Boundary Baseline and Death-Over Deviation: The Table I Built Before the T20 World Cup 2026

Hook: 30 off 30

Kensington Oval, Bridgetown, 29 June 2026. The board read South Africa 151/4 after sixteen overs, six wickets in hand, Heinrich Klaasen on 52 off 27. On my laptop a live win-probability script was running, built on the phase-by-phase run distribution of more than 4,800 T20Is logged between January 2026 and December 2026. The script said 86.4 per cent. Six overs later the board read 169/8. India had made 176/7 and won by seven runs.

Those seven runs were no accident. Four overs, two wickets, 18 runs — Jasprit Bumrah's spell broke the match baseline. South Africa's expected runs across the last six overs were 31.6; they made 23. The gap of 8.6 runs was the tournament's decisive deviation. I said as much before the final ball was bowled: this final was not the story of a batsman's heroism, it was the story of one bowling over breaking a baseline. Twitter wrote "Klaasen's storm"; my table wrote something else.

Where the gap between expected and actual runs crosses ten, the individual innings is noise in the audit log, not signal.

Context: the tournament you have to build a table for

The ICC Men's T20 World Cup 2026 begins on 7 February and ends on 8 March at the Narendra Modi Stadium in Ahmedabad. This edition has twenty teams across four groups of five in India and Sri Lanka; the top two from each group reach the Super 8; four teams then reach the semi-finals. The final was fixed in advance for 8 March. Colombo's R. Premadasa, Pallekele and Dambulla offer Sri Lanka's slower surfaces; Ahmedabad, Mumbai and Kolkata offer flat decks. Two pitch realities inside one tournament. That venue split is the largest variable in my baseline, so I keep it in two separate tables.

Methodology box: my sample holds 4,843 T20Is from January 2026 to December 2026, more than 1.12 million legal deliveries. Three phases: powerplay (1–6), middle (7–15), death (16–20). The expected-runs model is weighted by shot location, boundary propensity, match-up (spin or pace) and batsman handedness. Two supporting indices: the Dot-Ball Pressure Index (DBPI), the pitch-normalised share of dot balls per over, and the fielding residual, the expected runs lost to dropped catches and missed run-outs. Every number carries a confidence interval, and every claim has been placebo-tested — in data journalism the virtue is reproducibility, not elegance. The R code and dataset sit in my public repository.

Five things I will track hardest this edition: powerplay run rate, middle-over spin economy, death-over dot-ball share, fielding residual, and the decay of home advantage at neutral venues.

Core analysis: the baseline table and its breaking points

1) The phase baseline: where expectation sits still

Across my 4,843-match sample, the phase run rates land at 8.14 in the powerplay (95% CI: 8.07–8.21), 7.46 in the middle (7.39–7.53) and 9.92 at the death (9.78–10.06). The striking part is that between 2026 and 2026 these three numbers barely moved — the powerplay average shifted by 0.21 runs per over. Tournament cricket does not invent new behaviour; teams return to the old baseline and simply make faster decisions under pressure.

Deviation lives at the death. At the 2026 World Cup the tournament-average death-over run rate was 10.31, but in the knockout stage it fell to 8.94. Run production in the last five overs drops by roughly thirteen per cent in the biggest matches. The side that anticipates that fall is the side that strangles the scoring space.

2) Boundary share: a risky but real relationship

In the 2026–2026 sample, a team that drew more than 58 per cent of its total runs from boundaries in a T20I won 71.3 per cent of those matches. Below 50 per cent, the win rate dropped to 38.9 per cent. But the relationship is not linear, and I will show why shortly.

Boundary share is an outcome, not a cause — and anyone who reads it as a cause has not read the other side of the table.

3) The Dot-Ball Pressure Index: the true separator

When I split matches by DBPI instead of boundary share, the difference became far cleaner. Teams holding a death-over DBPI above 0.60 won 82 per cent of their matches at the 2026 World Cup; teams below 0.40 won just 27 per cent.

The final's numbers belong here. Bumrah took 2/18 in four overs at an economy of 4.17 — the lowest economy of any bowler with ten or more wickets in that edition. His real contribution was not in economy but in dots: his death-over DBPI was 0.73 against a tournament average of 0.52. Almost the whole 8.6-run deviation was written in one bowler's phase control.

4) Spin in the middle: the hidden variable of 2026

Everyone talks about batting-friendly pitches on India's flat decks. My data says something different about the middle overs. In T20Is played in India from 2026 to 2026, spin's average economy in overs 7–15 was 7.12 against pace's 8.03. On Sri Lanka's slower surfaces it falls further, to 6.74. The middle of this tournament belongs to spin, and that is where matches are settled.

