After Counting 538 Deliveries: Why the Scorecard Lost My Trust and How to Read Price Versus Value in the IPL Auction

March 2026, Johannesburg. Australia made 359/2, India folded for 234 — a 125-run defeat. That night I did not read the scorecard. I counted deliveries. Eleven nights, a borrowed laptop, and 538 deliveries logged by hand, because no vendor in India sold ball-by-ball data back then. The ledger showed what the scorecard did not: between overs 12 and 30, India's middle order played 71 dot balls — roughly one in three deliveries produced no run in that phase. My editor ran the number as a sidebar; it out-read the main report.

Since that day I have stopped opening match reports with the result. Every piece now opens with at least one number I generated myself rather than lifted from a scorecard. That is the first rule of my method: the scorecard is a compression artifact, not a fact. It is a summary of decisions made by people you never met — who chose which over, who stood where in the field, who was rested. Until you rebuild the total from delivery-level evidence, that total is a claim, not a truth.

Context: Where the numbers come from is half the battle

Cricket's data problem is not a shortage of numbers. It is that numbers pass through the eyes of people whose names we do not know. The broadcast graphic glows with net run rate, TV commentary reads out a "pressure index," and the viewer accepts it the way one accepts a weather forecast. I have learned to treat every source — from the official BCCI scorecard to a franchise's Twitter handle — the same way: provenance first, number second.

What I did by hand in 2026, software now does in seconds. But convenience has created a new problem — people now treat data as truth because it is on a screen, without asking which lag it sits in, what sample size it rests on, and what definition built it. In 2026 I built a pressure index that, by 5 p.m., only I and one intern understood. The graphic went on broadcast, and the next day people on Twitter were quoting it as though it were an official board rule. When a model nobody believes is a modelling problem, not a PR problem. I learned that lesson that week.

The same lesson takes me to the IPL auction table. Every February, crores scroll across the screen and emotion flows at the same rate. An all-rounder gets 12 crore because he won three matches last season, while an opener whose powerplay strike rate is league-top-five goes at base price because nobody discussed his "team fit." In my reading, the auction is not a money game but a portfolio-allocation game under a fixed purse and retention rules, where prices are mispriced in certain slots and that is exactly where the edge lies.

Core analysis: How the delivery-counting habit applies at the auction table

The number that causes the most confusion at auction is "runs last season." I always rebuild it. A batter's 450 runs split into two parts: the runs made when the team had already won or lost, and the runs made at the match's leverage moment. The first part beautifies the scorecard; the second wins matches. From 2026 to 2026 I did this split myself for six seasons — roughly 1,400 innings of delivery-level relogging — and the pattern was clear: batters whose leverage-run ratio looks lower than their raw runs tend to fetch high prices at auction, while those with a higher ratio go cheap. The market buys raw runs, not leverage.

Take a concrete, publicly available example: at the 2026 mega auction, Mumbai Indians retained Ishan Kishan for ₹15.25 crore, when his T20 strike rate and powerplay scoring rate were among the league's top openers. But the question is not whether that price was right or wrong — it is whether, at that price, the role he was asked to play (top order, long innings, taking risk in the powerplay) was consistent with the rest of that purse. A franchise that spends 15 crore on one opener has less room for bowling depth. That is arithmetic, not emotion — yet at the auction table it is handled like emotion.

My own team-valuation sheet, first built in 2026 and revised every year, runs on three columns: replacement level (the cheapest available option in the same role on the market), leverage weight (the average impact of batting or bowling in the match situation), and floor (the minimum price at which buying the player does not hurt the team). When I published those three columns after the 2026 auction, an analyst at one franchise told me his team was running almost the same model — but had never put it on paper. That is the real problem in cricket analytics: teams have "experience" that is actually an unaudited spreadsheet.

The delivery-counting habit helps here too. When a franchise says "we need a finisher," my first question is: in which over? A finisher in the 16th over and a finisher in the 19th over are two different animals. In the 16th you want a batter who can find gaps against spin, because a spinner is usually bowling then and the field is in. In the 19th you want a batter who can reach the boundary against the yorker, and the field is out. I broke down the death overs of the last five IPL seasons over by over — the correlation between 16th-over strike rate and 19th-over strike rate came out near zero. Two different skills sit under one "finisher" label, and the auction sells them at one price. That is where the market is inefficient, and where scouting can substitute for the zonal selector — if the scout has a delivery-level log.

Another example I often raise: at the 2026 auction, Kolkata Knight Riders bought Pat Cummins for ₹15.5 crore, a record-level sum for a fast bowler at the time. But if you spread that purse across the whole team — Indian core, overseas quota, backup wicketkeeper — the question is not whether Cummins was worth 15.5 crore. It is where else the team compromised at that price. Every price at auction is an opportunity cost, and that cost never appears on the scorecard.

That is why I build auction-purse simulators — a tool the reader can operate himself. I believe analytics is a product, not a pulpit. The TV commentary box and the regional-language second screen still argue from memory — "I remember in 2026..." — and that argument cannot be won until the viewer has his own calculator in hand. The day a viewer distributes a purse himself at the auction table, he will understand how expensive a 12-crore all-rounder really is for a team.

Contrarian angle: Correlation versus causation, and an accounting of my own error

Here I must make public a weakness in my own work, because my method's rule is: correct a wrong figure in public, with a date, and leave the original visible. In 2026 I built a pace-off-spin index claiming that spinners' economy rates in the IPL could be predicted from their "degree of turn." I calibrated it on 214 overs of the 2026 season and it showed 68% accuracy — which felt very good at the time. In the 2026 season it dropped to 51%. I spent a week understanding why. The answer was sample size and definition: I had taken "degree of turn" not from tracking data but from broadcast graphics, which were not consistent across matches. My input was wrong, not my model. And that is the most neglected lesson in cricket analytics: people rush to save the model when they should be watching the ball.

The same caution applies to auction valuation. A franchise says "we picked this player from last year's data." Fine, but last year's data was with that team, that field, that bowling attack. In a new team the same batter plays a different role. Correlation does not tell you the same performance will repeat in a new role — it only tells you it happened in the old role. This is the biggest oversight in post-auction reviews: teams look at "how it went," not "in what role he was used." Over the last three seasons I have counted at least seven cases where a player performed better in a new team than the old one, simply because he was sent to bowl in the right over or allowed to bat in the right position. The price was the same; the role changed.

The regional-language and print desk deserves mention here, because I am a foreign-born, sabermetrics-trained journalist working amid India's seventy-year print and radio lineage. Acknowledging that the Hindi and Tamil desks were doing verification work before the spreadsheets arrived is part of my method. I treat them not as a translation layer but as the real distribution channel. When I write about the auction, my reader is not only the English-newspaper reader — he is also the fan in Rajasthan who hears on Hindi commentary that "this player has a sharp mind" and believes the sentence without a number. My job is to put a number in front of that sentence.

Looking ahead: Next season's signal

At the next auction, when a price glows on screen, ask a question the scorecard never asks: what role did the team buy at this price, and what was that role's replacement level? The answer will often be uncomfortable. And that discomfort is the beginning of real analytics. I have a new column joining next season — "role-based leverage value" — and I am telling you in advance: last time it read 68%, and next time it may read lower. I will print it the same way, with a date.

The question is not whether data is right or wrong. It is whether you counted it, or someone counted it for you and pasted it on a screen.