Three Numbers That Didn't Match the Scorecard: Cricket Analysis's Verification Gate

The fan in the small west-side room at Bayuemas Oval was the loudest thing in the building. On the table lay a fourteen-page printout, three highlighters and three sticky notes. Each note carried one number: a right-arm seamer's death-over economy, a head-to-head line between two teams, and a date of birth. All three were marked verified in the packet. All three fell apart the moment I opened a scorecard. No player is named in this piece. The point is not a person; it is a process. The step every analyst rushes through is the step that breaks most often. A data packet rests on three kinds of claim: who played, when they played, and how the thing was measured. Bend any one of them and the most elegant analysis downstream is standing on a crooked floor. I found the story in a notebook margin, sitting beside a scorer's table with a pen and a stopwatch. Twenty-five years on this beat has taught me that the gap between the scorecard and the ground is where cricket's real reporting lives. Every tournament analysis has two layers. The first gathers facts: who did what, in which match. The second measures: who is better, who is worse, who holds up under which condition. A fault in the first layer does not stay small; the second layer magnifies it. That is why verification gates exist — before analysis, flag what fails the cross-check rather than quietly using it. Follow the data's journey and you can see where faults are born. A scorer writes on paper. The paper enters a tournament management system. The system feeds an aggregator's database, which feeds an analyst's spreadsheet, which becomes the packet. Something is lost at every handover. A wide logged in the seventeenth over arrives in the system as the eighteenth; at the back end it is charged to a bowler's account. One over of drift changes an economy rate and nobody notices. Tournament cycles make time the scarcest resource. Eleven days, fourteen matches, three venues, a fresh packet every morning. Verification is the first thing cut. Claims that look sweet at a glance survive; claims that require opening a scorecard hang in limbo. In Associate cricket the problem sharpens. Malaysia's domestic structure is small and full-time scorers are few. Bangladesh has volume, which means more errors but also more people to catch them. Malaysia has fewer matches and fewer hands, with the same expectations. A packet built in Kuala Lumpur passes through three or four pairs of hands, and spelling and definition shift at each one. The public record is not a clean baseline either. Scorecards are amended later, and the amendment is rarely announced. An archive keeps the old figure; new writing cites the old figure; the error settles into the look of fact. Cross-checking therefore means more than matching two numbers — it means knowing which one was updated and when. Three questions do the work. Where is the match list? What is the cutoff date? What is the definition — what counts as a run, what is the denominator of a strike rate? Without those answers a number is not a number but an estimate wearing a suit. The consequences are not small. A wrong economy rate enters selection debate, opposition planning, a reporter's copy, and finally a player's reputation. Cricket statistics are no longer decoration; they are the language of professional decisions. The first failure is window drift. A packet claims a bowler's death-over economy over twenty-four months. Open the match list and half of it is franchise cricket, where he bowled the seventeenth and eighteenth overs, and half is international cricket, where he bowled the nineteenth and twentieth. Different fields, different batter intent, different pressure. Blended into one column, the number describes nobody. Drift has subtler forms. ICC event warm-ups are unofficial and never enter rankings or career records, yet many feeds label them as T20s. Anyone pulling from a feed can absorb spells that exist in no official ledger. Rain-shortened matches are the same trap. In a chase cut to fifteen overs by Duckworth-Lewis-Stern, both the bowler's workload and the batter's intent change; placing that spell beside a full twenty-over spell without a flag makes the comparison meaningless. The second failure is definitional. Economy is runs conceded divided by overs. But which runs — off the bat, or including wides and no-balls? Some aggregators count only runs off the bat, and the same spell yields two different economies. Strike rates carry the same fog: who treats not-out innings differently, which feed counts wides as balls faced. Here is my real objection. Cricket now packages certain metrics as effort — dot-ball percentage, pressure index, impact score. They look superb because they can be bent to produce any conclusion on any surface. A bowler defending on a slow pitch, bowling a defensive line to a batter content to bat long, will post a beautiful dot-ball percentage while contributing almost nothing toward taking a wicket. Pointless running produces pretty numbers; so do pointless dot balls. A metric whose weighting nobody can explain is not analysis, it is upholstery. The third failure is the most irritating because it happens without malice: entity resolution. Mohammad, Muhammad, Mohamed — three spellings, three profiles. Bangladeshi names with multiple parts leave feeds disagreeing over which is the surname. Names of Malaysian players of Indian or Sri Lankan descent shift again in transliteration, and a small difference in spelling splits a record in two. Harder still is a change of association. A diaspora player moves from one domestic structure into another country's national side, and the earlier record does not fully follow. Part of a career vanishes between two pages of a database. Age-group records are worse, with the same name sitting under different birth years in under-19 and under-16 ledgers. A fourth failure goes unmentioned: the silent correction. A scorecard is fixed a month later, with no announcement. The old figure has already entered three articles, two podcasts and a team meeting. The corrected number is now the stranger, and the world is confident about the old one. The error does not merely persist; it acquires authority. Strangely, more data has not cleaned packets. It has done the opposite. A mistake once stayed in one notebook. Now the same mistake spreads across five dashboards in a day, each presenting it with greater confidence. More information is not more certainty; it is faster distribution of error. I verify with an old three-column method. The claim goes in the first column. The source — which scorecard, which feed, who entered it — goes in the second. The match set goes in the third. If I cannot write the match set, the number does not enter the packet. That habit has proved more valuable than any stopwatch. A good packet is recognisable by four things: a full match list, an explicit cutoff date, a definition for every metric, and a page headed 'what we could not verify'. That last page matters most. A packet honest about its gaps earns trust for everything else. Now the counterintuitive part. We assume verification failure is a technology problem — better databases, better software, artificial intelligence, and it will be fixed. My experience says otherwise. It is a labour problem. The person writing on a scorer's sheet at a remote ground, the volunteer uploading results at midnight, receives no funding. Money flows to the dashboard, not to the pencil. Investment goes to the layer where errors look attractive, not the layer where they are born. A second reversal is less comfortable. The packet is not always wrong and the public record not always right. When a scorecard is amended, a fresh packet can look wrong against a stale public page. Cross-checking needs a direction: who is being checked against whom, and which figure was updated when. Still, this reversal must not be overplayed. Not every mismatch is a conspiracy; some are typos. One test separates them — does the error have a direction? If mistakes repeatedly favour the same player or team, it is a process failure. If they scatter, it is carelessness. Failing that test, no accusation is fair, and the process can still be questioned without accusing anyone. What should you watch next? Ask three questions of any cricket number: which matches, which cutoff, which definition. An analyst who answers all three has earned a hearing. One who dodges them is offering an estimate, however dazzling. Then watch the next data release. Is the corrected figure published under a correction label, or quietly swapped? The first is professionalism; the second is a signal that older errors may still be hiding in the packet. The beat keeper hears what the highlights delete — but who hears what the scorecard silently rewrites? One question lingers. A number that cannot survive the walk to the scorer's table — what exactly is it doing in a selection meeting?