Empty Data, Full Tables: How an AI Analysis Pipeline Failed in the Sports Newsroom, and Why the Null Report Matters

Empty Data, Full Tables: How an AI Analysis Pipeline Failed in the Sports Newsroom, and Why the Null Report Matters

Part One — The Incident

The biggest structural change in sports journalism over the past few years has not happened on the pitch. It has happened on the desk. News agencies, club media departments and independent analytics platforms have split their content systems into two stages. The first stage extracts facts from raw text — which club, which player, which competition, which date, which number. The second builds a nine-dimension deep analysis on top of those facts: tactical, financial, results-cycle, league positioning, rules and governance, management and dressing room, risk, media narrative, and industry transmission. The first stage is called deconstruction; the second, deep analysis.

Between them sits a contract, and it is simple: as many facts as the first stage returns, that much analysis the second stage may produce. This contract matters because the entire analytical apparatus rests on facts, not on itself. What happens when the contract breaks was demonstrated with unusual clarity in a recent production run. The first-stage output arrived with the title marked not available, source not available, article type unclassified, the one-sentence summary empty, author stance not available, article purpose not available and — most damaging of all — an entirely empty list of information points. The only surviving data point was the domain label: football.

The second-stage analyst now faced two paths. The first was to fill the tables, because the tables existed and a filled table looks like work. The second was to mark every cell explicitly as unassessable for want of input. The professional answer is the second, and that is the subject of this report.

Part Two — What a Null Report Is

A null report is not an empty output. An empty output is what you get when a system returns nothing at all. A failed output is what you get when a system returns something wrong. The null report is a third category: a formally structured document in which the scaffolding stands complete, every dimension is present, every table is built, and every position is explicitly marked as unavailable. It costs as much effort to produce as a populated report, and it carries as much product value, because it states precisely what is missing and where.

The practice has existed for decades in industrial research, clinical trials and financial audit. If a clinical trial has an insufficient sample, the paper publishes a null result, not a guess. If a company's books cannot be produced, the auditor issues a disclaimer of opinion — a valid, recognised conclusion. Sports analysis has not yet institutionalised this practice, and that is its most significant professional exposure.

An example makes the difference plain. A null report says: the club's identity is unknown, therefore FFP/PSR exposure cannot be modelled. A filled-in but fictional report says: the wage bill stands at 78 percent of revenue, which places the club in the risk band. The second reads better; the first is usable. The difference is that the second invented a number, and every decision resting on that number — a contract, a valuation, a priority — will travel in the wrong direction.

Part Three — Nine Dimensions, Nine Voids

The tactical and technical dimension collapses first. The foundation of football analysis is the distinction between the formation on paper and the formation actually played. But that distinction can only be drawn if there is a match to describe, a team to name, a system to report. In this input there was no formation, no pressing intensity, no pass-completion figure, no expected goals, no PPDA. When the underlying text contains nothing, the question of whether a system is sophisticated does not arise. The financial dimension is in the same condition. Without a transfer fee, instalment structure, add-ons, sell-on clause, contract length or amortisation schedule, the panic-premium and above-fair-value calculations have no inputs at all.

Consider the results cycle. In daily journalism this is routine work — a losing run, pressure on the manager, movement in the bookmakers' sack race, supporter revolt. Each of those signals requires at least a league table, a fixture list or a sequence of results. Zero matches, zero points, zero standing: no trajectory can be plotted. The league landscape is more direct still. Assigning a tier from title contender to relegation candidate requires the name of the league itself. Without it there is no resource comparison, no talent-flow signal, no academy supply chain and no multi-club network analysis of the kind applied to City Football Group or the Red Bull model.

The rules and governance dimension produces the sharpest irony. FIFA's 115 charges, the points deductions handed to Everton and Nottingham Forest, the Juventus case — these are public precedents any analyst can hold up for comparison. But a precedent only bites when the club under scrutiny has a name. Here there is no club. Likewise tapping-up, third-party ownership, agent-commission transparency and FIFA Article 19 on the international movement of minors: none of these checks apply, because no transaction is described.

Management and dressing room is almost entirely person-based. Who captains the side, what friction a wage gap creates, whether homegrown and foreign cliques have formed, how much patience the owner has, whether the sporting director holds a transfer veto — all of it is inferred from interviews, leaks and wage data. With not one name in the input, the rows of the matrix have nothing to hold. The risk dimension is the same logic extended: injury, suspension, calendar load, goalkeeper depth, tactical staleness. Every risk needs at least a team or a fixture before it can be screened.

The media narrative dimension holds a particular interest. The entire hype-to-kill thesis — that early over-promotion of a player or a team seeds the eventual backlash — cannot operate without an object of hype. And source-quality grading, the one component of this dimension that needs no football names at all, is also blocked, because the source field itself is unavailable. Industry transmission says the same thing: upstream (academy, talent supply), midstream (clubs, competitions), downstream (broadcasting, commercial, derivatives). That chain only forms when an event starts it. No event, no transmission.

Part Four — The Real Risk: A Filled-In Structure

The danger here is not football risk. It is decision risk. Nine dimensions, more than thirty tables, checklists, graded ratings — the scaffolding looks so authoritative that a reader skimming the headings may assume the analysis beneath is equally solid. Reading 'tactical sophistication: insufficient information' and reading 'tactical sophistication: high' are not equivalent: the second is wrong, but the first feels weaker. A skimmer will mistake the density of the structure for the density of the analysis. That is the most dangerous consequence of this failure mode.

