From 36 All Out to the WPL: An Audit of Indian Cricket's Data Crisis

The afternoon of 19 December 2026, at the Adelaide Oval. India hold a 53-run first-innings lead. My in-play model gives them a 74% chance of not losing — 74, give or take nine. Forty-five minutes later, India are 36 all out. My model went silent. After that day, I started logging ball-tracking for every wicket that fell, through the end of the season, every single one — not so I could blame a player, but so I could rebuild the run-value term that had lied to me. What I found was not the secret of any one coach or any one batsman. It was the failure of a pipeline, and that pipeline's name is: selection.

In this tournament cycle, when everyone is swept up by flag and story, a different question dogs me, one more urgent than any formation or squad-depth truth: India has the world's richest cricket dataset — IPL ball-by-ball since 2026 — and yet the selection of its second XI is still done the way it was in 2026. This piece is an audit of that gap. I am not making a prediction; I am opening the ledger — premise, sample, window, confidence band, and the conditions under which it fails.

Context: When the system never recognised the data

The IPL began in 2026 and unknowingly erected an experimental system. Franchises started buying players at auction — and that too when data desks did not exist. In 2026, when I left a Kolkata daily's sports desk for a twelve-person pricing desk in Salt Lake, my first build was a "true run value" model trained on 812 IPL innings. I realised that what I was building as a new model, franchises had already been doing in the auction hall, without a model's name — as intuitional probability. In 2026 I retired possession percentage from my vocabulary and nobody at the desk noticed — the same lesson applies to Indian cricket's selection. The metric we are proud of is often a lagging indicator.

Now let's talk about that structure. India's main team is among the strongest in the world in any format — that is not up for debate. But the real problem lies where the camera does not go: Ranji Trophy, Vijay Hazare, and now the WPL. Here the data either does not exist, or is not curated in English, or has a sample so small it cannot support any decision. When I say India's real problem is not talent but the machinery that ranks it, I mean: we never model the dataset we have, and the one we do model does not show us.

The WPL began in 2026, and in 2026 India won the home World Cup. The commentary box acknowledged it quickly — but in modelling it remains unassimilated. I have the full ball-by-ball of the WPL, but when I try to map it against benchmarks from men's domestic tournaments, data-availability bias surfaces: the men's data is rich, the women's thin. That thinness is not a side-note, it is itself a finding. I built a small model on the death-overs data from the first two WPL seasons — 61%, give or take nine — and it is necessary to say what that nine cannot tell you: it does not capture travel fatigue, pitch age, or adjustment to empty stands.

Core analysis: Selection is version control, and we are not updating it

The night of 21 May 2026, Hyderabad. The IPL final. Mumbai Indians collapse to 129/8 in the first innings. My model gives Mumbai only a 34% chance of winning at the innings break — against Rising Pune Supergiant. Krunal Pandya scores 47 off 38, and the match goes down to one run. I spent three weeks rebuilding the death-overs wicket-value term. That night I learned that a model is not an event but a document — it must be versioned every season.

The same lesson does not apply to Indian selection, and that is my real point. India's second XI is the hardest team in world sport — that is not a cliché, it is a structural fact. Whether it is bowling depth, batting alternates, or the all-rounder list — India has at least three credible options for every slot. Yet the selection decision is settled by television time rather than sample size. The difference between a batsman who scores 900 runs in a Ranji season and one who hits four sixes off 14 balls in an IPL is not measured by anyone in the language of data — because the Ranji data is not curated in anyone's hands.

I apply here a reliance-test I apply to every report. If someone says "give the young player a chance," I ask: how many domestic matches' data, on what pitch, against what bowling attack, and what year is that data from? If the answer is only IPL highlights, then it is not an argument, it is decoration. And I do not use decoration.

Take one concrete example. In 2026, during the Tokyo Olympics — 3x3 basketball and skateboarding were debuting, samples of eight matches — I had to build a Bayesian shrinkage model. I priced India's men's hockey bronze at 4.5, and on 5 August it came good with a 5-4 win over Germany. But I printed the uncertainty band alongside: 4.5, give or take 1.2. The same method should apply to selection. Any selection claim — "this guy is ready" — should be published the same way: what is the premise, how big is the sample, and when will it fail.

Contrarian angle: Correlation is not causation, and selection data is caught in the same trap

In 2026, during the Russia World Cup, I ran the football book. My pre-tournament model gave Germany a 14% chance of failing the group stage — on 27 June 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. I filed a 2,400-word piece arguing that possession is a lagging indicator and that PPDA measures intent far better. I retired possession from my vocabulary and replaced it with field tilt, PPDA and progressive passes per shot.

Indian cricket's selection data has the same problem. Metrics like "strike rate" and "average" are convenient, but they do not capture the context-coefficient: crowds, travel, calendar, pitch age. In this subcontinent the sentiment-amplification coefficient is very high. Two good matches by a young star on TV build a narrative, and the crowd starts calling him "the next big thing" — while his sample is 90 balls. I decline any conclusion that explains a series from one innings. If the sample is small, I will say it is small, and I will defer the verdict — I will not dress a hunch as a finding.

One sharp point here. The India-Pakistan rivalry, which the commentary box makes the frame of every series, has not been played bilaterally since 2026-13. It is a marketing artifact, not an explanatory tool. That does not mean it is not a big match — it means stop drawing tactical or data conclusions from it.

Back to the data. In Qatar in 2026, Argentina led Saudi Arabia 1-0 and my in-play model had them at 94%. They lost 1-2. I published the variance ledger within forty hours — the model's calibration error measured across 5,000 historical matches, not an excuse. The same discipline should apply to Indian selection: every selection claim, win or lose, should be scored quarterly. When India were bowled out for 36 in Adelaide, my model was wrong — and I did not hide it, I rebuilt it.

Takeaway: The next-round signal

In the Wi-Fi age, when every ball generates data, Indian cricket's biggest risk is not that we lack data. The risk is that we have the richest dataset and our selection machinery still does not know how to read it. After the 2026 WPL win, the question that should be asked is not "is India ahead in women's cricket?" — the answer is yes. The question is: have we translated the WPL's ball-by-ball data into selection decisions the way we translated IPL data into auction prices? If the answer is no, then even our biggest dataset will remain just decoration — and I do not use decoration.