The BPL's Own xG: The Truth 1,248 Shots Hide From the Scoreboard

The Mirpur gallery had not filled up yet. Six overs gone, the scoreboard read 52/1. A former opener sitting beside me shook his head and said, “Good start.” I looked down at my sheet — my model had expected 41 runs from those same six overs. An eleven-run shortfall that nobody could see, because the scoreboard was not showing that number. It was hiding it.

Four balls in that powerplay died in front of the boundary rope, balls that would have cost two had the fielder stood three metres deeper. Two half-chances slipped out of ring fielders' hands. None of it leaves a mark on a scorecard. Yet the match's direction had already been written in exactly those four deliveries.

From that night an old habit died: I stopped writing match reports off the scorecard. Shot quality first, story second.

Context

Domestic cricket data in Bangladesh still means runs, balls and overs. A scorer's pen and paper, two cameras, and no record at all of where five fielders were standing. So when someone in this market uses the word “analytics”, my first question is: with which data? Labelled by whom?

In 2026, aged 24, I joined Dhaka-based Golpo Sports as a junior data analyst. The task was simple, even if it did not feel that way then: code 1,248 shots from a single BPL season — who bowled, what line, what length, how the field was set, how hard the shot was, what the crowd did. It took four months. That patience later became my profession.

The first realisation from that set: Abahani's 34 goals came from 27.6 xG, Sheikh Jamal's 29 from 31.2 xG. Read together, those numbers say the league rewards something different from what it believes it rewards. In Bangladesh, I taught a league to see its own xG; the work started there.

Cricket has no direct xG. So a mapping had to be built: every delivery needs an expected-runs value — batter's shot type, bowler's line, field placement, pitch character, whether it is a powerplay ball. xG here does not mean goal probability; it means how many runs this delivery has historically produced. Whatever the name, the question is one: what is happening outside the scoreboard?

Core Analysis

The powerplay model's real job is not predicting the score — it is measuring waste.

My powerplay model needs four variables to produce expected runs across six overs: the bowling type in the first two overs (swing versus seam with the new ball), the openers' three-season strike-rate baseline, the position of the fielder in the cover-point region inside the fielding restrictions, and ball movement. The number those four produce is not an average — it is an expectation. And the gap between expectation and reality is the real story of a match.

Across three seasons of domestic data, only two of the top five powerplay teams finished in the top four of the points table. In one season, the side ranked second for powerplay runs finished sixth. The reason is not complicated: sixes come from risk, and risk gets paid for in wickets. The true currency of the powerplay is not the six — it is the rate at which dot balls are avoided.

This is where my football background earns its keep. PPDA showed me Germany. At the 2026 World Cup, Germany versus Mexico: Germany's 26 shots produced just 1.3 xG, Mexico's 12 shots produced 1.1. Germany's PPDA was 6.9, meaning 18 transition chances were left open. Before the final whistle I wrote that Germany would not escape Group F. They did not.

What is cricket's PPDA equivalent? My answer: pressing means the density of pressure created in the field. In cricket that pressure comes from the number of ring fielders and strings of consecutive dot balls. So I built an index — the Dot Chain Index. Each over, the number of dot balls is multiplied by the number of fielders inside 25 yards. A high chain means the batter is under pressure; a low chain means the batter is in rhythm.

What emerged was counter-intuitive. Teams with a high dot-chain in the middle overs do not see their death-over run rate fall — it rises. As pressure accumulates, batters are forced to take risk in the last five overs, and two wickets in one over collapses an innings. What captains call economy control, the data calls pressure storage — and storage demands repayment. An opener of Liton Kumar Das's type leaving balls alone in the powerplay looks passive from the outside, but in the model it reads as balance held back for future runs.

One more thing has to be cross-checked. In 2026, during the empty-stadium period, I calibrated home advantage for Brentford across 306 behind-closed-doors matches. Home win rate fell from 43.1% to 33.8%, home xG differential dropped 0.21, and distance covered in the final fifteen minutes fell 5.2%. Empty stadiums taught me that home advantage is a variable, not a law. In cricket, crowd pressure feeds directly into fielding footwork — the dive, the run-out call, the timing of the catch. Fail to adjust this variable and any model drifts the wrong way.

