109 in New York, 106 in Kingstown: A Variance Audit of Bangladesh's T20 Batting

On 10 June 2026, at Nassau County International Stadium in New York, South Africa made 113/6 and Bangladesh 109/7 — a four-run margin. Six days later, on 16 June at Arnos Vale in Kingstown, Bangladesh were bowled out for 106 and then dismissed Nepal for 85, winning by 21 runs.

Three runs separated the two innings. The results were exact opposites. Television commentary called the gap pitch, luck, or a day that was not ours. Line the two innings up on my phase-split sheet and the damage sits in the same place — the dot-ball density in the ten balls after a wicket falls, and the boundary frequency between overs 7 and 15. In New York those two indicators read 64 percent dots and one boundary every 11.4 balls; in Kingstown, 38 percent and 6.9.

The xG map said 2.7, but Burnley. — Root: Chattogram xG blog after Burnley. A map is never a substitute for the match. Without a map, the internal structure of the match never surfaces either.

Context

In August 2026, when I started the Chattogram xG blog from Chattogram, I wrote about Chelsea against Burnley — Chelsea 2.3 xG, Burnley 0.9, final score 2-3. That piece fixed my rule: a match note opens with run rate, dot-ball percentage and boundary frequency, and only then tells the story. Earlier, covering the 2026 Wills Cup in Dhaka for Prothom Alo, I had learned the opposite order — event first, numbers after. Eleven years into this trade, the two have swapped places.

Cricket does not allow a clean mirror of football's xG. In football a shot is a discrete event; in cricket a single delivery carries a wide spread of outcomes — dot, single, boundary, wicket. So instead of expected runs I built a phase par score: pitch pace, boundary dimensions, dew and the average speed of available spinners, combined into one reference number.

On the New York pitch my par score was 119; in Kingstown, 97. Bangladesh's 109 sat seven runs under par in New York. Their 106 sat nine runs above par in Kingstown. Across the six group innings of the 2026 World Cup, Bangladesh finished above team par four times — and lost three of those.

The Asian context sharpens the numbers. On 28 September 2026, India beat Pakistan in the Asia Cup final in Dubai; on the slow, low-bounce surfaces of that tournament my tracking puts par scores 8 to 12 runs lower and spinner economy between 6.8 and 7.4. The 2026 T20 World Cup runs from 7 February to 8 March in India and Sri Lanka, where dew, slow pitches and spin-heavy middle overs will make this model matter more, not less.

Core Analysis

The ten balls after a wicket. I isolated this window from ball-by-ball data across Bangladesh's six innings in the 2026 World Cup. In ten-ball windows without a wicket, the run rate was 8.1 and a boundary arrived every 6.3 balls. In the ten balls immediately after a wicket, the run rate fell to 5.9 with 51 percent dots. The top-eight average in the same situation is 7.2 and 39 percent. Bangladesh does not slow down under pressure; Bangladesh sends two new batters into safe mode at once, and that safety is repaid with interest over the next four overs. This one pattern links the four-run margin in New York to the 21-run win in Kingstown.

Overs 7 to 15. This is where tempo is decided. Top-eight sides find a boundary every 7.2 balls in this phase; Bangladesh's figure is 9.4. Wicket loss is nearly identical (one every 21 balls). We are losing batters at the same rate but replacing them far more slowly. Towhid Hridoy was the most reliable asset here — a strike rate of 131 in innings lasting more than 34 balls, and 118 in the ten balls after a wicket, against a team average of 91. That gap is the message.

The match-up grid. Against slow left-arm orthodox spin, Bangladesh's top order scores 0.81 runs per ball and the middle order 0.74 — both below the top-eight mean. Against leg spin the picture flips: when an opponent bowls leg spin through the middle overs, Bangladesh's top order cannot manage the pressure, because early turn is limited and the low-grip scoop does not work on slow pitches. Rebuild the grid and one pattern appears — if the opposition introduces spin in the 12th over, Bangladesh's run rate drops by 1.4.

The variance band. Taking the gap between the 25th and 75th percentile of innings run rate, Bangladesh's variance is 2.8; India, Australia and England average 1.9. The problem is not capacity, it is constancy. The two extremes of form from Litton Das and Najmul Hossain Shanto in the same position account for a large share. Group-stage net run rate punishes that variance, because three small innings are not covered by one large one.

Death overs. Between overs 16 and 20, Mustafizur Rahman's economy is 7.1, Taskin Ahmed's 8.3, Tanzim Hasan Sakib's 8.9. A cutter-heavy attack travels well on slow pitches, but on a flat deck with dew, a chasing side's strike rate climbs to 157. On the 2026 Indian and Sri Lankan surfaces, that dew factor will carry the heaviest weight in my model.

The process. My pre-match checklist runs in four steps. One, set the par score and pitch factor. Two, hold a written plan for the ten balls after a wicket — who takes strike, who refuses to play dot balls. Three, pre-assign the match-ups for the moment spin arrives in the 12th over. Four, after the innings, log the gap between par and actual, with an error range.

In plain language: a par score is how many runs were safe on that pitch on that day. NRR means net run rate, the tie-breaker that decides tournament tables. DLS is the rule that resets targets in rain-shortened matches.

The Contrarian Angle

The familiar criticism is a low powerplay strike rate. My data points elsewhere. Bangladesh's powerplay run rate is 7.4, below the top-eight mean, but the gap is only 0.6. The middle-phase shortfall is 1.1, and the variance shortfall is around 0.9. With limited resources, buying an opener does not close the middle gap. This is where correlation and causation get confused — a 150-strike-rate opener raises the powerplay average, but if the side enters safe mode after a 13th-over wicket, the scoreboard does not move.

My own model deserves an open challenge too. With a six-match sample, the error range on window-level data is wide; the ten-balls-after-a-wicket sample runs two to three per innings, so 51 percent dots is trustworthy while exactly 51 is not. The Burnley lesson is worth keeping: a map can be wrong, but throwing the map away leaves nothing. A model without error bars and a story without a model are both incomplete.

Takeaway

In the 2026 World Cup in India and Sri Lanka I will count wickets lost in the powerplay, not runs scored, and my second indicator will be the dot-ball percentage in the ten balls after the 13th over. If that number drops below 40 within two weeks, Bangladesh will do more than survive the group stage — they will squeeze serious opposition in the last eight. So the question is simple: will the team write the crisis plan before the match, or again decide inside the match and call the difference between three runs and 21 runs luck?