Numbers Without Provenance: Auditing Asian Cricket's Statistic Protocol

The Empty Ledger

The file that landed on my desk that evening had eleven columns and all eleven were blank. No match name, no format, no venue, no player, no date, no source. One cell was filled, and it read: cricket_asia. In twenty-two years on this line, it was the first input I could not write a verdict against. Not overrated, not underrated, no percentage, no confidence band. I looked at the screen for a while, then closed the sheet.

The moment a model goes silent is the easiest moment to fake. You can guess, you can write with rhythm, and the reader will never know. I did not do that. This piece is the record of that silence — and of the protocol I built over the years so that when the source goes missing, the analysis stays honest.

The Question Is Provenance, Not Outcome

Cricket coverage has a strange inversion built into it. Everyone checks outcomes — the scorecard is public, the average is public, the strike rate is public. Nobody checks the claim. Nobody asks which window this number came from, which filter it passed through, which format it was counted in, and under what conditions it starts lying.

A statistic with no stated window, no filter and no provenance is not an argument — it is decoration. And you cannot read a match through decoration.

The reason is economic, not technical. IPL ball-by-ball data is free, clean, tagged over by over. Ranji data is thin, regional-language reporting is scattered, and the pre-2026 women's game is close to absent from the record. The result is that wherever the data is easy, that becomes the beat. The analyst goes where the light is, and then mistakes the light for the truth.

I made a rule against that habit: go deliberately where the data is bad, and make the thinness part of the finding. A label like cricket_asia is the test case. It tells you the subject is cricket and the geography is Asia. That is all. It does not tell you Test or T20, men's or women's, domestic or international, match or auction, sport or governance.

My desk runs a standing rule: every claim ships with four things — the model, the data window, the confidence band, and the condition under which the claim breaks. The rule makes me slow. It also made me the only man on that desk whose calls could still be audited a year later.

The Five Doors of the Audit

When the source is empty, the work does not stop — it changes shape. You stop writing verdicts and start writing protocol. I run every number through five doors.

Door one — format. A Test average and a T20 average cannot share a sentence. One bowler's economy rate is three different stories across three formats. An analysis that has not fixed the format has not started. That column was blank in my file, so the first door was shut.

Door two — window. "Last two years" and "career" are different worlds. Form shifts, actions change, pitches age. A strike rate without a window is a poster with no date printed on it.

Door three — filter. Does it include the powerplay? Does it include the death overs? Are the wickets that fell in already-won matches counted? Are knockouts separated? Change the filter and the ranking changes — I learned that in 2026, when I started reading a match as a sequence of prices.

Door four — venue and atmosphere. The home-ground multiplier in my register was 1.07 across six IPL seasons and more than twelve hundred matches. In 2026 the crowds left and it collapsed to 1.00. Empty stands taught me that atmosphere was a variable I had priced at zero. An analysis that does not know the venue is treating the environment as free.

Door five — luck. The toss, DLS, rain, and the decisions that made the DRS headlines. These have to be stripped out and held apart, or you will end up calling fortune by the name of skill.

Four of those five doors were shut in the empty file. So was the fifth. The only honest way left to write is to say the verdict is deferred, and to say which inputs will bring it back. Readers find that dull. I find it better than handing them a confident wrong answer.

Case One: 2026, the Sequence of Prices

In March 2026 I walked off a Kolkata daily's sports desk after thirteen years — the week the paper folded its standalone stats page. I joined a twelve-person pricing desk in Salt Lake that set odds for licensed operators in London and Valletta.

My first build was a ball-by-ball "true run value" model trained on 812 IPL innings. On 21 May 2026, at the innings break of the final in Hyderabad, it gave Mumbai Indians only a 34 per cent chance of winning. The score was 129/8 against Rising Pune Supergiant. Krunal Pandya made 47 off 38, and the match went by one run.

I spent three weeks rebuilding the death-overs wicket-value term. The lesson was not about the number, it was about the claim. Every number I trust carries a scar from a season I got wrong. The 34 per cent was not stupidity — it was the honest output of a model that had under-weighted death-over pressure.

That same year I changed how I filed. I stopped writing match reports as verdicts and began every piece with a stated prior, the sample behind it, and the conditions under which it fails. Editors got a standing "show the work" box — model, data window, confidence band.

