The Expected Runs Ledger: Searching for Truth Beyond the Scoreboard in Bangladesh's Domestic T20

On the night the domestic T20 points table was finalised, I reopened the spreadsheet on my balcony in Rangpur. Nine teams, 92 matches, roughly 19,000 balls of ball-by-ball ledger. The team that finished top of the table ranked eighth out of nine on expected runs (xR). The team that finished eighth owned the second-best xR differential in the league.

My first assumption was a formula error. I recalculated three times, reweighted the phase baselines, the wicket states, the pitch classifications, the dew factor, and still the gap held. The leader had won largely on two things: opposition errors in the death overs, and toss advantage. Add up the expected runs on every ball they faced, and they belong near the bottom of the league.

That night settled a principle that runs through everything I write: the scoreboard is an accounting artefact, not the truth. This article is an open ledger — how I built an expected runs model for domestic cricket, where it deserves trust, and where it is itself only a guess.

Context: where the ledger comes from

Ball-tracking data barely exists in Bangladesh's domestic T20. Hawk-Eye, ball-tracking feeds, field-placement maps: an annual subscription costs more than an entire domestic team's analytics department. I had three things: official ball-by-ball scorecard logs, handwritten notes taken from broadcasts, and an old laptop. The model had to be cheap, and that constraint should be stated rather than hidden.

The construction is simple. Every ball is sorted into a phase — powerplay (overs 1-6), middle (7-15), death (16-20). A baseline expected run value is then set by wicket state. Pitch type is layered on: Dhaka's slow, low surface; Chattogram's batting-friendly deck; Sylhet's turning track; Rangpur's two-paced wicket. Dew, day-night conditions and the toss add small corrections. Every baseline carries an uncertainty band, usually around 0.07 runs per ball.

The method dates to 2026. Aged twenty, after my own athletic career ended, I fed every Russia World Cup match into a manual xG spreadsheet I had built a year earlier. Croatia averaged 1.42 xG across seven matches but conceded 1.29 goals per game. France averaged 2.10 xG and conceded only 0.86. Before the final I wrote that Croatia's open-play xG was just 1.10 against France's 2.40, and predicted a France win. On 15 July 2026 at Luzhniki Stadium in Moscow, France beat Croatia 4-2. The blog drew 12,000 readers. I audited every shot of the 2026 World Cup and found where the model stumbles.

In 2026, with sport paused, I turned to the Bundesliga's restart. Across 306 pre-COVID matches and 92 post-restart matches, the home win rate fell from 43.3 per cent to 33.3 per cent, and home xG dropped from 1.54 to 1.31 per game. The report said plainly that 92 matches are not enough to rewrite home advantage theory. That habit stayed: every data piece I publish carries a paragraph on what the numbers cannot prove.

In 2026 I listened to press conferences and counted the pauses, not only the quotes. I waited until Italy had played all seven Euro matches before commenting. Their PPDA was 8.3, their xG 2.10 per game, and they conceded just 0.57 xG per game in the knockout stage. At the Tokyo Olympics I logged Pedri across six matches: 532 passes, 92 per cent accuracy, 11.8 kilometres per match. My personal rule since then is to wait seven matches before endorsing any new tactical meta. That rule slows me down and reduces my error rate far more.

Core analysis: three lies the scoreboard tells

The first thing the ledger exposed was a hidden staircase in the batting order. I took the three most celebrated innings of the domestic season. In all three, the gap between raw runs and xR ran between 18 and 27. The cause was the sixth bowler.

One batter made 68 off 45 balls, headlined as a brilliant finish. Split ball by ball, 41 of those runs came from 11 deliveries against part-timers and the sixth bowler. The other 27 came from 34 balls against frontline pace and spin — roughly a run a ball. The model puts his true contribution at 44 to 46 off 45. The widest divergence between raw runs and xR appears in innings where the frontline attack has already been used up. Selection, wages and next season's auction price all hang on reading that gap correctly.

The second lie concerns death overs. The league's run rate from overs 16 to 20 was 9.8 per over; the wicket-adjusted baseline was 8.4. That gap is variance as much as skill. A batter faces only 18 to 25 death deliveries across a season. A strike rate built on 22 balls is a sample, not a measurement. One 30-run over swings net run rate and swings the conversation. Using a run rate born from a tiny sample to make a large decision is turning luck into a scouting report.

