Where the Number Is Right and the Label Is Wrong: Three Failed Cross-Checks in Cricket Data

At 1:15 in the morning three rows on the monitor are red. The numbers from the Stage-1 packet are internally consistent — no typos, no rounding errors, units align. The problem is that none of the three matches the public record. One number was born in the right match but sits inside the wrong time window; one was built around the right human being, but two different players have been filed under the same name; one is perfectly accurate, yet the model version that produced it has since changed its definitions. The gate closes. The analysis stops before it starts.

Such nights are not new to me. In 2026, coding forty Bangladesh Premier League matches by hand into a spreadsheet at a Rangpur desk, I learned that cricket's errors rarely live in the number. They live at the boundary of the number — between over 0 and over 1, in the decision about which over owns a free hit, in whether a wicket ball belongs to the bowler or the batter. I began at a Rangpur coding desk, then let Russia's quiet teach me the rest: without a timestamp, no number is trustworthy.

Context: five hands between packet and article

A Stage-1 packet is a pile of raw material — ball-by-ball feeds, scorer sheets, newspaper archives, a few video timestamps. Stage-2 builds structure out of it. Between the two sit five hands: raw feed, scorer, aggregator, analyst, writer. Meaning can slip at every handover.

I first found that idea in Russia in 2026, studying France's 4-2-3-1. In the final they conceded 66 percent possession but only 0.8 open-play expected goals — two numbers that are meaningless unless read together. In 2026, coding ninety-two behind-closed-doors Bundesliga matches, I found the home win rate had fallen from 43.2 percent to 33.3 percent while away high turnovers rose eleven percent. When the stadiums emptied, I stopped listening for noise and started measuring silence — and that is precisely when I learned to attach a timestamp and a definition to every figure.

Cricket makes this discipline harder, because the game is itself an arithmetic exercise: overs, balls, runs, wickets, Duckworth-Lewis. A football match is ninety minutes; a cricket match can be forty overs, twenty, or fifteen. Every truncation creates a new denominator. And tournament cycles are where truncation happens fastest and messiest.

Core: the anatomy of three failures

Window mismatch. The first item looks immaculate — a bowler's powerplay economy. The number is true, but two rain-shortened matches have slipped inside it, where the powerplay was four overs rather than six. The runs-per-over arithmetic is correct; the word 'powerplay' is not. The number is not lying. The label attached to it is. The same trap waits in death-overs scouting. In a match where the innings ended in fifteen overs, overs seventeen to twenty do not exist — yet the aggregator files every ball of that match under 'death'. You have just identified a death specialist who never bowled in that phase.

Entity mismatch. The second item fails at the level of the name: two players sharing a surname, one player under two registrations, or a match later re-scored after a DLS recalculation, a code-of-conduct penalty, or a third-umpire reversal. The old figure survives on the internet; the corrected figure only enters the primary feed. Counting a no-result inside a head-to-head record is the same family of error. Cricket's most dangerous data is data that was once true.

Model version drift. The third item is the subtlest. Expected runs, expected wickets, win probability — these models are recalibrated every season. A 2026 figure and a 2026 figure cannot share a scale. Both carry the word 'expected', so readers assume they are comparable. Nobody is at fault, and the error still happens.

Against these three I use a four-step audit. Provenance chain: name and date every hand the figure passed through. The two-clock rule: record the match clock (which over, which ball) and the calendar clock (which date, which season) separately; one cannot substitute for the other. Denominator check: list the balls underneath the number. Definition lock: state the model version at the moment of writing.

The most important lesson of those four steps is that the error almost never lives in the value; it lives in the meaning. An aggregator placing 6.2 economy beside a bowler's name has not lied. If I sell that 6.2 as proof of a powerplay specialist, the lie is mine.

The loan fit report I published alone in January 2026 on Leicester City's move for Tete from Shakhtar Donetsk worked because the denominator was clean — 2.8 dribbles per ninety, inside a defined league-minutes window. Had cup minutes leaked in, the number would have stayed right and the conclusion would have gone wrong. The same discipline applied to Morocco's 4-1-4-1 at the 2026 Qatar World Cup: Sofyan Amrabat's 12.3 kilometres in the quarterfinal against Portugal is a meaningful figure because it is bound to a specific match, a specific role, a specific opponent.

Now add tournament pressure. Short cycles deliver packets faster and editorial demand faster still. Three matches in a day, three venues, three pitch conditions. The scorer is tired, the feed is late, the aggregator pours new data into last season's template. Under those conditions, window and entity mismatches multiply.

Contrarian: the aggregator is not the culprit

The instinct is to blame the aggregator, or to change the feed provider. In my experience the real fracture is the deadline. A packet that ships before verification makes the existence of a gate irrelevant. There is a second danger: a gate that checks numbers but not definitions manufactures clean lies. The figure is proven correct, the source looks reliable, and the meaning is still wrong.

Honesty is needed elsewhere too. Records are revised after umpiring reviews and DLS recalculations. Corrections in big matches get reported; corrections in small matches get buried. That produces an asymmetry in the public record — more scrutiny on large teams, less on small ones. This is not a conspiracy; it is the economics of attention. A record built against a smaller side can stay quietly wrong for years, and later become the foundation of analysis.

Fielding metrics carry the same trap. Runs saved only means something when you know how many balls entered your arc. Nobody saves runs on balls that never came, yet the table shows everyone equally.

Takeaway: what to watch in the next match

The next packet arrives within days, and with it three new red rows. That is normal. My advice is simple: before trusting a number, ask which balls sit underneath it, then ask which model version produced it. Data without a pitch is noise; a pitch without data is a missed chance. The only question that survives is this — when the tournament ends, which record will we trust: the one in the feed, or the one that survived verification?