That habit is why the blank cricket_asia sheet did not unsettle me. It was the cleanest case study of my own daily work: no input, no output.

Case Two: 2026, Retiring Possession

In June 2026 I ran the football book for the World Cup. My pre-tournament model, built on two seasons of xG and PPDA from the Bundesliga and Premier League, priced Germany at 14 per cent to fail the group stage. On 27 June in Kazan they lost 0-2 to South Korea and finished bottom of Group F.

Three days later Spain completed 1,029 passes against Russia and went out on penalties. That night I filed a 2,400-word piece arguing that possession is a lagging indicator and that PPDA measures intent far better.

In 2026 I retired possession percentage, and nobody at the desk noticed. I still repeat that sentence, because two things hide inside it. First, the metric was wrong. Second, nobody cared.

The second matters more. When you retire a metric and the desk stays silent, the temptation is to write about the silence instead. I did not. The audit is the protagonist; if the correction mattered, the correction is the story, not the silence that met it. I replaced possession with field tilt, PPDA and progressive passes per shot, and I began separating result from process in every match note. It made my tournament previews unfashionably dull and unusually accurate. Editors stopped asking for my post-match opinions and started asking for my pre-tournament numbers.

In cricket the translation is direct: we carry several packaged metrics that measure the volume of bowling rather than its quality. Dot-ball percentage is one. It does not say why the ball was dot — whether it was the batter's compulsion or the bowler's craft. Until that distinction is recorded, dot-ball percentage is decoration.

Case Three: 2026, Empty Stands and Thirty-Six

In May 2026 the Bundesliga returned to silent stadiums. I re-ran my home-advantage regression that week. The multiplier across six IPL seasons and more than twelve hundred matches was 1.07. Behind closed doors it collapsed to 1.00. I published the finding before the September UAE restart, and it held across sixty matches.

This does not mean crowds are meaningless. It means the crowd effect was exactly as large as I had measured it, and no larger. That remains my biggest lesson: when a variable is removed and the outcome does not move, change your story, not your data.

On 19 December that year, in Adelaide, India were bowled out for 36 while holding a 53-run first-innings lead. My in-play model had given them only a 26 per cent chance of losing. Thirty-six all out in forty-five minutes, and every model I trusted went silent. Josh Hazlewood took 5 for 8, Pat Cummins 4 for 21. I spent that winter logging ball-tracking for every wicket.

What came out of that log breaks cricket's oldest habit. We turn collapses into questions of character — "could not absorb the pressure", "body language dropped". The tracking says something else. Most of the wickets in that innings fell on a length above the stumps that was hard to leave, because the ball seamed off a lightly grassed pitch. That was geography, not psychology.

So I added a mandatory context-coefficient paragraph to every piece — crowds, travel, calendar, pitch age — and refused to compare pre-2026 and post-2026 numbers in the same sentence. Some find that separation over-cautious. It is not caution, it is basic hygiene.

Case Four: 2026, the Void Protocol and a Sample of Eight

On 12 June 2026, with Denmark-Finland on my second screen, Christian Eriksen went down in the 43rd minute at Parken. I pulled every Denmark-linked market within ninety seconds, then drafted a twelve-clause void protocol for non-sporting events. My firm still uses it.

Seven weeks later in Tokyo the opposite problem arrived. 3x3 basketball and skateboarding were Olympic sports for the first time, with samples of eight. I built a Bayesian shrinkage model, priced India's men's hockey bronze at 4.5, and their 5-4 win over Germany arrived on 5 August. I published the uncertainty bands alongside the pick.

That experience changed how I write. I now put the uncertainty inside the sentence — "61 per cent, give or take nine" — rather than hiding it in a footnote. Readers resented it for a year and then quoted it in every dispute.

In cricket the same problem arrives with every emerging dataset. The women's ball-by-ball corpus is not yet as dense as the men's. That does not mean the women's game cannot be analysed. It means it must be written with wider bands, and the bands must not be hidden. Those who apply the men's benchmark to the women's game are not analysing — they are comparing, and a comparison across two different samples is two different worlds.

