Without a Format Tag, No Cricket Number Means Anything — And That Is the Most Ignored Flaw

The sheet arrived at the production meeting with sixteen columns. Fifteen of them were filled — average, strike rate, economy, runs in the powerplay, runs conceded at the death, home-away splits, a twelve-month trend line. The sixteenth was blank: format. Nobody asked, because a blank column doesn't look like a technical fault; it looks like a place no one bothered to look. The sheet cleared every check.

That same week the same strike rate was used twice — once against a T20 benchmark, once against a Test benchmark. It fitted a story both times, because a number sitting without its unit does exactly what a price sits doing without its currency: it enters every comparison and gets verified in none. This is the central omission in cricket analysis, and it does not happen when a wrong number arrives. It happens when a necessary tag goes missing.

In March 2026 at Eden Gardens I was twenty-eight, logging scores on the night desk of a Delhi daily. Harbhajan Singh took thirteen wickets in that Test — 7/123 in the first innings, 6/73 in the second — VVS Laxman made 281, Rahul Dravid 180, the two put on 376, and India won by 171 runs after following on. My editor wanted four hundred words on Laxman's concentration. I filed a diagram and argued for the space. I had learned to read the space before I read the score, and from that night my sentences stopped saying 'he scored 281' and started saying 'he owned the corridor between point and cover'.

Cricket has four principal formats and each carries a different risk-reward structure. The ICC keeps three separate tables — Test, ODI, T20I — because the same batter has two different biographies. Suryakumar Yadav's T20I strike rate has sat above 160 and the same batter settles around an ODI average of 25. The wrists are identical, the shots are identical, the grade changes because the unit changes. Cricket's data architecture is dense now — phase splits, match-ups, dot-ball percentages, boundaries per over, NRR, DLS. Underneath all of it sits a layer the industry treats as housekeeping: metadata. Format, venue, toss, phase, sample size, source quality. When that layer drops, the structure above stays standing and starts lying.

In February 2026, inside the Delhi Daredevils auction room, the problem was clean: a five-million-dollar purse, fourteen league matches, and no historical data for pricing a role. In that room I stopped counting bids and started counting incentives. We built a role-scarcity sheet that priced 'death overs' and 'powerplay enforcer' separately from batting average. We bought Glenn McGrath, Mohammad Asif and AB de Villiers, finished fourth, and lost the semifinal to Rajasthan Royals by 105 runs on 30 May 2026. The lesson was about tagging, not numbers. Every auction room has a second clock running — a clock that measures role, phase and format rather than reputation.

Every cricket number arrives with a unit, and that unit is the format. Strip the unit and what is left is not data; it is advertising. An economy of nine is ordinary in T20, expensive in ODI, close to meaningless in Test. The question is never 'how much'. It is 'for which format, which phase, on what sample'.

Over recent months I have been handed analysis documents that were technically complete and substantively empty. Eight sections, sixteen checklists, 'insufficient information' in every cell. Such a document passes schema validation, so it travels onward. That is the silent null: empty but well-formed. And then the next layer feels pressure to produce something, and the easiest 'something' is a plausible name, a plausible number, a plausible price. In cricket terms, we argue about a strike rate for forty minutes and never once ask which format it belongs to.

19 December 2026, Adelaide: India bowled out for 36, their lowest Test total. I had forty-eight hours of noise to work against and I wrote the inverse of it — a batting failure, not a structural one, because the bench already covered the role gaps through Mohammed Siraj, Navdeep Saini, T Natarajan and Washington Sundar. 19 January 2026, Brisbane: India chased 328, Rishabh Pant 89 not out, the series won 2-1. That was not prophecy; it was a ledger with dates and role contingencies attached.

19 November 2026, Ahmedabad: India 240, Australia 47 for 3, then Travis Head's 137 and a 192-run fourth-wicket stand with Marnus Labuschagne, and a six-wicket win. My tournament model had rated India's middle-overs control the most robust structure in the competition — ten wins, then one match that ignored the structure entirely. I rebuilt the variance layer and now publish confidence intervals alongside calls: 'India control 68 per cent of the middle overs', not 'India will win'.

The industry's standard complaint is that analysts over-model. My objection runs the other way. The real failure is not in the model but in the column that was never filled. The step that was meant to extract entities read from an empty artefact, and the step that was meant to grade source quality depended on a field that was blank. That is not a modelling failure; it is an upstream failure. And a second error attaches to it: 'no risk identified' and 'no risk exists' are not the same sentence. The absence of a visible risk in an empty dataset is an absence of evidence, not a guarantee of safety. In cricket that distinction is the difference between a report and a silence. A match with no recorded phase data did not pass unexamined; it went unexamined.

In three decades I have imported plenty of frameworks — auction theory, portfolio logic, football pressing blocks. Each time the translation's failure points have to be published first. Cricket does not behave like a market once weather arrives: DLS rewrites a target mid-chase, the toss carries a very different weight across the first two sessions of a day-night Test, and a spinner's four-over quota is not commensurable with a first morning. Tactics are the patterns that survive contact with money, weather and ego. A pattern that collapses on contact will look elegant on a slide and save no runs on the field.

So I will pre-commit in print to what would change my call. If a franchise scouting department shows me a model that priced a role correctly with no format tag, no venue and no phase metadata, I will reverse my position and the correction will stay on the record.

The next time a strike rate appears on screen, the first question should not be 'how much'. It should be: which format, which phase, which venue, what sample, and what is the error band. A number that cannot answer those five is a graphic, not evidence. And where the ledger says one thing and the stands say another, the stands deserve to be logged with their own reliability rating — the crowd is data too, it just runs on a different format. The open question is who fills the next column: the person who built the sheet, or the person who forwarded it unread.