The Pressure Over Index: Where the BPL's Collapse Actually Hides

What happened in Chattogram last Wednesday was no mystery to the scorecard: 128/4 became 189 in twenty overs, sixty-one runs arriving in the last five. But the batter who walked in at the sixteenth over faced a different arithmetic — 62 needed, six wickets in hand, thirty balls left. My tracking sheet put that innings at a Pressure Over Index (POI) of 71.4, the highest of this season. Eight days earlier, in a near-identical equation, the same side sat at 42.1 — and won. The scorecard told two stories. The index told two ends of one story. This piece is about the second.

I have run Expected Truth out of Khulna since 2026. It began with a model built for the BPL; the work later spread into football, but my main desk never left cricket. Dhaka Premier League, BPL, domestic first-class — in all three I sit down with the same question: how much of what the scorecard shows is circumstance, and how much is skill?

Answering that needs a measuring stick. In T20 the biggest stick is the last five overs, because run rate and wicket risk rise together there. But "last five overs" is a boundary of time, not a boundary of pressure. A batter arriving at 170/2 in the sixteenth over and one arriving at 128/4 are not carrying the same weight. So I measure pressure by equation, not by clock.

The POI rests on six variables, deliberately no more. I know my own worst habit: add variables and the index looks elegant while its predictive power quietly dies. The six are required run-rate delta against the innings' first-ten-over average, wickets in hand, the set batter's dot-ball percentage, the incoming bowler's death-over economy, dew probability, and a pitch pace-and-bounce score.

Every ball is, to me, an immutable ledger entry. Once written it does not change; only new entries attach. That ledger discipline is what stops me from manufacturing a post-match story. When I watch from the ground as a batter absorbs a dot ball at the fourteenth over on purpose, the scorecard logs "slow batting" while my entry logs "wicket preservation" — and whether that preservation returns with interest in the next over is my actual question. I keep a method note at the end of every piece so the numbers can be rechecked rather than merely believed.

Baseline first. This season I have tracked 38 innings, 120 ball-entries each. Split the POI into three bands and the picture resolves. Below 45, low pressure: 17 innings, a 68 percent win rate. Between 45 and 65, medium pressure: 13 innings, 51 percent. Above 65, high pressure: 8 innings, 25 percent.

Note the sample — eight innings proves nothing, it only points a direction. But the pattern that keeps returning is this: in high-pressure innings, defeat does not come from run rate; it comes from wicket clusters. In six of those eight, two or three wickets fell between the sixteenth and nineteenth overs. Teams do not lose by failing to score. They lose the plot inside a single over.

This is where phase leverage earns its place. I divide an innings into 1-6 (powerplay), 7-15 (middle), and 16-20 (death). The middle nine overs usually produce 85 to 95 runs while conceding the fewest wickets, just 2.1. The death overs concede 2.9 wickets from half the balls. Which means your most expensive wickets are not protected in your cheapest phase; they are lost in your costliest one.

Reworking three seasons of ball-entries from Khulna, one thing kept surfacing. Sides strong in the middle do not win more matches than sides strong at the death; but sides that survive the death almost always had a strong middle. The relationship is not one-directional, yet the lean is unmistakable: a death-overs collapse is usually the final scene of a hidden middle-overs deficit, not the first.

A second index matters here — Recovery Efficiency (RE). When a side loses two wickets between the fourteenth and sixteenth overs, what share of its rhythm returns in the next two? This season, sides that lost consecutive wickets averaged an RE of 0.78, meaning roughly four-fifths of the lost rhythm came back. But the sides that recovered fully, with RE above 1.0, won 71 percent of those matches. The sides that could not won 23 percent.

Bangladesh's conditions give this a specific cause, and it is the finding I find most compelling. Our pitches are slow and low, and evening dew makes the ball hard to grip. So middle-overs spin, overs seven to fifteen, does more than choke runs — it builds reserved capital for the death. A side that spends four or five middle overs on spin for thirty runs keeps fast bowlers in hand for the sixteenth onward. A side that burns its quicks in the middle has no option left at the death, and where options vanish, the POI leaps.

Dew deserves its own look. In innings where dew probability sat above 70 percent, death overs averaged 10.4 runs per over; where it fell below 40 percent, they averaged 8.1. That gap is 2.3 runs an over, roughly 46 across twenty. And yet spinners were often given two more overs before anyone adjusted. Which means dew is not a weather report; it is a timing decision about bowling changes.