Four spinners' residuals I will track hardest in that phase: Rashid Khan (Afghanistan), Kuldeep Yadav (India), Wanindu Hasaranga (Sri Lanka) and Maheesh Theekshana (Sri Lanka). At the 2026 World Cup, Rashid Khan's middle-over economy was 6.29, roughly a run below the tournament average. Afghanistan's run to the semi-final has been told as a batting story; in my table it is a spin-residual story.

5) Home advantage: the empty-stadium experiment still paying out

In 2026, after the Bundesliga returned behind closed doors, I built an index from the first five rounds: the home win rate fell from 43.2 per cent to 21.1 per cent, home goals per game from 1.65 to 1.08. That index is the foundation of my cricket baseline today. The reason is simple: much of home advantage is actually crowd noise, umpiring bias below the level of awareness, and a player's familiar rhythm — not the venue itself.

In T20Is played in India from 2026 to 2026, India's home win rate is roughly 78 per cent. At neutral venues, where both sides are visitors, India's win rate in that same window falls to 63 per cent. That fifteen-point gap is, in my accounting, the tournament's largest hidden variable. In Sri Lanka it will thin further, because the crowd arithmetic changes.

Every empty stadium was a controlled experiment we never asked for — and in 2026 its results decide which venue is truly home.

6) The fielding residual: the number nobody puts on the table

A dropped catch in the death overs costs an average of 4.8 runs in my model, because the reprieved batsman usually enters a boundary-prone stretch over the next two overs. At the 2026 World Cup, 31 catches went down in the group stage; of the four sides with the most drops, three failed to reach the Super 8. That could be coincidence, but the 2026 and 2026 samples point the same way, so I no longer call it coincidence — I am enlarging the sample.

7) Bangladesh and the data-provenance question

Bangladesh's baseline has to be built differently, because two separate problems arrive together: a cricket problem and a data problem.

First the cricket. From 2026 to 2026 Bangladesh's powerplay run rate was 7.31, roughly 0.8 runs below the global average. The death bowling is genuinely strong — Mustafizur Rahman and Taskin Ahmed hold a death-over DBPI of about 0.58, above the tournament average. The strike-rate distributions of Litton Das, Najmul Hossain Shanto and Towhid Hridoy say the problem is not talent but phase transition. A slow powerplay creates extra risk against spin in the middle, and that is where Bangladesh are repeatedly caught.

Second, the data. There is a gap between Bangladesh's domestic ball-by-ball feeds and the ICC's tournament feeds in labelling and event timing, particularly for dropped catches and fielding positions. Associate-match feeds carry more missingness; at some venues delivery timestamps are rounded to the second, which introduces a small but systematic error into dot-ball counting. I learned this lesson building the 2026 index: however clean the model, if the pipeline is dishonest, the output is dishonest. So every analysis I publish now carries a provenance note — and that note sometimes becomes the story itself.

Contrarian: correlation is not causation

Anyone who sees the 71 per cent boundary-share relationship and concludes that hitting more boundaries wins matches will be wrong, because reverse causality is also at work. A team that is ahead bats on easier pitches, against weaker bowling attacks, with the field spread — being ahead is what raises boundary share. To test that endogeneity I re-ran the regression controlling for win probability after ten overs of the first innings; once controlled, the boundary-share effect falls from 71 per cent to 59 per cent. The effect survives, but it is overstated.

The baseline itself needs auditing. If I apply the 2026–2026 run rates to the 2026–2026 sample, the death-over baseline reads roughly 0.9 runs lower. The reasons are familiar: two new balls, changed fielding restrictions, bat technology, and the behaviour of the SG ball on Indian soil. If a tournament is played in India and Sri Lanka, England's or Australia's pace baseline does not transfer directly.

There is another trap, one that grows out of my own professional habit: mechanism-hunting. One abnormal result and we build a beautiful causal story — "death-over pressure", "tournament nerves", "momentum". I now pre-specify each mechanism before looking, then run placebo tests — I test a variable that should have no effect in the same way. If the placebo also looks "significant", my model has noise, not my story truth.

Momentum claims are, to me, a hypothesis rather than evidence. I operationalise them: does a side's trend in DBPI and boundary share across three consecutive matches predict the next result? Across the 55 matches of the 2026 World Cup the answer was essentially no — a correlation of 0.11, well inside the confidence interval around zero. The eye test is a witness; the data is the cross-examination.

Takeaway: what I will watch in the Super 8

Three signals stay on my table before the Super 8. First, the middle-over spin residual — the side that can hold spin below an economy of 7 in overs 7–15 will have a hand on the semi-final door. Second, the death-over DBPI — teams above 0.60 go deep in knockouts; that is my prior, and I am willing to be proven wrong. Third, the decay of home advantage at neutral venues — if India cannot hold that 78 per cent in Sri Lanka, the tournament's story turns elsewhere.

I do not chase narratives; I build a table and wait for them to arrive. When the trophy is lifted in Ahmedabad on 8 March, the question will remain — did the winner have the hardest bat, or the cruellest dot-ball counter?