15,842 Balls, 132 Matches and Forty-One Empty Cells: The Spreadsheet That Re-Taught Me the BPL

Rangpur, 12 December 2026, a quarter past midnight. Forty-one cells on my laptop screen were still empty.

That evening the final had finished at the Sher-e-Bangla National Cricket Stadium — Rangpur Riders beat Dhaka Dynamites by 57 runs to lift their first BPL title. I did not watch the match; I read the scorecard. And the scorecard taught me the first lesson I still carry into every data file: a scorecard tells you who won, never why. It records how many runs the chasing side fell short by, not which over the match actually slipped away. Nothing on the sheet holds that.

After that night I stopped writing match reports and started writing methodology notes — every claim with its sample size, its weighting choice and a stated error margin. My sentences got shorter, my footnotes longer.

Context: starting from the empty cells

Five BPL seasons, 2026 to 2026. On paper 132 matches; in reality 129, with three washed out. My table filled up with 15,842 valid deliveries: over number, bowler type, batter hand, field setting, result. No public expected-runs model existed for that league then, and ball-tracking was far out of reach. So I opened a blank spreadsheet and let the Bangladesh Premier League teach me.

I deliberately kept the method crude, because crude models confess their errors. Three weights: powerplay (overs 1-6) at 0.82 of the base run rate, middle overs (7-15) at 0.93, death overs (16-20) at 1.80. Separate venue factors for Sylhet, Chattogram and Khulna. Every figure was tagged measured, modelled or guessed. The last column was my own confidence level.

The trouble arrived with the empty cells. Forty-one deliveries had no over number in any log; commentary could only place them vaguely 'late'. My first instinct was to drop them. Then I understood: an empty cell is evidence about who did not collect the data. Some Dhaka and Bogura matches, small-venue league games — no cameras, no ball-by-ball record. The weakness was not in my sample size but in who was watching. Scouts skip exactly those venues, and exactly those venues produce the cheapest talent.

Core: three phases, three kinds of lie

The powerplay produced a real average of 43.6; my model said 40.1. The extra runs came from a quiet change in opening strategy — after 2026 sides stopped attacking the first two overs and began protecting wickets until the sixth. The model could not see a temporary shift because its baseline was an older average.

Death overs were my embarrassment. Reality 58.4, model 52.7. My model was under-pricing the last five overs by 5.7 runs. Digging for the reason, I found the league's death bowling had hardened into one mould: fewer yorkers, more cutters and slower balls. Spin usage in the death overs rose from 11 per cent in 2026 to 23 per cent in 2026. The smaller the ground, the more reluctant the fast bowlers. I had added a venue factor but never added who was bowling at that venue.

Middle overs almost matched — 47.1 real, 46.3 modelled. That is my only comfort: the league never changed its rhythm in the middle.

One more thing surfaced that no scorecard shows. In the 2026 season Rangpur Riders' powerplay scoring rose by roughly seven runs over the previous edition. Someone would call that team improvement. The ball-by-ball file says the change arrived the year Chris Gayle and Evin Lewis joined the same squad. A name breaks a model, because models do not play; models average.

I later watched a few matches in person — three straight games in Sylhet. The sheet said the spinners were controlling that venue. My eyes said the opposite: the spinners were bowling because the seamers had already bowled short out of fear. The fielding-position data agreed with my eyes, but it took me six months of numbers to trust them.

Contrarian: what that 58 per cent actually measures

Every outlet carries the same figure — home sides win 58 per cent of matches. Across four or five BPL seasons my venue-based win rate landed close: 56.9 per cent. Broken down, the story changes.

First, 68 per cent of those matches were won by the side batting second; toss and dew matter at least as much as venue. Second, scheduling itself creates the bias — home teams play more often at Dhaka, and Dhaka's average score is nine runs higher than other venues. Third, matches beginning after dark and ending late show an even stronger chasing advantage.

So what we call home advantage is largely a schedule, toss and floodlight equation, not a measurement of crowd noise. Where the crowds are biggest the home win rate is genuinely a little higher, but the pattern is not linear. And at venues without cameras, we never measured anyone at all.

I nearly fell into another trap: more powerplay wickets means a higher chance of winning. The data says the relationship is moderate — a correlation of roughly 0.31 — and weaker still in the 2026 sample. The reason is simple. Wickets fall to both sides, and an early collapse sometimes shrinks the target and makes the chase easier. Correlation and causation are not the same animal; in cricket they are often enemies.

Takeaway: the next signal

I still open that spreadsheet weekly, and weekly I find fresh empty cells. Those gaps are the real gold — they tell me where our eyes are absent. When the stadiums emptied, I started measuring what the crowd used to hide; in the same way, the dark corners of our small venues now hide the most information.

The question is no longer who will win. The question is whether next BPL season those empty cells multiply, or whether someone starts filling them. Because as long as our attention stays at the centre of the match, the truth stays at the edges — right beside the boundary line at deep cover, where nobody thought to leave a camera.