When the Live Feed Returns Empty: A Data Audit Lesson from a National Table Tennis Event

The scoreboard froze at 7-5, and for the next eleven minutes the tournament's live feed returned exactly one value: null. I was sitting in the press area of a national table tennis event, still keeping score by hand, eyes flicking to the screen above the stands where 7-5 refused to move. In my headset, the commentator kept going: the momentum had just shifted, the player on this side was losing rhythm. Shifted from where, when the score was frozen and nobody in the room had recorded another serve?

An empty data source has never stopped anyone from writing a conclusion. That was the first thing those eleven minutes taught me, and the reason I opened the system logs at two in the morning instead of going to sleep.

Table tennis runs at high frequency, and that density is exactly why it deserves serious analysis. A best-of-five match to 11 points generates 40 to 50 points in under 25 minutes, roughly two points a minute. Every point is a complete micro-event: serve type, spin, placement, rally length, how it ended. Across a table 2.74 metres long, no sport produces a denser stream of events.

WTT, the commercial tour launched in 2026, built its broadcast product around that density. The ITTF live scoring system feeds the on-screen graphics, and real-time data packages are sold to bookmakers with sub-second latency. That supply chain runs continuously, quietly, and almost nobody audits it until it breaks.

Three rule changes reshaped the entire statistical distribution that analysts now treat as the nature of the sport. In October 2026 the ball went from 38 to 40 mm, slowing flight and lengthening rallies. In 2026 the format moved from 21-point games to 11, with serves rotating every two points. In July 2026 celluloid was replaced by the 40+ plastic ball, altering bounce and spin grip. Each time, the old benchmarks lost validity and the measurement scale had to be rebuilt from scratch.

Based on my experience tracking matches at domestic tournaments and the WTT feeds I have cross-checked, I keep a five-item checklist for any table tennis data source: timestamp continuity, total point count reconciled against the score, serve-receive pairing consistency, rally-length distribution, and latency against the market. That night, items two and four failed. The other three passed. To an outsider, three out of five is an average score. To me, two failures mean the dataset is unusable for any distribution analysis.

The audit returned 34 missing points across eleven minutes. When I re-counted from the video, average rally length inside the missing window was 5.8 shots, against 4.2 in the segments the system captured. The stretch that fell out of the feed was precisely where both players settled into long exchanges, the most information-dense passage of the match.

Missing data is never neutral: it goes missing in the direction of the story people want to tell. A sample truncated exactly at the long rallies inflates the share of points ending inside three shots, inflates serve win rates, and turns a patient player into a reckless one. Three match reports published that night described exchanges the system never recorded. Nobody lied. They simply did not check.

That is why I never start with the score. Every player has a crack in the armour; my job is to find it before the opponent sees it. But that crack shows up in the distribution, not in the result. A player who wins 3-0 may well have lost 60 percent of the long exchanges and survived on six serve winners. The score says 3-0; the data says a different match.

I do not trust form. I trust form data. The two rarely agree. A season is a long chain, but people only remember the last three matches.

The counterintuitive angle sits elsewhere, and it concerns the 11-point format. The table tennis world still repeats that the 2026 rule produced more comebacks, that the game became harder to predict. The observation is right; the explanation is wrong. The probability that a weaker player wins a game to 11 is far higher than the probability of winning a game to 21, simply because the variance of a small sample is larger. No player got better. The sample got shorter.

Correlation is not causation. More comebacks do not prove table tennis became less predictable; they prove we are measuring a smaller sample. The same arithmetic explains why behind-closed-doors football in 2026 produced a surge in away wins: circumstances changed, variables changed, and what people called character was an unstandardised coefficient. Empty stadiums were the largest laboratory modern football ever had, and table tennis misses its own equivalent every time a feed drops and nobody writes it down.

One detail I saved for the end. When the tournament feed died, the market did not. Prices kept moving. Someone still had a secondary source, and someone was still pricing points the audience could not see. Live data sold to bookmakers is the darkest side effect of sports digitisation, because it turns an evening of competition into a stream of signals traded before the audience understands what happened.

If I could ask a data provider only one thing, I would ask about the nulls: where they are, how long they lasted, and who decided to skip them. The next round of the analytics race will not be won by whoever collects more numbers. It will be won by whoever dares to publish the places where they have none.