Thirty Needed Off Thirty, My Table Said 77 Percent: An Audit of the 2026 T20 World Cup Baseline

Kensington Oval, June 29, 2026. After fifteen overs of the final, South Africa were 151 for 4. The target was 177. Twenty-six runs needed off thirty balls, six wickets in hand, with Heinrich Klaasen and David Miller at the crease. I was watching the match with a notebook beside the laptop. My simulation log read 77.4 percent. Seven overs later the board said 169 for 8. A seven-run defeat.

I did not rewatch that passage out of heartbreak. I rewatched it because the gulf between 77 percent and zero percent is the most honest laboratory cricket offers. The story of that final is not a curse. It is the story of how a model reads one phase perfectly and the next phase wrongly, and those errors are the real asset, because a record of your own mistakes is the only defence against repeating them.

Context: which data I was working with

Before every tournament I download the ball-by-ball feeds of the previous three editions, split them over by over, and chart runs and wickets per phase. Then I keep the current tournament's feed in a separate bag so no single number leaks into the baseline. That sounds like over-engineering. Kazan in 2026 taught me something simpler: an analyst who folds his own results into his own baseline is not forecasting, he is narrating.

The 2026 feed had one obvious problem. The drop-in pitches at Nassau County in New York behaved in a way that venue's own historical database cannot explain. In other words, there was no baseline there, only a guess. So I put those eight matches in a separate bag and compared them to the rest. The gap: 6.1 runs per over in New York, 8.4 everywhere else. That 2.3-run gap cannot simply be blamed on the surface, because on the same ground India and Pakistan produced 232 runs between them, while Sri Lanka folded for 77.

The core: expected runs and expected wickets

The first xG model I built did not predict football; it predicted my patience. I applied the same logic to cricket, building expected runs (xR) and expected wickets (xW). Each delivery got four inputs: over number, batter's strike rate over the last twenty matches with a weight, bowler's current-spell economy, and the line and length the ball landed on. I then drew the historical probability of runs and wickets from the previous three editions. I also checked the reliability of length tagging, because missingness there runs near 9 percent.

The baseline: powerplay (overs 1–6) xR 7.8 per over, middle (7–15) 7.2, death (16–20) 9.6. The tournament's actual run rate finished at 7.8, so the headline matched. The distribution inside did not. Powerplay actual was 7.3, half a run below expectation. Death actual was 10.2, just over half a run above. Anyone can call this a low-scoring tournament. The structure says teams were scared early and reckless late.

The wicket picture is sharper. In Australia in 2026, an average of 6.8 wickets fell per innings. In 2026 that number was 7.9. Two gaps together, fewer runs early and more wickets throughout, mean one thing: front-loaded pressure. A side that could not reach 40 to 45 in the first six overs was effectively out of the innings, because pushing for runs in the middle cost wickets.

The phase table (per innings, 2026): powerplay runs 43.8, middle runs 64.8, death runs 51.0. Powerplay wickets 2.1, middle wickets 3.2, death wickets 2.6. Now hold that against one fact. Powerplay wickets above two, middle wickets above three. Normally the powerplay costs fewer wickets because the fielding side is attacking. Here it is the reverse. New-ball spells in this tournament were bowled to take wickets, not to contain. One sentence explains half the tournament's tactics.

Afghanistan: the largest deviation from baseline

Germany did not lose to South Korea; they lost to 28 shots and no goals. Afghanistan did not beat New Zealand by 84 runs through miracle. They beat them because they presented a length map on which there was no ball to drive. Afghanistan made 159 for 6; New Zealand were bowled out for 75. My xW model gave that surface a 0.6 percent chance of an all-out for 75.

Filtering the ball-by-ball log shows the mechanism. New Zealand batters were forced to leave 31 of 79 deliveries, a 39 percent dot rate. When the Afghan spinners bowled, average bounce sat just below waist height with a line starting outside off and shaping in. Hitting that line requires playing late, and playing late produces mis-hits. I isolated the fielding contribution separately: 11 runs saved. The tournament's average fielding residual was about 4. On that night alone, fielding added seven runs.

There is a trap here I try to avoid. Afghanistan bowled well, yes. But "bowled well" is not a measurement. The measurement: expected wickets 7.1, actual 10, a pure new-ball pressure surplus of about three wickets. How much of that surplus is pitch, how much is skill, how much is New Zealand's decision-making, I cannot split with this sample. I put that limit in front of the reader rather than hiding it.

