Cricket's Biggest Risk Isn't Bad Data — It's Invented Data

On 19 November 2026 in Ahmedabad, the Narendra Modi Stadium scoreboard read India 240 after fifty overs. Australia fell to 47 for 3 chasing 241. Normally, that is the end of the match. Travis Head made 137, Marnus Labuschagne finished 58 not out, and Australia reached 241 for 4 in 43 overs.

Most of what has been written about that final since ends on the same line: the data was wrong. I do not accept that. The data was right. The people reading it were lazy.

Here is the more uncomfortable part. The tournament's leading run-scorer, Virat Kohli with 765 runs, and its leading wicket-taker, Mohammed Shami with 24 wickets, both records for a single World Cup, were sitting in the losing dressing room. Two all-time records, one team, and that team lost the last match.

That is where the real story begins, because the problem that landed on my desk weeks ago does not show up on a scorecard. The xG autopsy began where the broadcast stopped and the silence started.

India had won all ten matches before the final. Home venue, home pitch, three spinners, Bumrah and Shami with the new ball. Conventional analysis called it a perfect setup. Australia had lost their first two games, to India and then to South Africa by 134 runs, then won nine straight to reach the final.

For people who forecast from numbers, Ahmedabad was a test. India's tournament aggregates were monstrous: Kohli's 765, Rohit Sharma's 597 at a strike rate above 125, Shami's 24. Australia's aggregates were far less glossy. So what happened?

Let me be clear about where I am standing. I am writing this in the middle of a transfer window, where cricket's news market is drowning in rumour — which club is signing whom, which agent is calling whom at midnight, whose contract has a release clause buried in it. The reader does not need another forecast. The reader needs a reliability filter: which piece of information survives the week and which dies by evening as a midday rumour.

That demand is exactly where the largest gap has opened.

There are two separate failures, and people merge them. The first is bad interpretation: correct numbers, wrong reading. The second is invented information: a number with no parent. The first is uncomfortable. The second is dangerous.

Ahmedabad is the first kind, and three numbers explain it.

The first is 240. India's score. On that pitch, in that light, batting was supposed to get easier. Instead the surface slowed, and 240 was below par. Nothing in India's tournament-long scoring rate connects to that total. That is the first crack: an aggregate strike rate is an average across eleven innings; a final is one innings, one day, one pitch.

The second is 47 for 3. Australia's position. Three down chasing 241 theoretically hands the momentum to the hosts. But something happened there that no tournament aggregate can capture: a decision by a partnership under pressure. Head attacked. Labuschagne absorbed. They added 192 for the fourth wicket. That is method, not luck.

The third is 192. Inside those runs sits a specific captaincy choice from Pat Cummins: win the toss, field, and use the new ball to pin India's top order to the surface. For anyone who dismisses the toss as pure luck — the toss is a coin, but the fifty overs after it are a decision.

So why did Kohli's 765 not carry into the final? Because 765 is a strength and a final is a sample. Turning strength into a sample is not statistics; it is conditions, pitch age, and the opposition's plan. Anyone who predicted the final from 765 had translated a tournament average into a single-match probability. That is not translation. That is assumption.

A number cannot be wrong, but a number without a guardian stops being a number — it becomes a word.

Now the second failure, which I think is the biggest risk in cricket journalism today.

A few weeks ago I ran an analysis pass on a cricket article. What came back was a single domain label: cricket_world. Nothing else. No title, no source, no author stance, no information points, no names, no date. The system had correctly stamped the category and never read the text.

What sat in front of me was a nine-section template: format analysis, player technique, team landscape, league economics, governance, risk, public narrative, industry transmission. Every field had a slot. Not one field had anything to fill it with.

The honest answer was to write the same sentence in every box: there is no verifiable information here, so nothing can be assessed. That was the only professional answer available.

The market has no price for that answer. The market wants a filled template. And the cheapest way to fill a template is to reach into a model's priors and invent — a team name, a fee, a quote. The output looks tidy. It has no parent.

