The Hidden Numbers of the Death Overs: A Cricket Translation of PPDA and Franchise Valuation in T20

Hook: What a Run-Up Reveals About a Match

At Sharjah that night I was not watching the scoreboard. I was watching the bowler's run-up length. Over 18.2, chasing side needing 31 off 16, four wickets down. The stands were loud, but the spinner at the top of his mark had cut his approach by nearly a foot and a half. Short run-up means less time — less time for the batter to make a decision. Seven balls, one run, one dot, one wicket. The over finished at two runs.

That over opened a new column in my notebook. If I had logged economy rate alone, I would have written "two runs, one wicket." What actually happened was seven consecutive deliveries that broke strike rotation, forced shot angles to change, and pushed a chasing side out of its own arithmetic. Economy rate cannot see inside those seven balls. My question started there: can pressure in T20 be measured at all? And if it can, what is it worth at a franchise auction?

Context: Why Economy Rate Lies in the Death Overs

T20 cricket is habitually split into three phases — powerplay (1-6), middle (7-15), death (16-20). The split itself is the biggest simplification. In the death overs, an over's meaning depends entirely on match state: runs required, wickets in hand, who is batting, which way the wind blows, whether the dew point is active. If the same bowler takes three wickets for twenty while the opposition needed sixty, then concedes eighteen while it needed thirty-five, the second innings is arguably the better performance. Economy rate reads both the same way.

The leverage logic I used in football with PPDA — Passes Allowed Per Defensive Action — was simple: the height of a pressing line tells you how much risk a team is willing to carry. Cricket's equivalent question is whether a bowler or bowling side is willing to carry risk, and what that risk returns. In the death overs the risk structure is strange. A missed yorker is four. A bowler who fears missing it and drifts wide instead invites extras and free-hit spirals. Over the last three years, bowling sides with a sub-10.5 death economy against chase have kept an aggressive-delivery share (stump line or into the batter) above 62 percent. Sides above 11.5 have slipped to 47 percent. Fear-driven bowling and wicket-seeking bowling are two different games, and the scorecard writes both in the same language.

Here I should name a habit of mine. Professionally I am a transfer market administrator. My work is largely about putting performance data into a structure that supports budget, retention and exchange decisions. What I learned first in that chair is this: when a metric travels from the field to the auction table, its internal uncertainty is the first thing to disappear. A franchise wants a number, not a confidence interval. The data monk's job is to put the band back — give the number, but never without the range.

Method: From PPDA to WPD — Translating Pressure into Cricket

In 2026, as a high-school student in São Paulo, I built the xG notebook for Corinthians on my Paulistão blog — 1.42 xG against 1.89 actual goals. That was not just a numbers exercise. It was building a habit: data table first, explanation second. The core idea of my cricket model is WPD — Wicket-Weighted Pressure per Delivery.

WPD rests on three inputs.

First, three-level leverage. Every ball carries a context score driven by required run rate, wickets in hand and innings position. A dot ball with two wickets in hand at 18.5 is not the same event as a dot ball with eight wickets in hand at over 12. I use a weight function where RRR multiplies wicket scarcity.

Second, outcome value. A dot ball is not worth zero. It is positive unless it locks the batter into a line and length that destroys strike rotation on the next ball. A wicket is weighted separately for the bowling side, then added to a follow-up style value across the next two deliveries.

Third, the price of aggression. This part concedes that a bowler did not take a risk. A wide, or a short and wide delivery, should not be forgiven. An attempted yorker that became a dead ball should not be punished as if it were a mistake of intent. I run an intent-bias control here — a cricket cousin of a confession model, using pitch maps and delivery data to separate what the ball was aimed at from what the bowler's plan required.

Add those three components and a bowler's death-over profile emerges that economy rate will never show.

Core Analysis: The Death-Over Data Chain

I built a chain from the last two franchise cycles and recent international T20 series. The sample is limited, so every claim comes with a band.

Link one — dot-ball shape. When a bowling side's bowled-or-intended-stump share exceeds 55 percent in overs 16-20, opposing batters' combined strike rate in that phase stays under 140. Drop below it and the strike rate climbs toward 160. Without the band the claim is incomplete: the relationship is stronger on large grounds (boundaries over 70 metres) and weaker on small ones (under 65 metres).

Link two — over-opening control. Death-over success comes from the first two balls of an over. Bowlers who keep at least 65 percent of their overs to four runs or fewer off the first two deliveries average a co-cycle death economy below 9.8. Here comes a large sample-size caveat: in one franchise tournament a bowler may bowl only 40 to 50 death balls. Pricing a player off fifty balls converts model error into a decision. I therefore apply a Bayesian correction — the most recent two cycles carry the most weight, but older data is discounted rather than deleted, because death-bowling skill rarely dies in six months. It tapers.