I keep a separate calculation for death overs. Economy in overs 17 to 20 cannot be explained by bowling alone. For each bowler I wanted a fatigue index — average time between boundaries, delivery-to-delivery gap, bouncer usage in the last two overs. It is the cricket translation of football's distance covered in the final fifteen minutes. Applied to death bowlers of the Taskeen Ahmed or Mustafizur Rahman mould, it shows why over seventeen needs a different bowler from over nineteen.

The model has a second use nobody has taken up yet: a waste map instead of a radar chart. For every innings we calculate which six-over block surrendered the most expected runs. Across the last three seasons, the biggest waste came in overs seven to ten — the phase where batters want to settle and spinners turn the ball. If the dot-ball ratio rises in those four overs, the probability of coming back into the match drops by roughly a third.

One layer had to be bolted on: fielder positions. Nobody keeps this data in Bangladesh, so I began keeping it myself — a screen capture per ball, every thirty seconds. Around that I co-designed a sheet with the scorers. Finding the intersection of what a scorer can record and what an analyst needs is the first job. Because the pipeline comes before the model. Here is my personal rule: an ESTJ builds the pipeline first and the poetry second.

Validation has to stay plain. Build the model on 2026-17 shots, then test it on 640 balls from the following season. Where average error sits below 0.11 runs per ball, the model is usable. Without that discipline, analytics becomes an attractive wall chart that cannot change a single result.

For selectors this has direct meaning. The auction buys the batter who can take fifteen off an over. The league wants the batter who does not follow three balls with three zeros. If the model and the auction speak different languages, the squad is built right and played wrong. In Bangladesh, I taught a league to see its own xG — that is what it means: making the language of the scoreboard and the language of the data one. The biggest lesson of that work is that he does not chase revelations; he calibrates until they appear. I now write every hypothesis down first and open the data second, so the temptation to build a story out of a pattern gets cut off.

The Contrarian Angle

Now the other side. A model is a mirror, not a prophecy.

A model saying a side should have won on xG is only saying that, on the historical budget, that innings usually produced more runs. That is not a verdict. Outcome variance in cricket is far higher than in football, because one ball can remove an opener and the second new ball can turn a match. I have watched xG abused at close range — people use it to explain in-game decisions, form and umpiring standards, which is not its job.

Another point must be remembered: the definition of pressing in cricket is not as simple as in football. In football, PPDA measures how many passes the opposition completes before pressure arrives. In cricket, pressing on the move is impossible, because fielders do not chase the ball. So what gets counted is dot balls and ring-fielder density. Without writing that definition down, the numbers can be bent anywhere — and that is the greatest danger in analytics.

The second trap: in a domestic league, what the data cannot capture is far bigger than what it can. Who prepares the Dhaka pitch, who cuts it on which day, who is fit before the auction and who is not, who gets how many chances in the age-group pipeline — these are not data, these are power. Pick a squad on data alone and the dark room you leave out is actually half the story of Bangladesh cricket.

The third trap is my own temperament. Finding something counter-intuitive feels like a reward. So now I pre-register every hypothesis, publish the base rate first, and only then say where the exception lies. Otherwise there is no difference between an analyst and a team's media officer.

Fourth: assuming the data infrastructure exists. Where four-camera ball tracking runs, Bangladesh's primary sensor is a scorer's pen. So models have to be co-designed, or practitioners will treat them as somebody else's next-day report. A model must be their mirror, not a slide in someone's lecture.

Takeaway

Next season, do not look at a team's six count in the powerplay. Look at the dot-ball rate between overs seven and fifteen, and how many fielders stand inside 25 yards during that window. If those two numbers live inside a side, the points table will do the talking itself.

And the question is the same for everyone: is your league ready to look at its own mirror?