The Data Integrity Gate: Why Three Cricket Claims Failed the Public Record

At 11:40 pm in a Sydney flat, with rain on the window and three columns open on a laptop, the cursor stopped on the third row. The left column held the claims pulled from a Stage-1 packet. The middle column held the public record: scorecards, series archives, published schedules. The right column held my own checklist, three boxes beside every claim — does the entity match, does the era match, does the denominator match. All three boxes were red.

This is not new. When I built my first xG model in a Sydney bedroom in 2026, I logged 1,248 shots in Excel and then watched France beat Argentina 4-3: France scoring four goals from 2.1 xG, Argentina three from 1.4. The eye test and the number disagreed. What I learned that night was not that numbers lie. It was that numbers are honest and the claims attached to them are frequently not.

What the gate actually is

My first lesson in this discipline came from football. In the first five Bundesliga rounds after the 2026 restart, home win percentage fell from 43.3% to 33.3%. Sydney FC beat Melbourne City 1-0 in an empty Bankwest Stadium in the A-League Grand Final. Using PPDA and distance covered, I found the home xG advantage had dropped by 0.25.

"Empty stadiums did not erase home advantage; they exposed its source."

The same logic applies to cricket. Australia won the 2026-25 Border-Gavaskar Trophy 3-1, but the series was really three different games. Perth seamed and bounced. Melbourne produced Nitish Kumar Reddy's century in completely different conditions. Sydney's final Test ended on January 5, 2026. Anyone who reads the series through a single aggregate average has collapsed three contests into one.

A data integrity gate is not a grammar check. It is a decision door. Before a number enters analysis, five questions must be answered: whose number is it, from which era, on what denominator, in which format, at which match state. Fail one and the number does not enter.

I have watched cricket for nine years and written about its numbers for six. Most of my wasted time has gone into metrics that look innocent but smuggle an assumption inside. PPDA is the classic case in T20. In football, PPDA measures pressing intensity because the duel for possession is spread across the pitch. In T20, balls per over fall, six-hitting risk discourages pressing, and spinners bowl overs 7 to 16. The same formula carries two different meanings.

Three failures

The three flagged items in the Stage-1 packet represent three distinct error types. None of them is an accusation against a particular writer. They are structural gaps that keep returning.

Failure one: era amnesia

The first error lifts a number out of the era that produced it. Cricket changes faster than most team sports because ball, pitch preparation, fielding rules, DRS, boundary dimensions and match volume all shift together.

Don Bradman's 99.94 average invites a better question than most people ask: 52 Tests, 80 innings, uncovered pitches, unlimited time. A duck in his final innings would have pushed the average past 100; instead a single innings decided a century's symbol. The number is extraordinary, but it belongs to a different game.

"Small samples are loud; large samples are honest."

The most common modern version of era amnesia is pooling pre-DRS and post-DRS data. The definition of lbw, umpiring tendencies, even batters' pad habits changed after 2026. Averaging wicketkeeper catches across both eras describes no player who ever lived.

The 2026 Champions Trophy final is a useful illustration. On March 9, 2026, India beat New Zealand by four wickets in Dubai, Rohit Sharma making 76. Put that match in the same table as a 2026 fixture and you lose Dubai's slow, dry surface and the evening dew.

Failure two: denominator fraud

The second error is the most common and least discussed. When you see a player's average in wins, strike rate in chases, or fourth-innings record, the first question should be: what is the denominator? How many innings, balls, matches?

After India won the 2026 T20 World Cup, numbers about Jasprit Bumrah circulated widely. On June 29, 2026, in Barbados, India beat South Africa by seven runs and his spell was decisive. But a tournament economy rate rests on a handful of four-over spells. The number is evidence of skill; as a forecast it needs its denominator stated.

Bumrah's 32 wickets in the 2026-25 Border-Gavaskar Trophy were remarkable, the highest by an Indian seamer in a series. Two conditions must travel with that figure: the five pitches offered seam and swing, and he did not play the final Test. Drop those and you get claims about invincibility in Australia that the data cannot carry.

Chasing splits are worse. A batter who bats often in the fourth innings accumulates sample; a batter never given the role accumulates none. Zero sample means a different role, not zero skill. When Royal Challengers Bengaluru won their first IPL title on June 3, 2026 in Ahmedabad, beating Punjab Kings by six runs after eighteen seasons, much of the clutch-player analysis that followed stood on very small denominators.

Failure three: proxy substitution

The third error is the subtlest. Nobody gives a wrong number. Someone presents a proxy variable as if it were the thing itself.

"I do not trust a number I cannot trace to a touch."

Dot-ball percentage in T20 is a good example. Low dot balls supposedly mean aggressive batting, until a batter takes a single instead of a dot to keep strike and hits bigger next over. The same percentage names two different strategies.

Fielding is the other familiar home of proxy substitution. Catch counts cannot judge a fielder because chances arrive from bowler pressure. A slip fielder does not control how often the ball comes to slip. When South Africa beat Australia by five wickets in the 2026 World Test Championship final at Lord's, reading that match through drops alone erases the story of sustained bowling pressure.

The five-step protocol

One, entity. Whose number is it? Name collisions across players, teams and tournaments are routine in cricket.

Two, era. Which year, which ball rules, was DRS in use, how far was the boundary, was it a day-night match.

Three, denominator. How many innings, balls, overs, matches, and how was that denominator constructed — meaning, what selection bias does it carry.

Four, format and role. Test averages and T20 strike rates do not share a table. Opening and finishing are different jobs.

Five, match state. Forty for three and 240 for three are different games inside one innings.

At Euro 2026, Italy's 65% possession, 19 shots and 2.1 xG only became meaningful against England's 0.8, with Jorginho covering 12.9 km to hold a PPDA of 8.7. Italy conceded four goals in seven matches, and I wrote at the time that the pressing was superb in a tournament and an open question across a season.

At the 2026 World Cup, Argentina lost 1-2 to Saudi Arabia with 2.3 xG and 15 shots against 0.3, and were caught offside ten times. The error that night was reading the result as proof about the process. The process was fine; the result was variance.

The contrarian risk

There is an uncomfortable side I should state against myself.

If the gate is too strict, nothing ever passes and the analyst retreats into permanent caution — never claiming, only qualifying. That looks like rigour and is actually indecision. A betting market requires a prior; declaring uncertainty forever builds no model.

A second discomfort runs deeper. We say a claim fails the public record, but the public record is not always reliable. Old scorecards carry wrong bowler names, wrong catch attributions, occasionally wrong results. The gate therefore audits two things at once: the claim and the source. The first is easy. The second forces the admission that our foundation may also be eroding.

"The model said one thing; the empty stadium said another."

That sentence taught me in 2026 that a number loses meaning without context. The reverse reading matters equally: an empty stadium is also a model. Whether home advantage comes from umpiring tendency, pitch preparation, travel fatigue or familiarity is not answered by empty seats alone. It is answered by match-level data.

A third discomfort is defending my own model. I like writing about Bumrah's 32 wickets because the number is beautiful. If I use it to claim he is equally effective in every condition, I am not an analyst but a supporter. Perth swung; Melbourne swung less. Admitting the difference is not weakness. It is method.

The next-round signal

That night I did not close the packet. I broke the three claims into three questions, wrote a limit for each, and fixed the conditions under which each claim would be withdrawn.

For anyone building cricket models in the next cycle, one question remains. When a number conflicts with the match you watched, which one do you call wrong — the number or your eyes? The answer depends on whether you know the denominator, whether you have matched the era, and whether you recognised the proxy. The analyst who can answer those three will not win arguments. They will simply make fewer errors, and over a long season that is the only measurable edge.