A perfectly formatted but hollow report is more dangerous than an error

On the morning of August 13, in Shenzhen, I opened a table tennis analysis report that the system had just pushed through. Nine sections, full skeleton. Tables aligned, column headers in bold, every cell filled with text. Three pages in, I still did not know which match, which player, which event, which date. Every data field carried the same value: N/A. No headline. No source. No athlete. No timestamp. Not a single index to hold on to.

Someone new to the trade would close the file and message back, "please resend." I sat still for ten minutes. In those ten minutes something more unsettling surfaced: this report had passed every automated validation step without being stopped. It was valid. It was simply empty.

In sports analysis, a report travels through four layers: source acquisition, fact extraction, entity recognition, and only then conclusion. The third layer is the least watched and the most lethal. If the system misidentifies a name, everything downstream collapses. If it identifies nothing at all, nothing downstream collapses — it just stands still. The standing still is what should make you wary.

In 2026 I backed Guangzhou Evergrande in the AFC Champions League quarter-final against Urawa Red Diamonds, on the strength of a higher xG figure. The hosts lost 0-1 at home and I lost 30,000 yuan. My error that day was not a shortage of data — I had all fourteen shots. My error was the missing positional weighting and situational context. It took me a year to rebuild my own positional database.

The 2026 lesson taught me that data needs context. This morning's lesson sits a layer lower: some days there is no data at all, and that is where the real danger lives.

Table tennis exposes that layer quickly. The WTT ranking system runs on a rolling 52-week window: old points expire on a fixed date, and a player can slide down the list without losing a single match. To read a player I need three layers — current ranking, points-expiry schedule, and the win rate against opponents from other associations. That third layer is the honest one, because it separates true ability from home-advantage and draw-advantage. A player like Ma Long entered the closing stretch of his career with six Olympic gold medals; from that point on, every season is a points-defence problem rather than a points-accumulation problem. With a file full of N/A, all three layers vanish at once. No ranking means no way to compute points-defence pressure. No bracket means no way to read draw difficulty. No roster means no way to measure the generational gap in the 23–26 age band.

The file of August 13 did not hand me a poor analysis. It handed me a correctly built frame, designed to hold conclusions, and entirely empty.

An empty model with a valid format is always more dangerous than an error message. An error message stops the process. An empty frame invites someone to fill it in.

And fill it in they will. An analyst under deadline pressure at nine in the morning will look at nine blank sections and tell himself the source must have glitched, so he will write from general knowledge. A language model handed that same frame will not stop — it will generate commentary that sounds entirely plausible about hip rotation peaks, service rhythm, draw pressure. Not one ungrammatical sentence. Not one true sentence.

In my trade, plausible commentary is the best-selling counterfeit on the shelf. It is not caught immediately, because the reader wants to believe it. It is caught only when someone traces it back to source — and by then a sum of money, a reputation, or a tournament slot has already travelled with it.

The empty frame still leaves one piece of information, and this is the part I find most interesting. It is not entirely blank: the domain label still reads table tennis, while headline, source, entity and timestamp are all gone. If the pipeline had failed at source acquisition, the domain label would have gone too. The label survived, which means the source came in, the routing still worked, and the fault sits narrowly at entity recognition. A narrow fault. Fixable in one run.

That is why I did not throw the file away. I also did not write a single word of conclusion from it. The rule I have kept since 2026: every conclusion must be able to state which information point it derives from. This morning the information-point count is zero, so the conclusion count must be zero too. It sounds obvious. In practice very few manage it, because the frame is already built and it looks a great deal like work in progress.

Numbers never lie — but they never tell the whole story either. I wrote that line in 2026, after building the model that identified Croatia as the only top-four side averaging a PPDA of 12.1 while still converting 18.2 percent of expected goals from counter-attacks. Croatia 2026 was never about believing in miracles; it was about remembering that probability was never destiny. Even that line comes with a condition: the numbers have to exist first. Without numbers there is no story to tell, and every substitute sentence is invention.

The industry reflex on encountering a file like this is to demand more data. I think that reflex points the wrong way. The problem is not a shortage of data; the problem is that the system is permitted to emit conclusions without an anchor. Fixing the anchor is far cheaper than collecting more data. It takes one hard gate: below three information points, or a null headline, return an explicit error status instead of a handsome, empty schema.

There is something easily missed here. The six risk lanes I normally scan for a player — injury, technical overhaul, equipment adaptation window, being decoded by an opponent, match density, public-opinion pressure — all lay dormant this morning. Dormant does not mean absent. Risk unassessed is not the same as risk that does not exist, and in sports reporting those two get written identically: the same blank cell.

An empty stadium is not an empty stadium — it is a laboratory. I used the crowdless matches of the pandemic years to separate the psychological variable from the technical one, and that was the period in which I learned the most about how people carry pressure. An empty dataset is a laboratory too, in a different sense: it shows you where your pipeline is weak when there is nothing to hold on to.

My most confident finding on August 13 had nothing to do with table tennis. It belonged to process. A system can emit a document that looks immaculate, all nine sections present, every field correct, and entirely devoid of information. Had I not read closely, I could have written a piece about a match that never existed.

Next cycle I will watch three things. The empty-extraction rate per run — above one half, that is a systemic fault rather than an isolated one. Whether the source-quality field ever populates — if it never does, every rumour-credibility judgement stays permanently blocked. And whether the timestamp gets stamped, because without it every points-expiry calculation is meaningless.

For now the report is still there, correctly formatted, correctly sectioned, and empty. I have left it on the screen. One day someone will ask me why I did not delete it. Because it is the cheapest evidence for an expensive point: the thing that looks like an answer is usually the thing hardest to verify.