A null report therefore needs an explicit flag: VOID — DO NOT CITE. If the document reaches a recruitment department, an editorial desk or a commercial team without that marking, the damage is real. A recruitment reader could conclude that a club is stable, when the truth is that the club's name was never captured.

Part Five — Diagnosing the Pipeline

Technically, the failure is highly informative because its shape identifies its location. Classification worked: the domain label football was returned. Extraction failed: information points were empty. That means the break occurred after classification — the stage that pulls entities and facts out of the text. The repair surface therefore narrows to a single stage rather than a wholesale rebuild.

One plausible cause is input inaccessibility: the article may have sat behind a paywall, arrived truncated, been written in another language, or come in a format text extraction cannot parse, such as video, audio or an embedded social post. In that scenario the classifier defaults to football on some surface cue while the extractor returns null. The hypothesis rests on thin evidence and can only be confirmed by logging raw input format and length across the next ten runs.

The fix that cannot wait is a completeness gate. The rule is blunt: if the information-points list is empty, the pipeline must not return a partially populated object — it must return an explicit error. A second rule: the entities field must be made a hard dependency of information points. At present it instructs the reader to identify entities from the information points above, while no points exist above, which makes the instruction self-referentially unsatisfiable. A third: the absence of time-sensitivity must not be a silent default, because without it no article can be placed on any news cycle.

Part Six — Editorial Ethics in the AI Era

The largest lesson is that in AI-assisted journalism the danger is less the wrong answer than the filled-in answer. The pressure to fill is structural. Thirty empty cells invite completion, and language models are extremely good at completing them — they can generate numbers, formations and quotations that read as entirely credible. This is especially hazardous in sports journalism because readers do not verify; they treat a scoreline or a formation as common knowledge.

There is a commercial dimension too. The sports media business runs on volume — articles per day, speed to publish, search visibility. AI pipelines are the answer to that pressure. But in the race between speed and truth, only one of the two can be accelerated. If a system is taught that output must be produced even on empty input, it will eventually do exactly that, and at that moment the newsroom loses its most valuable capital: credibility.

One further point deserves emphasis. 'Insufficient information' should not be read as weak analysis. It is the hardest and most professional conclusion available, because writing it requires the analyst to limit his own usefulness. An analyst who can answer every question will often answer questions for which he had no basis.

Part Seven — Industry Context

This incident is not isolated. AI-assisted content production has expanded rapidly across sports media in the past two years: match reports, transfer summaries, player profiles, previews and reviews. Much of it is technically accurate because it rests on hard data. Where data gives way to narrative, failure rates rise sharply — above all in transfer rumour. An agent's motive, a club's financial pressure, a contract's expiry date: all are easy to infer and easier still to invent.

The industry therefore needs a modest discipline: source-tier grading — Romano-tier, general media, tabloid — and a clear rule that no claim is published whose source cannot be named. That is not a technology problem. It is an editorial policy problem.

Part Eight — A Minimum Data Checklist

To convert this null report into a substantive one, the first stage must supply at minimum: article title, source and publication date; at least one named competition; at least one named club; at least one named individual, whether player, coach, owner or executive; at least one quantitative fact, whether fee, wage, contract length, table position, xG or points total; at least one dated event, whether match, signing, sacking, ruling or filing; the article type; and the author's stance. Three elements — a club, a player and one dated number — are enough for all nine dimensions to activate.

Part Nine — Conclusion

A report that states clearly that it has nothing is worth far more than a report that conceals what it had and what it lacked. The next decade of sports journalism will turn on that distinction. AI will be exactly as reliable as its input, and the capacity to recognise an empty input — a capacity the system must develop for itself — will be its hardest test.

Glossary

• xG (Expected Goals): a model estimating the probability that a given shot becomes a goal, separating process quality from scoreline.
• xGA: the defensive counterpart — the xG a team concedes.
• PPDA: passes allowed per defensive action; lower values indicate more aggressive pressing.
• Low block: a deep defensive shape conceding territory to deny space in behind.
• FFP: UEFA's financial control regime limiting club losses.
• PSR: the Premier League's profit and sustainability rules; breach triggers points deductions.
• Transfer amortisation: spreading a fee across the contract term to reduce annual reported profit.
• Sell-on clause: a former club's right to a percentage of a future fee.
• Contract year: the final year of a contract, associated with renewal brinkmanship and form volatility.
• Tapping-up: approaching a contracted player without the club's permission.
• Multi-club ownership: one group controlling several clubs, creating eligibility conflicts in the same European competition.
• New-manager bounce: the short-term uplift often seen immediately after a coaching change.
• Six-pointer: a fixture between direct rivals whose result materially swings the gap.
• FIFA virus: injuries and fatigue sustained on international duty.
• Hype-to-kill: the media dynamic in which early over-promotion seeds an eventual backlash.
• Null report: a formally structured output in which every position is explicitly marked unassessed because the required inputs are absent — distinct from an empty or failed output.

Disclaimer

This report is methodological throughout and contains no factual claim about any club, player, coach, competition or transfer. Sporting outcomes are highly uncertain; this material is offered for sports-information reference only and does not constitute betting advice.