Early-Season Numbers: Which Stat Tells the Truth and Which Is Just Noise

17th over. The chasing side needs 68 off 42. The captain hands the ball to a spinner — two left-handers at the crease, one short boundary, and the frontline seamer's quota already spent.

First ball flighted, the batter goes straight, four. Second ball grips and turns in, hits the pad, review, not out. Third ball dropped short, pulled, six. Fourteen off the over.

Within thirty seconds the verdict is published: don't bowl spin at the death. My disagreement isn't with the verdict. It's with the sample behind it. On the same night that one spinner goes for fourteen in the 17th, another spinner somewhere else takes two wickets in the 17th for four runs — both outcomes emerging from the same season, the same laws, the same captaincy logic.

I hold a record of four thousand two hundred balls on this exact pattern, hand-coded in 2026 off a laggy broadcast feed into a spreadsheet nobody had asked for. The most useful thing that record taught me wasn't about spin. It was that one ball, one line, one length produces two different numbers in two different atmospheres.

This piece points at that idea, but from the regular season's angle. The first three or four weeks hand you the most reliable and the least reliable information on the table simultaneously. Wins and losses are true because they are recorded. The reasons behind them are mostly false right now, because the sample behind them is roughly one over long.

Better to publish the falsifier first. If, over the next three weeks, those same spinners on those same grounds in those same chase situations — with no change in deployment — keep conceding above nine an over and their captains keep handing them the 17th, my argument is weak. If economy drifts back toward seven, then that fourteen-run over was an over, not a theory.

Data starts here, not at the result.

Context: what phase-adjusted actually means, and why it matters now

A bowler's tournament economy is an average, and the flaw of an average is that it never states the conditions it was built in. A spinner who bowls 40% of his balls in the powerplay and a spinner who bowls 70% of his after the 15th can both sit at 7.6 — and mean entirely different things.

So I code every ball into three phases: powerplay (1–6), middle (7–15), death (16–20). Starting death at 16 is itself a coding decision, not a habit. Several analysts start at 17; the difference matters when a spinner only ever bowls the 16th, and looks like two different bowlers under two definitions.

The second part of the coding spec is duller and it is exactly why I don't take a provider's dashboard at face value. What is a dot ball? A delivery the batter hit for no run? A wide? A leave? A pad-ball that becomes a leg bye? In my sheet a dot is a ball where the bat made contact or the batter shouldered arms and no run resulted; wides and no-balls sit in separate columns, because a wide is the bowler's error, not the batter's restraint.

Dropped catches are the third and most contested column. Does a drop count as a wicket? In my sheet it doesn't become a wicket for the bowler, but the runs conceded off that ball still do. It looks unjust. My job isn't justice, it's prediction — and for prediction what matters is repeatability, not biomechanics. Change that one decision and several death-over rankings move two or three places. That's why two providers can list the same bowler at different positions and both be technically correct.

I hand-coded four thousand two hundred balls before I trusted a dashboard. That isn't a boast about labour. It has a practical payoff: when a number looks strange, I know where to open it.

Now the sample arithmetic, which I publish before anything else. If a spinner has bowled six death overs across five matches, that's 36 balls. On 36 balls, an economy estimate can swing three to four runs per over in either direction — meaning a gap between 8.9 and 6.4 is usually the same box photographed twice. Until the gap exceeds the swing, the comparison is decoration.

That is the error sold hardest in every season's first month. It sells because it is interesting, not because it is right.

Environment comes next. When dew arrives, in which city, how deep the boundary sits, day game or night, how much travel the side has logged — all four move a death-over number as much as the bowler's ability does. So I never publish a death economy alone; I publish the ground, the hour and the innings context it was built in.

Fixture congestion is the least measured variable of this cycle. Two games a week is not something a medical staff can fix. When a quick bowls the 19th in three consecutive matches, the fourth match's yorker tends to sit a little higher — and that height shows up in the pitch map, not the medical report. Workload should be read off the calendar before the gym, and that applies directly to death-over numbers.

Core: three cases that built the method

Three months before my statistics finals in 2026, I hand-coded 4,200 IPL deliveries. The output was one thread on Bhuvneshwar Kumar's Purple Cap season — 26 wickets — showing that 78% of his overs 17–20 deliveries landed in the yorker band, and that his 7.02 death economy was structural, not streaky.

That sounds like a small observation. It isn't. When 78% of your balls land in one band, your output is an extension of your input; luck gets less floor space. It gave me a permanent habit: stop treating death economy as an output, treat it as an input. The question isn't "what is his economy", it's "how often did he bowl the ball he intended". The first question looks backwards. The second looks forward.

The model was clean; the finishing was not. That lesson arrived a year later.

In 2026, as a junior analyst at a Hyderabad sports-data vendor, I logged all 64 matches of the Russia World Cup ball-by-ball into an expected-goals model. Sixty-four matches gave me one xG model and a finishing lesson.

