Teaching a League to See Its Own xG: What the BPL Scorecard Hides in the Powerplay

In a BPL match last season, a side made 174 in twenty overs and lost. Three nights later the same side made 142 and won. The scorecard calls that T20 cricket being T20 cricket. The shot map calls it two unrelated stories.

Of the 174, fifty-eight runs came from six edges and two top-edges. Forty-three balls were played on lengths where my model gave that batter an expected value below 0.4. Of the 142, ninety-one runs came off the middle, grounded and timed. In the Mirpur press box I heard the usual verdict: 174 was a good score, the luck went the other way. Someone added that cricket rewards talent but still needs fortune.

I wrote a question in my notebook: if the scorecard cannot separate luck from skill, how does the league recognise its own best players? In Bangladesh, I taught a league to see its own xG. That was football. Cricket is next, and the road is far rougher.

Context: building a model where the data does not exist

In 2026, aged twenty-four, I joined the Dhaka outlet Golpo Sports as a junior data analyst from my flat in Rajshahi. I treated data as scripture. I hand-coded 1,248 shots from the 2026-17 Bangladesh Premier League in football: which foot, which body shape, how far out the keeper was, the delivery type. The result was unambiguous. Abahani Limited Dhaka scored 34 goals from 27.6 xG. Sheikh Jamal Dhanmondi scored 29 from 31.2 xG. Same league, same season, two opposite stories. I published a twelve-part series on shot quality. Traffic doubled and my xG table became a weekly fixture.

StatsBomb noticed the series, and in 2026 I joined the Russia World Cup as a remote event data analyst. In Germany against Mexico I logged Germany's 26 shots for just 1.3 xG, while Mexico's 12 shots produced 1.1 xG. Germany's PPDA was 6.9 — pressing high, but leaving eighteen transition chances behind them. I shipped the model before the final whistle: Germany would not escape Group F. Germany finished bottom. PPDA showed me Germany.

Dragging the same logic into cricket, the ground moved. In football one shot closes an attack and the event is singular. In cricket every ball is an event, tangled with length, match-up, field setting, pitch behaviour, dew and umpiring. The BPL has no permanent tracking rig, ball-by-ball speed data is not uniformly available, and scoring is largely volunteer-driven. Assume the infrastructure already exists and the model becomes decorative, useful in a slide deck and useless on a dressing-room wall.

So I built the pipeline backwards. An ESTJ builds the pipeline first and the poetry second.

Core: run quality, not runs

My first decision was unpopular: do not start with runs. Start with shot value. In the football xG model I used four inputs — shot location, delivery type, defensive pressure, body part. In cricket those became pitch length, ball line, batter match-up (left-right, spin-pace), and shot type (grounded, lofted, edged).

Ball-by-ball data did not exist for me then. So I coded from video, by hand. The matches I watched from Mirpur taught me to split the powerplay into four zones: stump line, off-stump corridor, wide line, body line.

The picture that emerged contradicts the league's working belief. A BPL scorecard counts runs but not the source of runs, so the table rewards fortune and punishes the wrong information.

In my model the gap between expected and actual powerplay runs was widest in two kinds of innings. First, sides that made 45-plus in the powerplay with more than sixty per cent of those runs off edges and balls outside the arc. Second, sides that made 35 to 40 but averaged above 1.1 in shot value, because they were batting on a surface where low targets were rational. The first group sits high on the table. The second sits low. On match conditions, the second group batted better.

Second decision: measuring bowling as pressing

A cricket analogue of PPDA needs a definition first — what counts as a press. In football PPDA measures how many passes a side allows before contesting. Cricket has no possession; every ball is contested. So I fixed three markers up front, published them, and let them be argued with:

First, the share of non-boundary dot balls in the powerplay. It is the cruellest measure of pressure on a batter, because the scorecard records it as a zero.

Second, the percentage of balls a bowler lands on the stumps in those six overs. Bowling wide is sometimes craft, sometimes fear.

Third, the field setting — how many fielders inside the ring, how many deep. That is a proxy for the bowler's conviction.

