What the Scoreboard Hides: A 66-Match Ledger, the Decay of Home Advantage, and the Signals for Bangladesh's Next T20 Cycle

Hook

Sher-e-Bangla National Cricket Stadium, 7:30 pm. The target was 162. The match ended in the 19th over, seven balls to spare, six wickets in hand. The scoreboard recorded a comfortable win; the stands recorded a photograph of laughter. On my laptop a different number was burning: 148/9.

That was not a team's score. It was my model's expected outcome for that pitch, those three batters, against that attack. The side that won that night, by my numbers, had fallen below a 29 percent win probability before the final ball. The win came from two free hits, one top edge and one dropped catch.

I am not writing this to shrink the win. I am writing it because almost every judgement we make about Bangladesh's T20 cricket still comes from the same place — what the scoreboard showed. Where the process went is a question that still does not get raised in our meetings.

Context: where the ledger came from

In 2026 I was in Rajshahi. My first job on a Dhaka digital desk paid BDT 18,000 a month, and my first assignment was to chart an entire Bangladesh Premier League season — all 66 matches, ball by ball. Shot location, body part struck, defensive pressure, wicketkeeper position, field placement. By Week 6 I understood that paper would not survive, so I rebuilt the sheet in Python.

The first pair of numbers that came out of that sheet still hangs on my desk wall: Abahani Limited Dhaka outperformed their expected goals by 11.4, and the league table showed them as champions. Nobody in Bangladeshi football had published those two numbers side by side. From that day 'deserved to win' left my vocabulary, and a methodology footnote appeared under every column I filed.

In cricket I have used the same skeleton, only the unit changed. Cricket has no goals, so I built the network model as expected runs, or xR, per ball. Eight inputs for every delivery: the bowler's career economy and historical strike rate in that phase, the venue's phase baseline, wickets in hand, the batter's role index, the pitch turn index, the situational pressure of the match, time of day, and humidity. The output is an expected run value; summing every ball gives the match's expected score.

Cricket has one advantage here: ball-by-ball data is far denser than football event data. In football you may log 1,200 events in 90 minutes; in a T20 there is a clear outcome on nearly every one of 240 deliveries. The spreadsheet does not lie to me, but it also does not explain itself — the explanation has to come from reading it against what the ground actually did.

In April 2026 my desk cut 40 percent of staff and my contract dropped to zero hours. I built my own scraping pipeline. When the Bundesliga restarted in May, I tracked 306 matches across five leagues. The result was clean: the home win rate fell from 43.2 percent to 33.6 percent in empty stadiums, and home xG dropped 0.11 per match. I published the dataset with the code attached and licensed it to two Asian outlets.

That experience taught me a habit: before a number goes to print, its collection method, sample size and context go with it. Read the rest of this piece under the same condition — every figure is from my own charting, and the limits of the sample are stated at each step.

Core: five threads, one pattern

Thread one — powerplay myopia and the arithmetic of late cameos

Charting Bangladesh's T20 matches ball by ball between 2026 and 2026 produces an uncomfortable picture in the first six overs. Our powerplay strike rate sat at a level only marginally better than sides like the United Arab Emirates or the Netherlands in the same phase. Yet in the last five overs our scoring rate is frequently higher than the opposition's.

Many call this the ability to accelerate late. Numerically it looks less like ability and more like correction. Through the middle overs we bat at a strike rate that requires a boundary roughly every two overs; the pressure accumulates and bursts in the last five. The trouble is that late-over scoring works in league play, but in knockout cricket, when the opposition fields two death specialists, that bursting space is sealed in advance.

Take a 2026 sample from my sheet: of the matches where our powerplay total was under 40 with no wicket lost, 58 percent still went off-equation by the last five overs. A slow powerplay does not invite defeat on its own, but it drives the error tolerance of the next two phases to zero. The sample is small, so this is a hypothesis — but it is the hypothesis the next cycle should test.

Thread two — venue baselines, and the gap between result and process

Bangladesh's three main venues — Sher-e-Bangla in Dhaka, Zahur Ahmed Chowdhury Stadium in Chattogram, Sylhet International Cricket Stadium — produce three different games. Across the ball-by-ball logs I could gather from 2026 to 2026, the combined sample approaches two hundred matches. Their phase baselines are nowhere near equal.

In Sylhet the expected runs per ball against spin in the middle overs is lower than in Dhaka, but at the death the boundary dimensions and the outfield push the value of fours and sixes up. Chattogram has a higher first-innings average than Dhaka, and a sharper second-innings fall. Pool the three baselines and you get this: the same total means three different things. 155 is respectable in Dhaka, ordinary in Sylhet, and often short in Chattogram.

That is where the result-versus-process gap opens — the gap I have hunted in every piece since Kazan, June 27, 2026, when Germany lost 0-2 to South Korea with 2.31 expected goals against 0.78. One loss on the scoreboard, one controlled match in the numbers. Cricket's closest equivalent is the win where the winning side trailed over by over and got home on the opposition's wickets lost and extra wides. By my count, between 2026 and 2026 there are at least nine such matches in which the winning side's expected runs were lower than the loser's.

