When the Football News Data Pipeline Comes Back Empty

On Monday morning the football desk received a document that had the full shape of an article and none of its substance. Title, source, one-sentence summary, article type, information points, entities involved — every field had fallen back to N/A, Unclassified, or an empty list. Exactly one field survived: the domain label, football. The instinct is to read this as a thin article. It is not. Four independent fields degrading to their defaults at once is the signature of a pipeline failure, not of low information density. The tell is the missing title: if even the document's title layer could not be parsed, the problem sits below the extraction stage entirely. The surviving domain label admits two readings. Either the classifier and the extractor are separate calls and only the latter failed, or the label came from a source-level default rather than from the text. In both cases the conclusion is the same: nothing beneath it can be trusted. This matters because every Stage-2 action depends on the information-point layer. Argument structure, evidence strength, implicit assumptions, stakeholders, timeliness, source credibility — all of it is downstream of a layer that is currently empty. With zero input, any output is fabrication, and fabricated analysis is worse than no analysis. An editor cannot separate the invented passages from the observed ones once both sit in fluent prose. Three causes account for this pattern, and they should be checked in order of cost. First, the body was never captured — paywall, JavaScript rendering, robots blocking, or an encoding fault. The missing title points here. Second, the model's response was truncated or the structured parse failed, so the schema's defaults were emitted; the survival of the domain label while everything else is empty fits a split between classifier and extractor. Third, the article text never reached the prompt: the template variable was not substituted, and the model filled the schema on empty context. The cheapest diagnostic is two numbers: the character count of the captured body, and the length of the prompt actually sent to the model. Zero characters means the fault is upstream in capture. A populated prompt with empty output means the fault is in parsing or schema. Either number immediately halves the search space. There is a second defect worth naming: the label football is far too coarse. Match report, transfer rumour, tactical analysis and opinion column carry entirely different credibility tests. A match report is judged by its data source, a transfer rumour by the reporter's tier and club relationships, tactical analysis by the coherence of its reasoning, an opinion column by whether the stance is declared. Without this refinement, source-quality assessment collapses into generalities, and a rumour and a confirmed fact enter the same structural slot. When the information points are rebuilt, the critical discipline is separating factual claims from the author's judgement. Stage-2's evidence-strength and implicit-assumption work rests entirely on that distinction. Mixing the two at entry produces systematically biased conclusions, and the bias is invisible because it looks well organised. The minimum viable input is short: title, source, publication or event time, article type, and three to ten information points each tagged with attribution — reported by whom, or asserted by the author. If the piece carries an explicit stance, one sentence captures it. The alternative is to hand over the full text or the body paragraphs plus URL, so Stage-1 can be redone with controlled quality. Football adds a specific risk: facts expire fast. Transfer windows, injury status, manager quotes and match statistics all have validity periods, and analysis without a timestamp is not merely incomplete but wrong. The editorial lesson is that an empty result should be treated as an alarm, not a conclusion. A pipeline that silently writes N/A and moves on buys speed at the cost of trust. The honest move is to expose the empty output and investigate.