The Empty Cell Also Testifies: Cricket's Silent Data Failures and the Discipline of Analysis

In 2026 I watched a BPL match on television from Khulna, logging ball-by-ball by hand in a notebook beside me. The habit was new then; I had launched a data newsletter called Expected Truth from this city, and my first project was building an xG model for the BPL. In that match, the third ball of the fourteenth over arrived while the scorers' terminal hung. My handwritten log had six deliveries; the official feed had three. The feed returned nine minutes later, and by that night a television graphic published a 'momentum curve' that had not been drawn from those six balls. The curve was not a lie. Its underlying data was incomplete, and nobody knew.

That night one idea became permanent for me: the most dangerous number in cricket is not the wrong number but the absent one that still stands on a chart. A wrong number gets caught, argued over, corrected. A missing number never introduces itself; it quietly occupies someone else's seat.

This is a methodological report against that silent occupation. There is no match result here, no praise or blame for a star. There is one question a data journalist must answer every day: when the information is absent, what is the professional decision?

Context: three layers and their silent failures

I divide cricket's information flow into three layers. Layer one is capture — scorers, ball-by-ball feeds, tracking cameras, ground reports. Layer two is extraction — isolating information points from that raw material: who scored how many, which over took which wicket, what a bowler's economy was. Layer three is interpretation, which is my job.

The problem is that a failure at layer one never issues a declaration before it reaches layer three. An empty scrape, a paywall, an encoding error, a disconnected terminal — each becomes a zero cell without announcement. The analyst then fills that zero with whatever he likes, and the reader never learns the cell was empty.

Professional information processing has a name for the remedy: null handling. Where no citable data exists, you do not speculate — you write 'insufficient information'. That is not an admission of weakness; it is methodological transparency. An empty cell honestly left empty always carries more information than one filled with colourful guesswork.

Cricket offers unusually strong temptation here. A T20 match contains 240 legal deliveries. Losing four is 1.67 percent of the total. The number looks trivial — unless those four balls are the eighteenth over where the match is decided. Then 1.67 percent can rewrite the economics of the whole game.

This is why I have published a methodology note with every piece since 2026. The habit cost me early: I delayed one article 48 hours to perfect a model. I later imposed hard deadlines and hired a freelance editor to stop my perfectionism from outranking news value. The rule never changed — if the data is not there, write that the data is not there.

Core: four kinds of absence

In my experience cricket data disappears in four ways, and each needs different treatment.

One — capture failure. The feed cut, the scrape returned empty, the paywall blocked. This is mechanical and the most detectable. The 2026 match above belongs here. The remedy is simple: verify the raw log, reconcile the delivery count. If 236 of 240 balls are present, write separately about the other four.

Two — definition failure. The data exists but nobody agrees what it means. What is a 'pressure over'? Is it the last five overs with a required rate above six? Or an over that begins with three wickets down? I have built concepts like pressure overs and recovery efficiency myself, and I always write the definition first — because a definition fixed afterwards bends the number to the author's convenience. This failure is intellectual, not mechanical.

Three — sample failure. The data exists but is insufficient. You cannot determine form from three innings or declare a venue specialist from one series. At the 2026 World Cup I tracked all seven Croatia matches — fourteen goals from 9.6 xG, a +4.4 overperformance. Because the tournament is bounded and the sample complete, you may discuss overperformance. From those same seven matches you cannot explain Croatian football culture.

Four — selection failure. The subtlest kind. The data exists, the sample exists, but the sample itself is biased. The balls a batter missed are in the tracking data; where are the ones he missed that the umpire did not call no-ball? Televised innings have granular data; the rest do not. The problem is not the numbers but the selection of numbers.

Separate these four or your analysis converges on the same error: we cannot see the model's limits because the limits hide inside the empty cell.

Three of my own numbers, and what they do not prove

In 2026 I tracked Abahani Limited Dhaka's title run — 34 goals from 26.8 xG, a +7.2 overperformance — and logged their PPDA in a 2-0 win over Sheikh Jamal Dhanmondi Club. Those numbers were clean because the data flow was intact.

In 2026, using tracking access obtained through Russia, I analysed the Bundesliga's restart behind closed doors. Across 83 empty-stadium matches, home teams' points per game fell from 1.54 to 1.21, average goals from 3.1 to 2.7, and Bayern Munich's PPDA tightened from 7.2 to 6.4. The index was later cited in five academic preprints.

All three demonstrate what a model can do when its data base is complete. None proves the model travels. Anyone who reads Croatia's +4.4 as a forecast of future overperformance will be wrong — overperformance is a loan, not a gift, and the model cannot name the repayment date.

