What the Empty Cells Confess: Bangladesh's Death-Overs Crisis Is Not the Bowling

At the 2026 T20 World Cup, while Bangladesh were batting in the Super 8 stage, I kept drawing a new column in an old spreadsheet from my home in Rangpur: “balls faced by batters four to seven in overs 16 to 20.” Drawing the column took ten seconds. Filling it took weeks of stubbornness. The cells were not blank — blank would have been a mercy. They were almost empty, and what few numbers sat there were far too thin to argue about.

Bangladesh played seven matches in that Super 8, won three and lost four. Those numbers are not in dispute. But the phase Bangladesh shouts loudest about — the death overs — rests on data so shallow that the argument has been circling the same drain for years. I am not going to break that circle today. I am going to measure its diameter.

I have watched matches for a long time. In 2026 I opened the batting and kept wicket for Udity Club in the Dhaka league. Then coaching, then, in 2026, a move into the BCB media set-up — The Daily Star called it “the fine cricket writer turned media manager” — and then back to the press, writing analysis. That road gave me one habit above all: when the eye makes a claim, check it against the notebook. The eye says Bangladesh are bad at the death. The notebook says Bangladesh rarely reach the death. The gap between those two sentences is the subject of this piece.

Bangladesh's only genuine laboratory for T20 cricket is the Bangladesh Premier League. In the eight or ten weeks of January and February, domestic cricketers face more T20 balls than they do in the remaining eleven months combined. Fortune Barishal won the title in 2026 and again in 2026 — that is not a model output, it is a fact. But winning consecutive titles and building an international batting core are two different jobs, done by two different people, on two different clocks.

In 2026 I opened a blank spreadsheet and let a domestic league teach me — then it was football, the Bangladesh Premier League, 132 matches, 3,410 shots, my own distance-and-angle weights because no public expected-goals model existed for that league. A 9.4-goal gap surfaced, and three betting syndicates emailed me within a week. That experience taught me two things. One: the cruder the domestic data, the more decisions rest on reputation. Two: what cannot be measured gets discussed morally, not analytically.

Coming back to cricket, I opened the scorecards from Barishal, Khulna and Dhaka and applied the same method. The scoring bands of the Sher-e-Bangla National Stadium in Mirpur, the Sylhet International Cricket Stadium and the Zahur Ahmed Chowdhury Stadium in Chattogram are so different that a single strike-rate figure compared across tournaments is close to meaningless. Sylhet can host a 210 chase; Mirpur may defend 155. Yet nobody at an auction discounts for venue. A batter's price rises on his name and a fat average, not on matchup value.

That is the first layer of missingness. BPL ball-by-ball data gets resold, but phase-level and venue-level matchup data — which batter did what against which bowler in overs 16 to 20 — is never published. Who collects it? A live-scoring vendor, two betting-feed companies, and a handful of in-house analysts at academies. The data exists; ownership of it and the power to act on it sit in different hands. The people who pick teams do not see the data. The people who see the data do not pick teams. The emptiness is not accidental. It is structural.

Now let me open my own model. It is crude. The sample is small, so every figure carries a wide error margin, and I will not hide it. I split a batting innings into three phases — powerplay (1-6), middle (7-15), death (16-20). I counted the bowling side directly off the match. When I tried to count the batting side, the empty cells walked to the front of the room.

On the powerplay, Bangladesh's numbers are coherent. They are not mid-table stragglers in the first six overs; against the top eight or ten sides in the competition, they compete in this phase. The story changes in the middle overs. Between overs 7 and 15, Bangladesh's strike rate sits clearly below the tournament's leading sides, while their wicket-loss rate is also lower. They move slowly and they keep wickets in hand. I am not calling that strategy wrong. I am calling it insurance — and insurance is paid for in instalments, at the death.

If I count the balls actually faced in overs 16 to 20, Bangladesh sit near the bottom of the significant sides in my model. The reason is obvious. When five wickets have fallen by the 17th over, the set batter does not have a licence to swing. The man who comes in at six stands at the non-striker's end and watches three-over blocks vanish. In the innings ledger, he is marked down as having failed. That is not a number about him. That is a number about a place he was never allowed to occupy.

Here is the statistical puzzle. My crude model confused me, because in trying to measure Bangladesh's batting failure it ended up proving that Bangladesh's batting opportunity had shrunk. The death-over problem is not created in the death overs. It is created in the stubborn savings policy of overs 7 to 15. We blame one phase for an outcome that is really the residue of the phase before it.

My model was crude, but the empty cells confessed more than the runs. The empty cell asked me a question: is Bangladesh's death batting bad, or is it unpractised? Those are different diseases with different treatments. If it is bad, it is a skill problem, and a bigger sample will not save it. If it is unpractised, it is a strategy problem, and dosage will. With the data I have, I cannot choose between them — which is exactly why the argument has been frozen for eight years.

By now I watch Bangladesh twice: once with my eyes, once with phase splits. The eye says the boy standing at the non-striker's end looks vacant. The phase split says he did not bat in that over. Watching twice lets me separate my own eye's bad habits from my model's blindness.

