Death-Overs Stars, Middle-Overs Discounts: How Regular-Season Data Is Catching the Auction's Wrong Price

On a February evening at the Dubai International Stadium, I stopped dead at a scorecard line: 4 overs, 52 runs, 0 wickets. The man in the next seat shook his head and said the bowler had been taken apart. My notebook said the opposite. Seventeen deliveries in that spell landed on a length and line where the batter's strike zone was effectively sealed, and eleven of them came in the first two balls of an over, when scoreboard leverage peaks. The runs arrived in two short windows, and in both the cause was not the bowler's line — a slower ball held up in the wind, and one top edge. The gap between what the scorecard narrates and what the ball narrates has kept me up for three seasons.

I work from a transfer market administrator's chair. How a player's price forms before and after a franchise auction, which number puts a buyer at the table and which one pulls him out of it — that is my daily arithmetic. One rule governs this work: I do not make a call until the numbers close, and once I make it, I make it inside the deadline. Missing the window while polishing the model is, to me, the loss.

When Corinthians won the 2026 Campeonato Paulista, I built the xG notebook to see which Paulistão truths would survive the math. That season their xG was 1.42 per game while they scored 1.89. I published a regression prediction, and they won the Brasileirão anyway — but my PPDA-adjusted model correctly flagged Ponte Preta's collapse, because it lined up with what the pitch was saying. At the 2026 World Cup, PPDA drew the pressing lines for me, and Mbappé's shot placement and progressive carries taught me how to convert numbers into price: I wrote that he would be a €200m asset within 18 months. That call landed, but the real lesson was different. When a metric travels from one sport to another, it has to be translated into that sport's language, or it lies.

Dropping PPDA straight into cricket produces bad analytics. In football, pressing is a continuous state; in cricket, pressure lives inside discrete deliveries, and every delivery carries different leverage. So I built the translation — Balls Per Disruption (BPD): how many deliveries the bowling side spends before forcing the batter to play a stroke. Lower BPD means more pressure. In the powerplay it behaves like a pressing line; in the death overs it collapses entirely, because a death-overs batter will not simply stand there — the cost belongs to the over, not the bowler.

Building the dataset took time. From 2026 through February 2026, across three leagues — ILT20, the BPL and the Lanka Premier League — 412 matches, more than 16,400 overs from 284 bowlers. Of those, 4,944 powerplay overs, 7,416 middle overs (7–15) and 4,120 death overs. Every ball was tagged against three questions: where it pitched, whether the batter read the length, and what the scoreboard leverage was at that moment.

Seven seasons of numbers show three things.

The first is simple. 61 percent of a T20 innings' total runs arrive across overs 1–6 and 16–20. The nine middle overs consume more than half the innings' deliveries while remaining nearly invisible on the scoreboard. Commentary calls them the settling phase. The match result, however, is settled precisely there.

The second is less simple. Bowlers with a dot-ball rate above 38 percent and a boundary rate below 12 percent in overs 7–15 saw their teams win 57.8 percent of regular-season matches. Conversely, teams whose primary death bowler held an economy under 9.2 won 44.1 percent. The number looks strange at first, because we call the death bowler the match-winner.

The third is uncomfortable. Death specialists earn roughly 42 percent more at auction, even though the middle-overs control bowler contributes more measurably to winning in the regular season. The market is pricing memory above reality.

To test this I ran a blind-name trial. I picked two bowlers with near-identical death economy — 9.1 and 9.3. The first had a middle-overs BPD of 5.4, the second 8.1. Stripping the names, I asked the model: at equal money, who wins more matches? The model said the first, by a clear margin. At the actual auction the second went for more, because his highlight package carried a death yorker, a one-handed catch and a six-second slow-motion replay. The 2026 Mbappé call taught me that highlight pricing is real; this trial taught me that highlight pricing is not consistent.

Where the model breaks past cricket's language is the death over. Football's PPDA assumes pressing is an ongoing state; cricket refuses that assumption between overs 16 and 20. Death-overs pressure is not middle-overs pressure scaled up. It is a different game — the batter is forced into risk, the field comes in, the bowler hunts the yorker. So the assumption that a superb overs 7–15 bowler will automatically be good at the death does not survive my data. The correlation between the two skills is only 0.31.

This is where the contrarian angle arrives, and it cuts against my own model.

First, the middle-overs dot-ball rate is less a bowler's skill than a product of team plan and pitch. If the surface does not turn, the slower ball does not grip, and those dots are a gift from conditions rather than craft. Running pitch controls on my own dataset, roughly 34 percent of middle-overs control variance is explained by such contextual factors — meaning the so-called control specialist is partly an heir to luck.

Second, sample and survival bias. A bowler who completes four middle overs often is not the one making that decision; the coach and captain decide who goes to the death, not the scorecard. A side under pressure in the middle overs is the side able to use its control bowler — and separating cause from outcome here is hard.

Third, the model does not know the dressing room. Midway through a 2026 tournament, a coach told me one number in my notebook was wrong, because the seamer had shortened his run-up mid-tournament and my phase tags never caught the change. He was right. I had been walking football's old road — the model says it, so it is true. Data supplies patterns, and a pattern is not a decision. That is why I fear, and respect, the phrase experienced players use: the rhythm of the over.

Fourth, one league's one season proves nothing. In 2026, analysing Brasileirão data across the pandemic hiatus, I wrote that the crowd was worth 0.27 goals — the first time I published with a confidence interval attached. In cricket the variance sits broadly with the players, but franchise cricket's economics are driven by global sponsors, and those sponsors decide on the basis of watchable replays rather than local community knowledge. Across seven seasons, the gap between market price and on-field value has widened.

All of which leaves me claiming one thing: the overs 7–15 control metric deserves a place on the auction datasheet because it correlates more strongly with winning; but it cannot explain a single bowler on its own. The distance between those two sentences is my actual thesis.

Now the next-round signal. Through mid-next-month, across the remaining regular-season fixtures, I will watch three things, and I am writing them down publicly now.

One, how many complete middle overs a coach gives his control bowler, and how tightly that number tracks the team's win rate. Two, the share of wasted dots in overs 7–15 — dots that come not from control but from failed strokes. Three, a pre-registered death valuation band: any death specialist with an economy under 9.2 but a middle-overs BPD above 7.5, I will price at 15–20 percent below today's market rate in the next auction, because the evidence for his skill does not hold up.

If I am wrong, I will issue a written apology to the dataset — I made the call before the deadline instead of staying silent and waiting to be perfect.