Against Australia, the same architecture. Australia were bowled out for 127, and seven of their eleven batting innings finished below a strike rate of 100. Gulbadin Naib's spell of 4 for 20 in four overs had an xW of 1.7. Two wickets in four overs is not a per-match event, it is a per-tournament one. That makes it rare, not mystical.

India's middle overs: silence with a home advantage

In 2026 I counted the silence and found it had a home advantage. India's innings in 2026 show one striking feature. Rohit Sharma and Virat Kohli struck at around a run a ball in the powerplay, yet the side never collapsed through the middle. India's middle-overs run rate was 8.1 against a tournament average of 7.2. That 0.9-run gap produced 201-run totals, which in a low-scoring tournament is close to unassailable.

Against Australia in the Super Eight, India made 205 for 5; Australia stopped at 181 for 7. I mapped the fielding restrictions. Of the 48 balls India bowled in the middle overs, 29 had a fielder on the boundary rope and seven had four men in the inner ring. That aggressive ring layout slowed singles, and slow singles halve a chase. The structure is repeatable, not personal. India made 119 against Pakistan on a difficult surface; Pakistan stopped at 113 for 7.

In the semi-final, India made 171 for 7; England folded for 103. England's dot-ball rate was 46 percent. England did not lose that match to batting failure so much as to a failed plan. They tried to play the Indian spinners into the leg side, but the ball was shaping away, not in. Nine catches behind the wicket tells that story.

South Africa: where the death plan and the run rate separated

South Africa's death bowling ran ahead of baseline all tournament. Their economy in overs 16 to 20 was 8.3 against a tournament average of 10.2. Defending 163 for 6 against England in the Super Eight, they won by seven runs. That is not luck, it is an entry: expected runs in the last four overs were 38, actual was 26.

Back to the final. Why my 77 percent was wrong matters more than the number. My model assumed Klaasen and Miller at the crease could sustain 8.8 per over. But Jasprit Bumrah and Hardik Pandya's death economy in that spell was 5.2 and 6.0 respectively. My model underweighted bowling context. That is a model failure and I logged it.

There was also something no model captures: running between the wickets. Two run-outs occurred, one late. A run-out is not an entry-level variable; it is haste, lapsed concentration and nerves combined. I have no expected value for it, so I run the model without it and register that omission as a limitation.

Pakistan: where the baseline broke hardest

Pakistan's group matches showed a pattern I have rarely seen at this scale. Powerplay run rate 6.3, a full run below the tournament average. Middle overs lifted the strike rate but wickets fell at 0.22 per over. Against the United States at Nassau County both sides tied on 159, and Pakistan lost the Super Over.

The eye test is a witness; the data is the cross-examination. The eye test says Pakistan played badly. The data says that on that surface, losing four wickets inside the first six overs for 35 drops your win probability to 23 percent, and yet the match stayed alive to the last over because of the bowling. Pakistan's Super Over plan was the exact inverse of their 20-over plan, meaning they overwrote their own performance model.

One new addition I built across the tournament: a powerplay run conversion ratio. Roughly, an innings becomes defensible when powerplay runs, divided by four, exceed the wickets lost. Pakistan's ratio was 0.86. India's was 1.42. South Africa's was 1.31. This single ratio tracks almost every result in the tournament, and it is the most usable thing I built.

Bangladesh: the true weight of a Super Eight place

I will not dress Bangladesh's Super Eight run as a cultural triumph. The data across those matches says otherwise. Powerplay economy was 6.7, excellent. Death economy was 11.1, very poor. The whole tournament lives inside that gap. With a new ball they stayed in matches; with an old ball they left them. Chasing 196 against India they stopped at 146 for 8. Against Afghanistan they were bowled out for 105 in a match decided by eight runs under DLS.

I computed one number broadcasts never show: the share of deliveries faced by Bangladesh's top order after the sixth over. In the middle overs, 18 percent of balls they faced were against spin. India's figure was 27 percent, Australia's 24. Bangladesh's batters leaned on the sweep and the short-arm play against spin, and the sweep is the riskiest shot in the book. The rise in mis-hits follows from that choice.

