BPL's Own xG: The Mirror That Showed the League Its True Face

From the press box at Mirpur's Sher-e-Bangla Stadium, I watched the last ball of a match from block six. The scoreboard said Abahani Limited Dhaka had won by five wickets, and the reporter beside me was already typing his headline. My laptop said something else. Abahani's expected runs in that innings were 142.6; they actually made 134. They won that match not on shot quality but on three fielding errors and two death-over overthrows.

That night, back on my Rajshahi balcony, I opened a notebook and asked myself a question I had never framed this way in seventeen years of watching cricket: if the BPL is a reward system, whom is it actually rewarding? Technique, or the nerve of the last five overs? The gap between what the scoreboard says and what shot quality says became my work for the next two years.

In Bangladesh, I taught a league to see its own xG. That line sits on the first slide of every talk I give, because in early 2026 our cricket conversation was full of "deserved wins" and "unlucky losses" — words of emotion, not measurement. I wanted the numbers to speak, but in a language a Mirpur dressing room could actually use.

In 2026, at twenty-four, I joined the Dhaka new-media outlet Golpo Sports as a junior data analyst from my Rajshahi flat. My job was singular: code every shot of the BPL 2026-17 season. That was 1,248 shots. For each one I logged six fields: where the ball pitched, where the bat connected, shot direction, shot height, nearest fielder's position, and outcome. I treated the data as scripture. I did not yet know that this attitude would later teach me how easily data can be abused.

Within the first month I learned what data infrastructure actually means in Bangladesh. No Hawk-Eye. No ball-tracking. Two to four cameras per match. Scorers writing runs on paper, video files arriving two or three days late. A model built in London and copied blindly in Mirpur would fail. So my first decision was structural, not technical: sit down with scorers, coaches and the video operator and agree on a minimal collection protocol. Which six fields can everyone capture accurately, together? Fix that first, build the model second. I did this because a beautiful model that is not co-designed with the people at the ground will be abandoned within two months.

I called the model xR — expected runs. The logic is simple. What matters is not how many runs a shot produced, but how many it should have produced in the league's average environment. A cover drive on a slow, low Mirpur surface is not worth the same as the same drive on a flat Chattogram deck. So I built venue-specific baselines, and split each baseline into six-over, twelve-over and twenty-over phases.

One clarification, because I made this mistake once myself. When I began translating football's PPDA into cricket, I assumed that adjusting for pitch conditions and ball age would be enough. It was not. In Bangladesh's league, humidity, dew and evening light change the behaviour of the ball so much in the second half of an innings that the same batter plays the same shot for six in the first ten overs and holes out in the last five.

The series ran in December 2026, across twelve parts. The table in part one is still discussed today: Abahani Limited Dhaka scored 34 against a model expectation of 27.6; Sheikh Jamal Dhanmondi scored 29 against an expectation of 31.2. The units are the model's own. The interesting part is that the side topping the points table was not the side ahead on shot quality; the side that lost was the side with the more durable batting profile. The outlet's traffic doubled, and the xR table became a weekly fixture.

When shot quality and run production separate, the culprit is not talent — it is structure. Abahani's 34 against 27.6 says their batters played better than expectation; if that skill does not convert into a trophy, the question becomes how durable it is. Sheikh Jamal's 29 against 31.2 says the process is working while the harvest is not. Two types of teams need two types of decisions: one needs finishing resources, the other needs patience.

In part four I published shot maps. A pattern emerged that no scoreboard shows: four of the league's top five run-scorers took more than half their runs on the leg side, and much of that came over fine leg, exploiting the short square boundary. The league was quietly subsidising one type of stroke because of the size of its grounds.

In part seven I did the work that has served me best. I tried to translate football's PPDA into cricket. PPDA measures how many passes you allow per defensive action — a lower number means aggressive pressing. Cricket has no passes, so I first had to define what "press" means.

I kept the mapping simple. A fielding side presses when it pushes the bowler fuller and faster, brings catchers closer, and forces the batter off his preferred shot. Those three events I called "pressure events." The metric then became BPPDA: how many balls a fielding side spends to create one pressure event. I publish those mapping assumptions every time, because a number with hidden assumptions serves nobody.

PPDA showed me Germany. At the 2026 World Cup in Russia, working as a remote event-data analyst, I logged Germany's 26 shots against Mexico for just 1.3 xG, while Mexico's 12 shots produced 1.1 xG. Germany's PPDA was 6.9, and the match generated 18 transition chances. I shipped a thread before the final whistle: Germany would not escape Group F. Germany finished bottom. — Root: Used PPDA to predict Germany.