The new definition of failure is not the wrong number. It is the parentless number.

In a transfer window this failure spreads fastest, because rumour fuel is cheapest here. A name, a fee, and 'sources say' is enough to build a story. And the machine filling the template cannot tell which source was real and which was somebody's guess posted on a forum last Tuesday.

From my own experience I have built three tests I apply to every claim.

The first: who is this number's parent? Who said it first, when, and did they see it themselves or hear it from someone? Second-hand numbers are cricket's most common contamination.

The second: does the format tag match? Blending Test, ODI and T20 figures is treating three different sports as one. A bowler's T20 economy does not prove an ODI weakness. A number without a format is meaningless.

The third: can the sample carry the claim? Declaring a batsman's age-curve decline off three innings is the most common dishonesty in cricket analysis. Home averages hide away weaknesses. You do not need a model to know that — one away tour with the lower order tells you.

I learned these three the hard way.

In 2026, after Sydney FC beat Melbourne Victory on penalties in the A-League Grand Final, I posted a twelve-tweet thread. Sydney had 0.9 xG, Victory had 1.4, and Sydney scored from a set piece. I wrote that Sydney's dynasty was variance, not dominance. Four thousand retweets, a paid column, and a lesson: the number is the easy part. The hard part is writing down what evidence would break your own claim.

That lesson worked the next year in Kazan. On 30 June 2026, France beat Argentina 4-3, and I argued that Mbappe's two goals exposed Argentina's back three rather than simply showcasing his speed. Then I spent three days at Croatia's training base watching their 4-1-4-1 press and predicted they would reach the final. Presence on the ground corrects the data; the data does not correct presence. I packed for Russia in four hours and unpacked my assumptions for years.

In 2026 I made my English-language commentary debut on the Bangladesh women's ODI series against India. In a commentary box you have one over to be right, and it teaches you, politely, never to let an unverified line leave your mouth.

So I keep a rule: my receipts ledger and my draft stay separate. The ledger records who said it, when, and how I checked. The draft holds the story. Mix the two and the story starts to sound like a receipt while not being one. That is the most dangerous kind of lie.

The second rule is the two-stage rule. Publish the fast dispatch, then follow with the audit. The fast piece exists to survive the seventy-two-hour news cycle. The audit exists to survive the following week. Two different pieces, two different responsibilities.

Now, where I could be wrong.

First, invented information may be a transitional problem. Verification tooling is improving quickly, source tracing is being automated, platforms are adding labels. If a batch of a hundred AI-assisted cricket briefs passes a source-traceability audit at above 95 per cent in the next twelve months, I will withdraw the claim.

Second, and more uncomfortable: perhaps the real risk runs the other way. Perhaps the damage is done less by invented information than by excessive trust in clean information. Ahmedabad is the proof. The cleanest dataset in the world pointed at India, and the match went elsewhere.

Third, perhaps 'information gain' is itself the disease. The pressure to give readers something new pushes us toward the exotic claim, and exotic claims are not always true. The boring truth gets no traffic.

Fourth, my null-extraction receipt may simply be a bug in my own pipeline rather than an industry symptom. One receipt is not a pattern. I take that objection seriously, which is why I am filing this as a warning rather than a law.

Still, the fear stays. A system that stamps the category but never reads the text is not a cricket problem. It is our problem.

My predictions are two, and both are testable.

First, within twelve months at least one major cricket outlet will publish a correction for a fabricated statistic or a fabricated quote in AI-assisted copy, and the correction will come from reader pressure rather than an editor's audit.

Second, at least one cricket board or league will add a source-disclosure requirement to media accreditation, the way financial journalism already has.

And the last question is for you. When a number arrives in your feed with no parent, do you stop, or do you share? Bad data has always been with cricket. The new thing is parentless data. Parentless data is never disproven. It just spreads.