This is my central claim: Bangladesh's T20 record is still not an accurate measure of our true strength and weakness, because the gap between the raw scoreboard total and a venue-adjusted expected total averages eight to eleven runs. Spread across one team's run, that gap is invisible; spread evenly across three or four matches in a series, it becomes points in the table.

Thread three — the workload ledger: who bowled how much

After 2026 I built a pipeline that places domestic league ball counts and international T20 and ODI ball counts side by side. This ledger is rarely printed, because clubs and the board keep separate books and nobody owns the job of reconciling them.

In my count, across a normal T20 season two of Bangladesh's top four fast bowlers cross a thousand deliveries when domestic and international loads are combined. The workload in cricket is rougher than minute management in football because training and travel are not counted separately. A fast bowler's shoulder, ankle and lower back do not accumulate load linearly across a three-to-four-week series — they accumulate in jumps.

I am not writing this to argue that rest decisions should be made on a spreadsheet. I am writing it because selection committees typically carry two numbers into the room: strike rate and economy. The workload index is not there. Yet death-bowling effectiveness tracks that index directly — bowlers who send down 22 to 24 overs in consecutive matches post the highest rates of wides and no-balls at the death. The relationship is stable in the sample, but it is not causal, and I say so plainly.

Thread four — auction price and the accounting of primary skill

In BPL auctions I have seen a recurring pattern that mirrors the one I see most often in football's transfer market. In football a goalkeeper who can strike a long kick is frequently valued above his core shot-stopping ability. Cricket's equivalent is the bowler who can hit sixes lower down the order — his auction price often rises above his per-match breakthrough or dot-ball capacity.

I have set two seasons of auction prices against the following season's performance index. Of the six players whose price rose furthest above what their primary skill predicted, nearly all were paid for at least one secondary skill: batting depth, fielding, or captaincy. Only two of them finished the following season among their team's top three performers in their primary skill. The sample is small, so this is not a verdict — it is a hypothesis needing two more seasons. The question stays: every transfer window is a ledger, and every rumour has a decimal point — whether the owners read that decimal point is the real question.

Thread five — the base rate of selection: the hidden cost of the safe pick

Our selection debate usually swings between two poles — experience against youth. In the numbers it is a question of variance rather than variety. Between 2026 and 2026, young batters given debuts showed a far wider gap between their first ten matches' strike rate and their next ten than established players did. They were not scoring less on average; they were scoring in larger swings.

In a short series where two or three matches decide everything, a high-variance pick puts the side at risk. In a long tournament with a seven-or-eight-match knockout path, that variance raises expected utility — provided the team can build a stable frame around it. In one 2026 combination I modelled, two senior batters plus three young ones finished ahead of an all-senior combination on expected points. That is not a large-sample result, but it belongs in the selection room.

Contrarian: correlation is not causation

The weakest point in all of the above is the one I will state first. The number I published in 2026 showing the fall in home win rates in empty stadiums has, for many readers, become proof that crowds are the single cause. That is wrong, and at least four confounders are operating at once.

First, pitch preparation changed in the same window — post-pandemic seasons generally produced drier, more spin-friendly surfaces, where home spinners kept their own edge but batting-friendly surfaces vanished. Second, the schedule was abnormally compressed; fewer rest days hurt the home side less than the touring side only in theory, because touring sides were not travelling. Third, opposition analytics units improved in this period; whether or not a crowd is present, the home pitch pattern now sits in written scouting reports. Fourth, the ratio of neutral-venue to home matches shifted.

With four variables moving together, one number cannot separate them. What I printed said this: in empty stadium environments, the home advantage index fell. Why it fell, that sample does not say. The spreadsheet never lies to me, but it also never turns my guess into proof.

A second caution about this piece: that first 66-match sheet is still my favourite, and that makes it the most dangerous. We look at a pattern across numbers and say the pattern across 66 matches is this. But 66 matches cannot separate selection, venue and variance into three distinct layers. Whenever a single hypothesis explains everything, the sheet has become a mirror of my own bias. For anyone who bets on forecasts, this is the worst pathology, because the pattern will sound sweet precisely when you are looking for a reason to be angry.

A third caution, against decontextualised numbers. No index is published in comparison without its collection method, its sample and the market incentives around it. I repeat this with every figure, because the biggest gap in Bangladesh's data landscape is not a shortage of numbers — it is a shortage of documentation about where the gaps are.

Takeaway: four signals for the next cycle

In the next T20 cycle I will watch four things rather than twenty matches. Powerplay strike rate, which if it stays under eleven runs an over keeps stacking pressure in the first two phases. Venue-adjusted expected runs for the home side, which still answers a question even though the empty-stadium question is old. The death-over ball-count ledger for fast bowlers, where crossing 22 overs produces a missing piece roughly every three matches. And the ratio of money spent to primary skill at auction. Twenty years have taught me one thing: the scoreboard is true, but it is not simple — and the number that refuses to admit its own error is the one that will betray you in the next series.