I applied the same logic to the BPL. In the powerplay, sides that keep BPPDA low — setting aggressive fields to squeeze the bowlers — concede nine to fourteen fewer runs on average. In the middle overs, the same aggressive setting backfires: singles and twos flow through the gaps, and pressure events dry up.

That yields a conclusion clubs resisted at first: pressing is not a moral virtue; it is a time-dependent decision. A side that fields with the same intensity all innings wins the powerplay and loses control in the middle.

I have spent the most time on death overs, because that is where the league's money and the league's mistakes are thickest. Death bowling has a routine of yorkers and slower balls, much like a set-piece routine in football. Teams rehearse it, but rarely change it to match the strength of the opposing batting order.

In my data, BPL sides that altered their yorker-to-slower ratio based on a batter's footwork profile saved six to nine runs per innings in the last four overs. Six to nine runs is often the difference between playoffs and elimination.

In 2026, during the pandemic shutdown, I consulted for Brentford FC. My job was to measure how much home advantage shifts behind closed doors. I analysed 306 behind-closed-doors matches across the Bundesliga, Championship and Serie A. Home win rate fell from 43.1% to 33.8%; the home xG differential dropped 0.21; distance covered in the final 15 minutes fell 5.2%. I called the adjustment CrowdNull.

Empty stadiums taught me that home advantage is a variable, not a law. The same holds in cricket. The home side's knowledge of a Mirpur surface compounds with crowd pressure and subtle behavioural leanings. In closed-door domestic matches, home teams' death-over economy rose by roughly 0.32 in my paper — without a crowd, a larger share of the pressure lands on the bowler.

The biggest error in imported analytics in Bangladesh is ignoring ground reality. Foreign models say: attack hard in the powerplay. But when a Mirpur pitch turns low in the second innings, attacking in the first six overs means losing two wickets and leaving the middle overs dark. My argument is that a league must be judged by its own data.

An ESTJ builds the pipeline first and the poetry second. The pipeline here is simple: what the scorer captures, what the coach sees, and what I code are not three truths — they are three cameras on one truth. As long as the BPL runs two to four cameras per match, our models must rest on manual coding and pre-registered hypotheses.

The auction and the age-group pipeline are two big variables. Franchises pay most for finishers, because death-over skill is visible. But if we judge Under-19 or first-class data only on runs and strike rate, we become victims of strike rate — because a 150 strike rate on a small ground comes from boundary dimensions, not from timing.

I keep umpiring and review data separate, because it silently changes results. In domestic league coding I found a mild venue-linked lean in LBW decisions, clearer in second innings. This is not a story about corruption; it is a story about human decision-making, and the job of data is to make that human lean visible.

Data scarcity is worst in the age-group pipeline. For a sixteen-year-old batter we have twenty innings of scorecards and no shot maps. Judging him on strike rate there is shooting arrows in the dark. I want clubs to keep shot maps for at least six matches — not just outcomes, but how late the bat arrives.

Now the line I say even about my own model. Metrics like xG are already being abused, because a metric cannot explain in-game decisions, player form, or umpiring standards. We produce a number and start calling it fate. What does Sheikh Jamal's 29 against 31.2 really say? It says they scored fewer runs than expected. It does not say they were unlucky, that the field setting was wrong, or that a finisher was carrying an injury. Correlation and cause are not the same thing.

When I see a side win five straight matches while trailing on xR, I do not write "they are lucky." I write that we need to check whether their death-over press-breaking pattern is durable. The decision belongs to the selector, the coach or the analyst; the model's job is only to sharpen the question.

The comeback story sits right here. A fast bowler returning from a torn ACL often has a poor death-over economy, and we dismiss it as lost form. The reality is that the mental block is harder to fix than the body — the legs return to their old positions, the courage to commit does not. Data can measure that missing courage, if we code which deliveries he went full length on, not just pace and economy.

So my rule: pre-register hypotheses before the season, publish base rates first, and state the model's limits openly. I sit with coaches and players to agree on the mapping — if they do not accept what we are calling a "press," the number stays on paper.

Next season my eye will be on three things: middle-over BPPDA, the xR differential of batters graduating from Under-19, and death-over economy once crowds return — because CrowdNull was a measure of crisis, and we still lack the measure of a return to normal. If selectors want one number, let it be the xR differential, not strike rate. One question still hangs: if a league cannot recognise itself in its own mirror, whose trophy is it anyway?