France beat Argentina 4–3 in Kazan. My model had it 2.4 to 2.1. Seven goals separated the model from the scoreboard. For three weeks I treated the scoreline as testimony against the model and rebuilt the shot weighting — not to declare the model right, but to see how far it trusted itself.

Then I ported the logic to cricket, refusing to let the borrowed word do unearned work. The cricket-specific translation is expected runs per delivery by line and length. When a batter makes 500 in a season, the question isn't that he made 500; it's how many he should have made off the balls he faced. In one case I modelled, a 500-run IPL season was running 31% ahead of its own shot quality.

Thirty-one percent ahead isn't a scandal. It says the finishing outran the model — and assuming that generosity repeats next season is the mistake. Regression isn't a forecast. It's a warning.

Before I rank a player, I ask what the ball was doing. That question keeps taking me to the same place: a poor death economy is often poor deployment, not poor bowling.

The 2026 season is the cleanest demonstration. The IPL moved into three empty stadiums and I moved with it, consulting for the Hyderabad franchise. Mitchell Marsh went down with an ankle injury in the season's first match, leaving no seam-bowling replacement.

I pushed hard for Jason Holder. Not romance — a variable. Empty stadiums made dew a variable I could no longer ignore. Crowd noise receded; dew did not. And dew is invisible until you watch the ball hit the surface in the second innings.

The model was blunt: in Dubai chases, spin economy climbed from 6.9 to 8.4 after the 16th over. A run and a half per over, across the last five, changes the shape of a chase. The right reading isn't that spin becomes bad. It's that the same spinner's same ball to the same batter is now a different ball, because the ball is wet and the grip has changed.

That season turned into three must-win games, a win over RCB in the Eliminator, and a loss to Delhi Capitals in Qualifier 2. I won't claim the dew model won those matches. I'll claim it answered the question everyone was asking — who bowls after the 16th — and that answer blocked at least one wrong decision. A model's most honest contribution is usually that: fewer wrong decisions.

My job is to make the model legible to people who must act. So I walk into the room with a sentence, not a decimal: "If the dew comes, spin is our most expensive option in the 17th." Coaches decide off sentences.

Four audits for this regular season

The first audit is the sample. Every time a number surprises me, I count the balls behind it first. A strike rate off 20 balls is an opinion; off 200, a claim. Both get printed in the same font, and that is the trade's biggest failing.

The second is phase. A batter's overall strike rate says nothing about his utility; his middle-overs rate says whether he can hold an innings together, his death rate whether he can end one. Fusing them into a single number is convenience, not analysis.

The third is environment. Dew, boundary dimensions, day-night, travel — separate all four, because three of them aren't in the bowler's control and neither is the fourth.

The fourth is the market. The IPL auction is a public pricing experiment and it can be graded. Franchise scouting has quietly replaced the zonal selector, which makes the auction the rare case where a market publishes its own bets. I treat each buy as a hypothesis with a settlement date two seasons out — then score the room.

Contrarian: rule out the boring explanation first

This is where I have to restrain myself. Correlation is not causation, and in cricket that error is most common among the people looking hardest at data.

A spinner's ugly death economy has at least four boring explanations with nothing to do with ability. One: the captain only bowls him once the game has already gone and the batter is taking risks, so the sample itself is biased. Two: his overs come from the end where he bowls into the wind. Three: the ground is small. Four: he keeps drawing left-handers because the captain is playing match-ups, and a match-up is a strategy, not a flaw.

Until those are ruled out, reaching for the counter-intuitive conclusion is just brand-building. I know that being right against the room built my reputation, and that is precisely why it is my biggest weakness.

The other problem is where the argument happens. The commentary box is the last unaudited room in Indian cricket. Data desks won the tactical argument years ago; the holdout isn't ignorance, it's broadcast economics.

Ad breaks reward milestones over match situations. That's why a hundred still outranks a phase-shifted innings in the running order — a hundred can be sold in thirty seconds, a 22-ball-43 in the 43rd over cannot. This isn't a naive audience; it's a business model. Until the price of a break is tied to milestones, the box will sell milestones. My job is to name the economics, not to sneer at the viewer.

Which is why I never build a piece around a hundred. The milestone is an outcome to be explained, never a cause to celebrate.

I learned to count what the broadcast camera leaves out. The camera follows the ball; it doesn't show how deep slip stood, how far the keeper came up, or why the captain moved the field in the 16th. Start counting those three and a lot of numbers become questions — which is good news.

Takeaway

Over the next four weeks I'll watch one thing: not who bowls the death overs, but whether the balls they bowl are the balls they intended. The day a spinner lands four of six in the 17th on his chosen line and still goes for twelve, I'll change a column in my sheet. Until then, death economy is a receipt, not a verdict.

The question isn't whether spin works at the death. It's whether your team knows which weather its 17th over is being bowled in — and the answer isn't in a dashboard. It's in an old Hyderabad spreadsheet and the column where I first wrote down what the ball was doing.