an organic public reaction. Fans remember the goal; I remember the numbers behind it — and also the silences on the channels that are not talking about it.

On industry transmission, the document set out a very clean three-layer structure: upstream is the publisher with patch cadence and event licensing; midstream is clubs, events, and streaming platforms; downstream is sponsorship, derivatives, and mainstreaming. It also pinpointed exactly why transmission analysis depends on the publisher: Riot's fortnightly cadence, Valve's infrequent major updates, and Tencent's season-based model produce ecosystem dynamics that differ in kind, not merely in degree.

That is the principle I apply to football: never compare two leagues with different power structures without declaring it. The English Premier League and V.League have the same number of players on the pitch, but not the same financial structure, not the same broadcast-rights allocation, not the same fan behaviour. Transplanting a governance model from one to the other while ignoring those three differences is the fastest way to produce an analysis that sounds excellent and is entirely wrong.

So what do we learn from a document that should have been a failure?

First lesson: a null-check gate is far cheaper than a fabricated analysis. If the information-points field is empty, or the title is N/A, stop and raise an alert; do not generate. Writing that check costs a few minutes. Distributing a fabricated analysis in this industry is measured in audience trust — and trust cannot be refinanced.

Second lesson: "N/A" is not "no risk". This is the most important terminology note in the entire document. Anyone reading a report and seeing an empty box must immediately ask: is this empty because there is no issue, or empty because nobody checked? Those two answers lead to entirely different decisions.

Third lesson: a process must have a Plan B, and the Plan B must be prepared before it is needed. In 2026 in Qatar, my data system collapsed thirty minutes before the Argentina versus Netherlands quarter-final. I did not wait for a fix. I immediately sourced a backup from the FIFA website, printed three marked pages, and used Argentina's average of two yellow cards per match to commentate. After the match, I proposed building a cloud backup data repository. The editorial board adopted it within a week.

That is the operating pattern I want to see in every analytics team: do not complain about a broken system, identify the process fracture, and patch it before the next piece goes on air.

There is one question I always ask of any analytical report before signing off: if every number in this document were deleted, what would remain? If the answer is "a plausible-sounding story", the document does not deserve publication. If the answer is "a framework traceable to source data, plus a list of what is still missing", then it is a document that can withstand studio pressure.

That document, despite empty inputs, passed that test. It did not give me a story. It gave me a data-recovery specification, a priority list, and a warning that if this chain runs automatically without a gate, it will generate exactly what our profession cannot tolerate: confidence without basis.

Do not ask who will win; ask which way the data is leaning — and if the data is not yet there, be the first to say so before anyone has time to pretend it was.