Combined, they show something counter-intuitive: the side conceding fewest powerplay runs is not always the side taking most dot balls. Sometimes it bowls the highest share of wide-line balls, and against timing-dependent pairs on a slow, low Mirpur surface, that works. The Mirpur wicket keeps returning to this conversation.

Third decision: measure the running, not only the runs

Empty stadiums are a permanent lesson. In 2026 I consulted for Brentford and analysed 306 behind-closed-doors matches across the Bundesliga, Championship and Serie A. Home win rate fell from 43.1 per cent to 33.8. Home xG differential dropped 0.21. Distance covered in the final fifteen minutes fell 5.2 per cent. I called the adjustment CrowdNull, and Brentford used it to rewrite set-piece routines before pushing on in their promotion campaign. Empty stadiums taught me that home advantage is a variable, not a law.

In cricket the translation is running between the wickets. In tournaments played to empty stands in 2026, the appetite for twos after the eighteenth over fell. Small sample, short innings — but the signal is clean. With a crowd, who supplies that extra second of running? The fielder. Without one, that second returns slowly to the fielder as well.

That observation does not transfer cleanly. Cricket lacks football's GPS coverage. So I timed four venues from match video with a stopwatch and a spreadsheet. In my model the relationship between powerplay pressure score and attempted run-outs at the death is visible, but it is correlation, not cause. That distinction matters, otherwise someone will next season announce that returning crowds will improve Bangladeshi fielding by six per cent. That is a guess, not a finding.

Fourth decision: what happens when a coach holds the model

I put the sheet on a coach's desk. One page — shot value per batter in the powerplay, length zones, dot-ball share. He read it for three minutes and said: your sheet says my best batter was poor today. But I watched him stand on his knee; the footwork never arrived.

That sentence stopped me. A model can read an innings and cannot read a body, and in the accounts of players returning from injury, that is exactly where the largest trap sits. A bowler coming back from an ACL shows different economy in his first five matches and his next five, but the month in between is written nowhere on a scorecard. The mental block is harder to fix than the body, and nobody has built the instrument.

Contrarian: a model that testifies against itself

The largest risk in this work is enthroning xG. In Russia I called Germany's exit before the final whistle, but the model knew nothing about the dressing room — who was carrying a knock, who had stopped talking to whom, who had slept. If that is true in football, it is truer in cricket.

An xG-style index cannot explain umpiring standards, form or in-game decisions. A top edge that lands two inches inside the rope is not counted as a miss by the model; the scorecard counts it. If a batter abandons the slog sweep, the model debits the batter and credits the bowler. The model sees both; the table shows one. That is why I never write that a batter did not deserve a hundred. I write that his average shot value was this, and the surrounding innings averaged that.

The second risk is specific to Bangladesh. Our data environment is not built; it must be co-designed. Scorers, coaches, video operators, club management — without a shared collection structure, every model stays a mirror and never becomes a window. Trophies are won on the field, decisions are made in the dressing room, and the consequences surface three seasons later in an age-group side. Getting there starts with a pipeline, not a playbook.

The third risk is contrarianism as a habit. If someone writes the opposite of consensus every week, that is personality, not analysis. My rule is simple: publish the base rate first, pre-register the hypothesis, then look at the data. The analyst does not chase revelations; he calibrates until they appear. When someone draws a conclusion about Tamim Iqbal or Litton Das from one innings, I go back and cut the length zones from their previous ten. Usually the answer does not hold. A middle-overs read on Mushfiqur Rahim or Mahmudullah has to be match-up based, not run-rate based; and at the death, Mustafizur Rahman's or Taskin Ahmed's cutter only works when the ring is set correctly — the setting matters more than the name.

Takeaway

For next season I am releasing one signal in a limited way: the first four matches of the table will lie to you. In that window, look past points and check powerplay dot-ball share against clean-contact share. A side with strong expected powerplay runs but a low total will climb within three weeks.

And to anyone who thinks xG in cricket is a dressed-up story, one question: if the scorecard is the truth, why did the 174 lose and the 142 win?