Link three — match-state translation. This is my favourite space. I separate two metrics: pre-required and post-required death. Pre-required means the chasing side is held above 15 RRR; economy combines with strike-spray control. Post-required means the batting side is already winning at a canter. Those balls are few but distort economy rate disproportionately.

A concrete recent example. Across a U19 and UAE domestic T20 cycle I tracked a pattern: bowling sides that move away from line-and-length in the death overs toward toe-crunching plus wide-yorker consistency gain 7 to 10 percent in WPD, but their economy rises 0.4 to 0.7 runs when wides increase. Weighting both against win probability, the first effect outweighs the second — on one condition: the square boundary must be protected. The realistic explanation is that toe-crunching pushes risk to the boundary edge, and a short square boundary turns that edge into a single.

Link four — spin's death role. Many analysts treat a spinner in the death overs as an extra option. The data says a second or third spinner's death success depends on his powerplay water-saving role. If a spinner keeps a powerplay economy under 6.5, his death role as a blue-mind disruptor works, because batters cannot read him. If his powerplay numbers are average, bringing him on at the death invites noise. This is the cricket version of the trap where a metric receives excess trust.

Link five — the invisible value of fielding. A saved boundary against a conceded one is worth roughly 0.4 to 0.6 expected runs at the death. My model folds death-field positioning quality in as a safety factor, because aggressive delivery plus poor fielding equals catastrophe.

Contrarian Angle: Correlation Is Not Causation

Two good stories sit in front of this model. One will hold. One may fall.

The first: PPDA drew the pressing lines, and Mbappé's shot map said the rest. In 2026, tracking France's PPDA at 12.4 and Mbappé's 0.18 xG per shot, the logic was plain — shot location plus progressive carries equals future value. That call worked. Which is exactly the trap. A World Cup is a sample, not a law. If I transplant that same logic into cricket — "this bowler's death-dot ratio is 60 percent across one T20 series, therefore he carries £800,000 annual franchise value" — I am deriving a rule from a seven-match series with no external validity. The gap between correlation and causation is called marginality: tournament-specific capacity, ball conditions, opposition quality.

The second: the 2026 empty-stadium home advantage study. In Brazil's Série A, home win percentage fell from 52.1 to 42.6, goal difference dropped 0.27, and distance covered stayed flat. That study taught me to test every equivalent variable before blaming one. Cricket's death overs work the same way. When a bowler's economy suddenly rises, my first thought is not "the form is gone." I check whether his share of death-phase overs changed, whether ground dimensions changed, and how often he bowled in lost-match contexts.

The third and least comfortable: just as gegenpressing was solved by mid-table sides through athleticism, death bowling in cricket is undergoing the same translation. The slow-ball trend, the yorker trend — these are becoming a pure physical skill: how fast can you hit the same length repeatedly. This does not mean cunning is finished. It means intelligence now invests elsewhere — not beating a batter's logic, but breaking the pre-move logic he carries into the delivery. The bowlers still standing are the ones who knew, before releasing, which line the batter would not leave.

The biggest caveat is overfitting. We hold one cycle of data in which a bowler delivers 120 to 150 death balls. Fit five features to that sample and the model looks clean in training, then fails outside. I cap personal models at three features and run an out-of-sample check before any public call. Without that discipline, a data monk is a dangerous thing — he can be wrong with total confidence.

Franchise Valuation: The Auction Misprice

As a transfer market administrator I have watched mispricing happen when people read a production metric as a valuation metric. In cricket, wicket count is a production metric. The valuation metric is WPD-adjusted contribution: expected runs saved against replacement-level death bowling, weighted by the leverage of the overs in which they were saved.

The step most auctions skip is replacement cost. A franchise signs a death specialist not against zero but against the next best option on its own roster. If the internal alternative already delivers 8.6 death economy at 55 percent stump intent, then paying a premium for a 9.2 economy at 62 percent intent is a deliberate purchase of variance, not of value. Teams pay for variance all the time without naming it.

The second skipped step is durability under role drift. A death specialist who is asked to bowl overs 14, 16 and 18 rather than 18, 19, 20 loses the fixed structure that produced his numbers. My model adds a role-drift penalty, because what looks like a decline is often a change in assignment.

The third step is the honest one: publishing the range before the auction closes. A forecast without a pre-registered assumption is not a forecast; it is a recollection dressed up after the fact. My own rule is to write down the triggers — if this bowler is used in two or more pre-required overs per match, his WPD holds; if he is used as a middle-overs enforcer, the profile decays.

Takeaway: What to Watch Next

In the next series, watch one number before you watch the score: the share of death deliveries aimed at the stumps. If it rises while wides stay flat, that bowling side has found something structural. If wides rise with it, you are watching risk without control, and the wins will be loud but shallow.

The second signal is the first two balls of every death over. Teams that win that mini-contest consistently will top the table long before the standings say so. And when the next auction arrives, ask a simple question of every death specialist on the board: was his economy earned against a required rate, or against a settled match? The answer is worth more than the number itself.