A wider example helps. The highest team total in international T20 cricket is Afghanistan's 278/3 against Ireland in 2026, per ICC records. In my entries that innings ran 63 in the powerplay, 114 through the middle, 101 at the death. At the other end, the highest individual score remains Aaron Finch's 172 against Zimbabwe at Harare in 2026, per ICC records. Finch's dot-ball percentage that day was freakishly low, but that was a single explosion, not evidence of structure. I do not chase outliers; I follow them until they confess that they are outliers.

Back to Wednesday. Before the POI reached 71.4 my model had issued three warnings. First, the set batter's dot-ball percentage over his last ten balls had climbed to 41 — he was scoring, but he was spending. Second, the incoming bowlers' death economy was 7.1, below the league average. Third, dew probability stood at 74 percent, meaning spin grip was fading. Stacked together, my model gave the chase a 31 percent chance. They made 61. The model was roughly in the right place, and wrong in the right place.

What the ground keeps showing me, the numbers catch late. Watching domestic cricket from a balcony in Khulna, one habit stands out: our batters watch the first ball of the sixteenth over rather than attack it. In European leagues the set batter pulls or scoops that ball, because the bounce invites it. On our pitches it skids low and becomes a dot. Pressure climbs in the next over, the batter tries to repay his own mistake, and gives his wicket away. The collapse therefore begins not with a bad shot but with a decision taken one ball late.

Three myths are worth breaking. First, that a death-overs specialist means a batter — in reality it means the decision to bank overs in the middle. Second, that more wickets in hand reduce pressure — in reality they do not spread the burden, they concentrate it on one pair of shoulders. Third, that dew treats both sides equally — in reality, a spin-heavy attack suffers and a pace-heavy attack is nearly gifted.

I am registering a pre-committed prediction, because without pre-registration a prediction is only an opinion. Three rules. One, definition: high pressure means POI above 65. Two, window: the remaining eight matches of this season. Three, revision rule: if the high-pressure win rate exceeds 40 percent, I will concede my band threshold was set too low and correct it. The call: in the playoff race, sides using at least four middle overs of spin will win more than 55 percent of their high-pressure matches.

Now the part where I stand against my own index. The relationship between POI and victory, presented as I have presented it, slides easily into a story about teams cracking under pressure. But correlation is not causation — and here the alternative explanation is less attractive and more likely.

The alternative is this: the POI is not measuring pressure, it is measuring overall team quality. A side with weak death bowling and a brittle middle will naturally carry a higher POI, because three of its variables are direct products of that weakness. "Low wins at high POI" then becomes not a story about pressure but a restatement of good teams beating bad ones. To escape that trap I did two things. One, I held batting and bowling quality as separate control variables so the residual POI effect could show itself. Two, I restricted the sample to closely matched sides, within five percent on overall rating. In that narrow sample, just 11 matches, the gap shrank to 68 against 39 percent — same direction, smaller magnitude.

The second caution is more uncomfortable. My model is blind where it has no variable. Dressing-room chemistry, a captain's call, a young bowler's nerve — none of these have an entry in my ledger. At the 2026 World Cup, Croatia scored 14 goals from 9.6 xG, and my model gave France a 58 percent chance in the final. The number was right, but how to measure that squad's belief was never in the model. The numbers did not break the model; they exposed where the model was blind. Cricket says the same thing: a POI of 71.4 tells you how hard the situation is, never who is standing in it.

The third caution is about time. Eight high-pressure innings prove nothing. This is a starting point, not a verdict, and I wrote the threshold down in advance precisely so the temptation to redefine it after seeing results stays out of my hands. One more admission: the cleaner an index becomes, the more easily it displaces reality. Numbers on paper and nerves on grass are not the same thing, and readers should hold that thought.

Over the next three weeks I will be watching three things: the number of middle-overs spin used, the dot-ball tendency of the set batter at the sixteenth over, and the timing of bowling changes before dew settles. None of these is a highlight, none appears on a scorecard. Yet this is exactly where the next collapses are already written. Expected truth is not a verdict; it is a guess I am willing to break myself. So the question is simple: does your team read the scorecard, or write it?

Method note: the six POI variables, thresholds and revision rules were registered in advance. Ball-entries were cross-checked against the cricsultan.com database. Sample: 38 innings and 4,560 ball-entries this season. Root: 2026, Khulna, Data Monk.