The third lie is the comfortable chase. Six-wicket wins read as ease. Of the league's seven six-wicket chases, four had their wicket-adjusted chase xR below par until the seventeenth over. The win arrived in the final three overs against a weak spell that erased the arithmetic. The scoreboard calls it simple; the ledger calls it fragile.

I opened the transfer ledger and found a fee was never merely a number. Of three domestic contracts, the largest fee went to a 28-year-old top-order batter with a 128 strike rate: 362 runs in 11 matches and a 41 per cent dot-ball rate. The cheapest signing was a 31-year-old middle-order batter on a 119 strike rate: 298 runs, a 29 per cent dot-ball rate, and 11.4 runs per over in the last five across seven innings. Cost per run favoured the cheaper man heavily. Wage analysis starts with the quota, then the balance of the batting order, and only then the strike rate — reverse that order and the budget burns.

Workload is where domestic cricket is most blind. On deadline day I learned that paperwork is the only language the market respects, and no sheet lists a tired bowler. I built a risk score from zero to one hundred using 14-day rolling overs, spells per match, back-to-back spell frequency, the April-May heat index and travel hours. One case: a 22-year-old left-arm seamer bowled 214 overs across formats in eleven weeks. His risk score moved from 41 at the start of the season to 78 by week ten. Workload risk is not a feeling; it is a sum — rolling overs, spell counts, heat and travel hours.

A further structural finding: spinners concede 7.1 an over in the powerplay but 9.6 in the seventeenth. Yet the seventeenth is usually handed to a seventh-choice seamer running at 10.3. The most expensive overs go to the least proven hand.

Home advantage: crowd, pitch, travel, schedule

The home side wins 57 per cent of domestic league matches. The 2026 Bundesliga study suggests crowds matter — a drop from 43.3 to 33.3 per cent is not trivial even at 92 matches. But that finding cannot be transplanted whole, because in Bangladesh the home advantage comes mostly from the ground, not the stands. Ranked by materiality: pitch preparation first, since the home side chooses the surface; toss and dew second, since chase becomes easier under lights; travel third, with away sides often on six to ten hour road journeys; crowd last, because domestic attendances are in the low thousands.

Contrarian angle: correlation and causation

The xR model did not solve my problems; it clarified them. It cannot see field placement. Without ball tracking, I cannot separate a batter who pierced the field from one who slogged and was dropped. The model rewards outcomes. A lucky batter keeps mistiming into gaps and his xR rises — a modelling blind spot, not a skill.

The second gap is opposition quality. Phase baselines ignore who is bowling at the other end. A 7.2 economy in the death overs against a weak attack is not the same as 7.2 against a strong one. My bowler-quality index uses economy residuals, which is somewhat circular: I measure a bowler by his own results. Admitting the circularity avoids arrogance.

The third gap is match state. Required rate changes decision-making, and the model ignores it. The 2026 lesson applies directly: Croatia's open-play xG of 1.10 against France's 2.40 still produced a 4-2 France win because the model cannot see game management. The cricket equivalent is over-by-over chase management.

I also documented my own failure modes. Audit paralysis: verify everything and publish nothing. Sample-size purism: comparing one format against another from a different era erases pitch, workload and budget. Context inflation: adding variables until the signal disappears. My rule is to validate core variables first — phase, wickets, pitch — then add marginal ones, and to refuse large decisions built on cheap proxies.

Takeaway: signals for the next round

Three things over the next three weeks. First, death-over bowling load: for the two teams with compressed fixtures, any seamer under 22 passing 26 rolling overs in a 14-day window goes into the ledger regardless of results. Second, the table-versus-xR divergence: no verdict on any team before seven matches, and any gap above six points after that is itself the story. Third, the wage-to-xR ratio in the transfer ledger: when a club moves a batter mid-season, I check dot-ball rate and middle-over xR before fee and reputation.

Popular narratives demand fast answers. The ledger moves slowly. But the table forgets, and the ledger does not. So the question stands: if the scoreboard misleads us, who should the next auction trust — one innings from last month, or a nine-month data trail?