The second layer of missingness is at the auction. In the BPL auction, top order is bought on runs and average. Suppose a No. 2 bats fifteen overs at a strike rate of 125. Suppose a No. 6 faces four overs at 160. In a real crisis the second man matters more, but the first man has more runs, a better average, a higher price. Who benefits from keeping the data incomplete? When the cells are empty, the decision falls back on television memory and highlight reels. That is where bias nests.

The stadium sits on one side of that bias too. If the Mirpur surface suppresses scoring, then a 125 strike rate at Mirpur is a better skill than 125 at Sylhet. But the venues change in an international tournament, and if venue weightings do not travel, the previous tournament's experience becomes useless. The 2026 cycle lands in India and Sri Lanka. Subcontinental tracks, yes — but not one track. India's surfaces can behave like one-day pitches, Sri Lanka's slower squares turn. Anyone picking a squad on domestic data needs to know which domestic data belongs to which venue first.

The third layer is subtler. There is a stock of spurious correlations sitting between domestic estimation and international output. “140 at 36 means you are a finisher” is a ladder many people will not kick down. In reality, a strike rate in overs 16 to 20 is not a fixed property of a batter; it is an innings-level variable — how many wickets are in hand, what rhythm the bowler is in, how much pressure the scoreboard carries, and which pitch it is on. In a World Cup knockout, those variables turn violent. Trusting a domestic strike rate as a proxy for an international finisher means controlling for venue, opposition and situation while pretending to read an innings.

So a number arrives from the BPL — and who translates it? Bangladesh currently have batters like Towhid Hridoy, who can hold a strike rate through overs 7 to 15 and also open up between 16 and 20. Litton Das has two distinct patterns, one lower down and one at the top. Shariful Islam's powerplay skill is a separate asset entirely. These are different tools, and the real work is seating the right tool in the right phase. I am not saying the tools are bad. I am saying the accounting is being done on incomplete data.

Which brings me to my most important reservation. I will not shout about the death-bowling line-up. Mustafizur Rahman, Taskin Ahmed, Tanzim Hasan Sakib — that unit is Bangladesh's comparatively strongest phase. More broadly, Bangladesh's death bowling is not worse than its rivals'. The model even hints they will survive this phase in the tournament. But bowling survives only if a batting side gets far enough ahead. And the batting-side crisis cannot be measured in a mobile model — it only casts a shadow in those empty cells.

Why has the finger been pointed at the wrong phase for so long? Because a phase-batting crisis is mathematically vague. A bowling crisis has a clear picture: runs conceded and wickets taken between overs 16 and 20. Viewers therefore stand in the last frame of the innings, where the name is written. The first frame — the obstinate blocking of the 11th over by two or three batters — does not catch the crowd's eye.

Silence is not zero; it is a new baseline with its own residuals. Bangladesh's death-over data is silent — and that silence says Bangladesh rarely get there. The argument is conducted over the silence, not over the numbers spoken loudly.

So what is the fix? I will throw three proposals, and none of them is brilliant. First, change the primary selection index. Not “how many runs does he score with many balls,” but “how many balls did the No. 4 to 7 face in overs 16 to 20.” Nobody measures this now. Start measuring it and some interesting names may rise, and some current names may fall. Second, make the savings policy of overs 7 to 15 explainable. If a side can lift its middle-over strike rate by five to seven, that returns as five or six extra balls at the 17th over. Third, target subcontinental tracks in the India-Sri Lanka cycle — weight the batters whose strike-rate gears do not change across those conditions.

Now I want to surrender one liability. In an earlier piece I saw Germany's pressing decay before a tournament — their PPDA drifted from 8.9 to 12.6 in qualifying, and they went out in the group stage. Forty thousand people read it. But my model ranked them third-favourite, so I hedged the text and lost the argument on results anyway. That taught me what an honest model means: it is one with an appendix at the end, reading “here is where my model could be wrong.” This piece's appendix sits below.

The appendix, or where my model could be wrong: I may be holding the venue effect too large when the real effect is small. I could not capture the magic of Fortune Barishal's consecutive titles in the data, which may spread bias. I reduced the batter-versus-batter option to a single dimension — in reality cricket contains sliding doors I do not model. And my biggest weakness is silence itself. On a small sample a number always tempts, and I can be wrong too.

Yet one thing I trust. This crisis is not effort-based, so it will not be fixed by more sweat. It is a definition crisis — we are calling something vague a bowling problem. Modern T20 batting is entirely a balance between wicket preservation and attacking capacity. Bangladesh has leaned on the insurance side for a long time. The trouble with insurance is that it never returns the full premium; it only reduces the loss.

And a model is a monastery: you enter to escape the noise, then hear it clearer. I have closed the empty column for now. In the 2026 group stage, a handful of innings will show whether Bangladesh's No. 6 is standing at the non-striker's end or walking out to bat. Perhaps then it will be proved that the long argument was never about bowling. It was about representation we never measured.