Zero Facts, Full Template: The Silent Failure Inside a Football Data Pipeline

When a football analysis engine is asked for a nine-dimension report and returns a document in which every heading sits in place and every table is drawn, yet not one cell holds a hard fact, the problem is not the analysis but the data moving through the frame. The episode is a warning for sports journalism and the data-driven sports industry alike, because both now rest on the same automated pipelines. The shape of the incident is simple. First-stage deconstruction returned a blank title, blank source, unclassified article type, an empty one-sentence summary, no author stance and no stated purpose. The list of information points was entirely empty. The 'entities involved' field instructs that entities be identified from those information points, which do not exist. The 'source quality' field asks the same of source fields that were never captured. Time sensitivity is recorded explicitly as not assessed. Three causes are plausible and none can be verified from this input. First, an upstream extraction failure — a paywall, a JavaScript-rendered page, a PDF or OCR failure, or a truncated API payload. Second, a source that genuinely returned a stub: an index page, a photo caption, or an empty live-blog shell. Third, a deconstruction model that scaffolded its template before content arrived, in which case the 'not available' values are artefacts rather than findings. The most telling signal is circularity. Entity extraction is normally the most robust step in football deconstruction and rarely returns nothing on a real article. When both 'entities' and 'source quality' are defined as derivations from a list that is itself empty, the schema cannot stand on its own base. The structure also reveals that it expected per-point source attribution — a design feature that produced no output. The second serious omission is time. Football information decays in days and weeks: lineups, injuries, table positions. A results assessment is uninterpretable without knowing whether it describes September or April. The absence of a publication date is not a small gap; it attaches an unmeasurable staleness discount to the whole report. This is the silent failure: well-formed but empty output, indistinguishable at a glance from a legitimate finding of nothing. A human reviewer and an automated consumer can both fall into the same illusion. A bot reading this document would conclude the analysis ran and found nothing, when in fact the analysis never ran. Downstream risks are concrete. Transfer-rumour credibility grading depends entirely on this dimension, and a blank source name shuts it down completely — an authoritative journalist and an aggregator mimic each other's phrasing, so prose style cannot substitute for source tier. Likewise financial fair play, profit and sustainability rules, the contract-year effect and multi-club ownership conflicts cannot be checked, nor dismissed, without a named club and a named governing body. There is an ethical boundary here too. Writing a thousand polished words of 'original' coverage on an empty input is not technically difficult. What is difficult is believing it is acceptable. Fabricated tactical analysis, invented fees and imagined contract clauses read exactly like the real thing, and they settle permanently into datasets. The correct output is a structured refusal, not padding. The fixes are cheap and obvious. First, add machine-checkable flags to first-stage output — fact count and entity count. When the fact count is zero, hard-fail the handoff rather than passing a valid-shaped empty object downstream. Second, treat 'entities' and 'source quality' as independent extraction targets, a separate NER pass and a source-domain lookup, not as derivations. Third, require an absolute date and an explicit as-of snapshot for any table or injury data; relative labels such as 'recent' are worthless. Monitoring signals should be fixed as well. If the source field is populated in fewer than ninety per cent of items, rumour grading is systematically unavailable. Repeated 'not assessed' returns on time sensitivity are a direct hazard for time-critical stories. And segment failures by source type: if they cluster on paywalled, non-English or live-blog content, the fault lies in the ingestion adapter, not the model. One point must remain clear. An unassessable risk profile is not a low-risk profile. Unknown risk is, for decision-making purposes, strictly worse than identified and quantified risk, because identified risk can at least be managed. What this report genuinely supplies is therefore not a judgment about football but a systemic fact. The conclusion is not that the system analysed badly. It is that the system analysed nothing with great tidiness, and that tidiness is its chief weakness. A real sports report will exist the day the input returns with a name, a competition and a date. Until then this document should be read as what it is: not content, but a work order for repair.