France beat Croatia 4-2 in that final, and my pre-match model gave France a 58 percent win probability. The model leaned correctly, but 58 percent means 42 percent Croatia. Those who advertise 58 as prophecy conceal the 42. That is not model failure; it is failure in reading the model.

An integrity protocol: five pre-registered rules

Rule one — completeness threshold. If a match's ball-by-ball log is under 98 percent complete, publish no phase-level index from it; scorecard-level facts only. In a 240-ball match that means 235 balls.

Rule two — definition lock. Fix the definition before publication and do not change it mid-season. If you must change it, republish every figure computed under the old definition so readers can see the difference.

Rule three — no imputation. Never estimate missing deliveries, by any statistical method. If someone does, declare it explicitly and mark the imputed portion in a different colour.

Rule four — revision policy. When proven wrong, publish a correction and state the cause — process error or new information. Without that distinction readers cannot tell who erred: the model or cricket.

Rule five — minimum sample. Declare the minimum sample before claiming a trend. For phase indices I use eight matches; for bowling profiles, 600 legal deliveries. Below that, publish the number but never the word 'trend'.

Follow these five and an uncomfortable consequence follows: you publish less. Attractive stories get dropped because the data is incomplete. What does get published is not estimate but evidence.

Three tests for model blindness

Reproduction. Can someone else reach the same number from the same data? If not, the fault lies in method, not data. My methodology notes exist for this — so a reader can disagree with my conclusion and still rerun my arithmetic.

Counterfactual. If the number flips, does the story change? If the same story survives a changed index, the index is ornament, not cause.

Blind spot. Where is the gap between what the data says and what the ground says? This is my weakest point. A data journalist drifts toward dismissing dressing-room knowledge. Taskin Ahmed's fatigue in a spell, Shakib Al Hasan's shoulder, Tamim Iqbal's mental load — none of that appears in a ball-by-ball log. So my protocol carries a mandatory clause: every index-driven piece must include at least one non-numeric testimony — a player quote, a coach's remark, or direct observation from the ground.

Contrarian: imputation is invention in polite dress

Cricket's data industry operates on a silent compact: fill the empty cell. Broadcast graphics cannot have gaps, fantasy platforms cannot run incomplete matches, sponsors need reports. So missing deliveries get estimated — by model, by average, by nearest comparable situation. The polite name is imputation.

My position is plain. Imputation is invention in polite dress. When I say six balls were estimated, I admit I have no evidence for six balls — yet the chart will show them as though they were truth. Readers cannot see that seam, because imputed segments are usually drawn in the same colour.

An empty cell is not a failure; a falsehood in an empty cell is. If a match yields 235 deliveries, analyse 235 and say so. Readers respect honesty; they resent being shown completeness over incomplete data.

There is a second, harder reversal. An empty cell is not automatically a catastrophe. Sometimes data is missing not because the system failed but because the event sits outside the sample — a remarkable innings, an impossible catch, an unexpected win. That is cricket's randomness, not the model's error, and conflating the two is the gravest sin. The numbers didn't break the model; they exposed where the model was blind. But not all blindness is the model's fault.

One warning against myself. This protocol risks a state where integrity becomes an excuse for publishing nothing. Silence is not honesty unless the reason for silence is written down. So my rule: if you print an empty cell, print why it is empty, what is missing, and what would fill it. An empty cell does not speak on its own — it must be made to speak.

Takeaway: a verifiable signal for the next round

No summary. A pre-registered signal instead. Over the coming domestic T20 season I will track three things: the completeness rate of the ball-by-ball log in every televised match, publicly; which broadcaster or platform marks imputed data separately in its graphics; and which team is first to publish its own data-integrity appendix — naming the matches where data was incomplete and the indices withheld.

My forecast, deliberately weak: within two seasons at least one franchise will do this, and for reputation rather than commerce. Supply of data is rising; supply of trust is not. The first team to manufacture trust will sell its numbers better — even if it has fewer of them.

The question is yours. Do you want a chart that is complete, or a chart that is true? If you must choose one, what will your club, your platform, your newspaper choose?

---

Methodology note: This piece makes no match-specific claim, because the source analysis supplied no information points — every structural field was marked insufficient information. The figures used (Abahani 2026: 34 goals, 26.8 xG; Croatia 2026: 14 goals, 9.6 xG, Modric 72.3 km; Bundesliga 2026: 83 matches, 1.54 to 1.21; Bayern PPDA 7.2 to 6.4) come from my own archive and were previously published. The 98 percent completeness threshold is a pre-registered proposal, not an established standard. I don't chase outliers; I follow them until they confess. Expected truth is not a verdict; it is a sample still pending.