Analyzing Zero Data Points: Why No Tennis Report Can Be Written From an Empty Shell

Last week an automated analysis pipeline delivered an output to my desk whose stated subject was 'tennis'. Four of its five fields read 'not applicable', the information-point count was zero, and the sentence meant to identify entities confessed that it came 'from the information points above'—where there was nothing above. This is not analysis. It is an empty shell with a subject label hanging inside it and nothing else.

I have spent 39 years in sports broadcasting and writing, from a schoolboy's voice at Radio Metrowave to my own microphone in Chicago. In that time I learned one thing: an empty framework never fills up into an honest report. If you force it, you either write common knowledge about the word 'tennis' or you fabricate events the source never contained. Both contradict my trade.

Zero information points means zero object of analysis. This is not a philosophical subtlety; it is a measurable state. Every layer of analysis—claim verification, entity matching, timeline checks, source-quality grading—stands on information points. With none, none of those four tasks can be performed. What remains is inference, and I do not label inference as analysis.

What we do have is only the domain label 'tennis'. A label makes no factual claim; it only says which field the subject belongs to. From it one can extract priors, not analysis. Possible entity classes: players, coaches, competitions (Grand Slam/ATP/WTA/ITF), governing bodies, ranking systems, court surfaces, equipment, sponsors, broadcasters. Time-sensitivity prior: high. Rankings, results, injuries, schedules, retirements and comebacks decay within days to weeks, so any specific ranking or record citation must anchor to a date. Source-quality prior: unjudgeable. Official sources (ITF/ATP/WTA/tournament sites) beat mainstream sports media, and mainstream media beat self-media and aggregators—but this ordering depends entirely on the actual source, which is absent.

I recall the debut episode of my 2026 podcast 'Split Times'. The IAAF World Championships 100m final in London: Justin Gatlin's 9.92, Usain Bolt's 9.95, Bolt's farewell run. I opened with a reaction-time regression model built in R. But notice what I had that day: names, times, dates, edition, reaction-time numbers. There were information points, so the model could stand. Today's empty shell contains not one of those ingredients. What would the model stand on?

My 2026 Russia World Cup experience taught the same lesson. To build an expected-goals model across 64 matches I needed shot data, positions and timing for every match. France's counterattack efficiency of 1.8 xG per transition did not fall from the sky; it came from raw data on every transition. And my pre-tournament bracket model ranked Brazil first and France second; after the final I audited in public the two variables that had mispriced Brazil. My credibility rests on forecasts that can be scored, and scoring requires information points.

In 2026, when stadiums emptied, I tracked serve-plus-one statistics across 300 crowdless matches, separated noise from signal, and argued in an essay that crowd absence flattened home advantage by roughly 3 percentage points. I filed it three weeks late because I kept rerunning the model. But the model was possible because information points existed. With an empty shell you can sit for thirty days and nothing emerges.

Now the contrarian angle. Someone may say that from a domain label one can still write general tendencies—time-sensitivity in tennis, source ordering, entity classes. I say: those can be written, but they are not a report; they are a methodological memorandum. Passing a methodological memorandum to a reader as an account of events is, to me, the moral equivalent of misrepresenting data. I log this limitation rather than hide it—first entry in my error ledger.

The more important question is why the empty shell arrived. If this layer is the output of an automated pipeline, the failure is probably an input that never arrived, or a silent extraction failure that returned an empty array without raising an error. In my experience silent failure is the most dangerous kind: no alarm sounds, and downstream someone assumes the analysis was completed. My 2026 lesson was that the model said one thing and the stadium said another—and when that happens you write the stadium, not the model. Here too: the pipeline says 'ready', but the product is empty. Write the product, not the pipeline's promise.

Looking forward, my proposal is simple. Either provide the source text or a segment of it—I will break it into information points, extract entities, tag time references and sources, and restart the analysis myself. Or provide the minimum structure: title, source name and type and link, publication date, article type, summary, author stance, purpose, and a list of information points where each claim carries entities, a time reference and a source attribution. The information-point list cannot run empty, and source and publication date cannot be omitted—otherwise both source quality and timeliness remain guesses.

An empty shell is not a report; it is an invitation to write one. I have accepted the invitation, but I will not write if nothing is put inside. Because the reader's time is my scarcest resource, and writing without information is its greatest waste.