Case Five: 2026, the Variance Ledger

On 22 November 2026 Argentina led Saudi Arabia 1-0 at Lusail and my in-play model had them at 94 per cent. They lost 1-2. I published the post-mortem within forty hours — not an excuse, a variance ledger with the model's calibration error measured across five thousand historical matches.

The same week, my pre-tournament note on FIFA's added-time directive had already run — 27 minutes were added in England-Iran. For the rest of the tournament I wrote about Morocco's PPDA instead of the favourites' storylines.

That produced a structure. I launched a public variance ledger — every flagged edge, won or lost, scored quarterly. It pulled my copy away from hot takes toward repeatable claims. I stopped printing single-number forecasts; every prediction now ships as a distribution with the tail case named. Two editors dropped me. The readers who stayed were the ones who paid.

The Contrarian Angle: Correlation Is Not Causation, and Glossy Data Lies

This is where I slip most often.

The real danger is false precision. Models print decimals, and decimals read as authority. A 51.3 per cent edge sounds like knowledge when it is a coin with a story attached. The fix is plain: publish the interval beside the point estimate, and write the sentence that says what the number cannot tell you.

The second danger is data-availability bias. We write where the data is cheap. That produces a quiet asymmetry in cricket coverage: fifty thousand words a week on the IPL, five hundred on the Ranji Trophy, and almost nothing on domestic women's cricket. That is editorial laziness, and the cure is to write deliberately where the data is bad.

The third danger is the "nobody noticed" grievance. When you retire a metric and the desk stays silent, the temptation is to write about being ignored rather than about the metric. I have felt that pull many times. I have suppressed it every time, because the reader you want buys your correction, not your complaint.

The fourth danger is contempt for the commentary box. I won the tactical argument years ago and the box still talks in milestones. It is easy to slide from auditing the claim to dismissing the person making it. My rule is firm: audit the claim, not the speaker. A wrong number on television is a dataset problem, and the fix is a better number, not a better sneer.

This is also where correlation and causation separate. One statistic did the rounds this year: teams that ran more distance won more matches. It is true and it is useless. The team that is chasing the game runs more, because it is running after the ball. The relationship between distance and victory is inverted, and what we sell as an "effort metric" is often a description of defeat. Pointless running also produces pretty numbers.

Selection as Version Control: The Problem Is the Ranking Machine

If I had to pick one structural weakness in Indian cricket, it would not be a shortage of talent. India's real problem is not talent but the machinery that ranks it. The hardest team in world sport is India's second XI, and there is no public, auditable protocol for selecting it.

This is where the IPL auction matters. For me it is not carnival content but a price-discovery system. In an eight-hundred-million-dollar market, franchises were doing what data desks later claimed to invent — probabilistic selection. Coverage still prefers leaks to ledgers. One line sits in my register: the transfer fee is a rumour, the wage bill is a fact, and the sell-on clause is a confession. An auction price does not tell you a player's value; it tells you the market's opinion, and the market's opinion has to be corrected every year.

And the dataset we assimilated last is the women's game. The Women's Premier League from 2026 and the 2026 home World Cup win changed the sport, but the commentary establishment acknowledged them far faster than it modelled them. Names like Harmanpreet Kaur and Smriti Mandhana are everywhere, yet the data scaffolding behind them is not as dense as it should be. That is a sporting-political failure, and it will shape selection for the next decade.

In the same way, the "India-Pakistan rivalry" we sell at every tournament has not existed in bilateral cricket since 2026-13. A fixture that is not played cannot be the basis of analysis. It is a marketing product, and I have no habit of understanding cricket through marketing.

Takeaway: What to Watch Next

The empty file was not a failure for me. It was a reminder that the most honest state of analysis is the one in which it says — not yet.

Over the coming months I will track three things. First, whether Asian cricket coverage starts printing a number's window and filter alongside it — a small change with a large effect. Second, how fast the women's ball-by-ball corpus thickens, because the next selection argument will be fought there. Third, whether we go back to treating crowds as a variable, or remember the lesson of 2026.

There is a version of this game that only the numbers saw, and I owe it a hearing. A hearing does not mean the number wins. It means the number gets to state its source, its window, and the condition under which it fails.

The day the filled sheet arrives, I will write the full verdict — with the prior, the sample and the band. Until then, I will let the empty ledger stay empty.

Count first. Believe second.