Bangladesh's fielding residual was actually positive, around seven runs saved, which looks good on paper. Seven runs in the field disappear against what they conceded at the death.

Contrarian: the pitch story is half true

Now the uncomfortable part, where I testify against my own baseline. For the first two weeks the consensus was that the pitch controlled everything. I ran a placebo test on that claim. Method: I removed the eight matches where both sides were bowled out for under 130 batting first, then recomputed run rate across the rest. The run rate moved by just 0.2 per over. The pitch was not the whole story; bowling and match-ups spoke louder.

Another trap: the toss. Many claimed that winning the toss and fielding decided matches. I checked. Teams batting second won 49 percent of matches. A fair coin gives 50. The toss was close to inert. What actually worked was the ability to take wickets while the ball was new, paired with a disciplined if unglamorous boundary-ring configuration.

I do not chase narratives; I build a table and wait for them to arrive. But I also accept that some narratives become true before they enter the table, because narrative itself is a performance variable. The pressure on South Africa in that final does not enter the model, yet it entered their batting. My honest answer is that I have no measurement for that moment, so I record the absence of measurement as the finding.

Expected wickets against actual wickets: the team gaps

This table shocked me most, so I did not trade it on Twitter at the end of the tournament; I kept it in my notebook. Across the tournament, expected wickets against actual wickets by team: Afghanistan, xW 48.2, actual 72, gap plus 23.8. South Africa, xW 64, actual 71, gap plus 7. India, xW 66.5, actual 69, gap plus 2.5. Pakistan, xW 59, actual 48, gap minus 11. Sri Lanka, xW 52, actual 44, gap minus 8.

Afghanistan's gap is roughly 24 wickets. It is tempting to file that as very good bowling. I suspect a portion comes from mislabeled data, because I remain unsure the spinners' lengths were recorded correctly in that feed. A model that is not certain should treat its best result with the most suspicion.

India's gap is just 2.5. That tells me India won no lottery. They stayed honest to their own baseline, and that honesty was worth slightly more than the alternative on final night.

Actual against expected runs, phase by phase

One more table makes the tournament obvious. India, powerplay: xR 48, actual 46. Middle: xR 56, actual 63. Death: xR 60, actual 69. Pakistan, powerplay: xR 47, actual 36. Middle: xR 53, actual 54. Death: xR 51, actual 45.

The difference lives in one place: the powerplay. India scored two fewer than expected; Pakistan eleven fewer. At the death India scored nine more than expected, Pakistan six fewer. Had Pakistan's eleven powerplay runs been banked for the death overs, they would likely have made the Super Eight.

I like this analysis because it leaves no room for praise or contempt. Nobody is good or bad; someone finished a phase nine runs short, and that decided the match. Very few things in cricket explain themselves so cleanly.

The rehearsal of silence: what I watched without seeing

For years I have pulled the ball-by-ball feed beside me during every tournament match. The habit has a cost: I can never simply watch cricket. On the night Afghanistan beat Australia, I already knew Australia's death economy was 9.8 before a ball was bowled. After the match a friend asked whether I had seen how brilliant it was. I realised I had seen that passage only as numbers.

I accept that price. Without it I would not have recognised the 77 percent trap. Staring at numbers makes it easy to believe everything is explained, and that is the worst greed of all. A table is more honest than a human account, but a human builds the table, and humans err.

Takeaway: signals for 2026

The 2026 T20 World Cup will be played in India and Sri Lanka, spin-friendly, low-bounce, minimal grass. Three signals I am willing to bet on. First, the powerplay matters more, because a new ball spares batters from turn, which is the biggest cushion for Bangladesh, Sri Lanka and Afghanistan. Second, my death-over xR baseline will sit above 10, because short boundaries in India amplify clearing the rope. Third, fielding residual gains weight, because higher friction reduces fast singles and the boundary restriction becomes the only brake.

One more signal for the file. Sides that raise the number of deliveries their top order faces in the middle overs, meaning those who keep the top order at the crease longer, will reach the 2026 semi-finals. For Bangladesh that is the best and the hardest path at once, because that number can be coached, but not overnight.

Look back at that Kensington Oval evening this way: 26 needed off 30, six wickets in hand, a 23 percent chance. Probability is never destroyed. It merely becomes real because someone makes a mistake in some over. My job is not to announce verdicts after the match. My job is to price that mistake correctly the next time.