From a 44-Match Notebook to a BPL Data Audit: What Hand-Coded Numbers in Rangpur Actually Say

One winter in 2026 I opened a spiral notebook in the seventh row of the western gallery at Rangpur Stadium. I was sixteen. Across all 44 matches of that Bangladesh Premier League season I drew the same four columns — event, location, minute, outcome. Where the shot came from, which way the pass went, in which minute, and what it produced. No local outlet published anything beyond goals and cards, so I filled the gap myself. Eight years later, sitting in a Rangpur tea stall looking back at photographs of those sheets, it struck me that this grid was my first model, and that the model's errors are the real subject of this piece.

I began with 44 matches, a Rangpur notebook, and a suspicion of easy numbers. Out of that season's sheets one pattern surfaced that no Bangladeshi reporter had named: 61 percent of Abahani Limited Dhaka's open-play goals originated in the left half-space. In newspaper reports that was 'excellent finishing'; in my column it was 'structural advantage down the left.' Eleven people replied to the photographs; one was a university coach. I did not understand it then, but my writing rules changed in that moment — I stopped filing any match report without a numbers sheet attached.

The question now is simple: is a hand-coded notebook really a basis for conclusions, or is it just a good story we pass off as data?

Why the notebook's column structure still matters

The first paid byline taught me that a model is only as honest as its assumptions. Watching all 64 matches of the 2026 Russia World Cup on a 21-inch television, I logged roughly 1,200 shot coordinates into a Google Sheets xG model whose column logic came straight from the Rangpur notebook. Croatia's three consecutive extra-time matches — Denmark, Russia, England — were my test case. In the England semifinal I calculated 143.6 kilometres covered, the tournament's highest. A Dhaka football site published my 3,000-word breakdown and paid me 4,000 taka.

That 4,000 taka was not just money. It was proof that a public, reproducible model outargues opinion. From that day I began attaching methodology footnotes to every piece.

But this is where the first crack appears. Hand-coded data has a problem nobody wants to admit: observer bias. When I record the direction of a shot, my eye already knows which team is attacking, who is ahead, who is behind. Whether I would reach the same decision watching the same match a second time is a test nobody runs. In 2026 I did not run it.

What 44 matches showed, and what they did not

Now the real question. Hunting patterns in that notebook, I found two things. First, a relationship between wind direction and the goal being defended at specific venues: at Rangpur, afternoon wind generally blows west to east, and the side defending the western goal in the second half gains more from long balls. Second, a side that concedes inside the first 15 minutes attacks more in the final 15 — not because of quality, but because the defensive line pushes higher.

Here is where caution belongs. Forty-four matches means 44 observations, and I arrived late to 11 of them, meaning no data for the first 20 minutes. Whatever emerges from any statistic, does it survive base rates? Home teams won roughly 43 percent of matches that BPL season. My notebook suggested the side with the wind advantage won 59 percent of the time, but the sample is so small that the gap could be noise. I did not publish that number then. I still do not.

For comparison: in 2026, when 83 Bundesliga matches were played behind closed doors after the pandemic restart, I coded that dataset. The home win rate fell from 43.3 percent to 33.3 percent. That was 83 matches, one clear question, and one measured change. I turned it into a sociology term paper, 'The Twelfth Man Is a Variable.' Two journals rejected it; a blog post of the same argument was read by 9,000 people.

Place those two statistics side by side and the lesson is clean: 44 matches generate hypotheses; 83 matches test them.

Domestic cricket in the mirror of global leagues

One habit irritates me when writing about the BPL: data is either measured against IPL standards or skipped entirely. Both are wrong. IPL bowling data is unreadable without pitch behaviour, dew factor and franchise budgets; for the BPL you must add irregular scheduling, shifting venues and wildly unstable attendance.

Sitting in the stands I noticed something television never captures. When the gallery is half empty, a fielding captain's verbal instructions carry less, and a bowler's run-up rhythm shifts. There is no easy way to measure that, but it is true that these variables never reach the scorecard. Empty stadiums taught me that when an environmental variable looks small, its data does not become small — we simply do not collect it.

Take one case that tests this argument. In the 2026 BPL, Khulna Tigers' powerplay scoring rate was about 7.8 per over in the first half of the tournament and dropped to 6.4 in the second. Loss of form was the accepted explanation. But a match-by-match look at the schedule shows three of their five second-half matches were played in afternoon heat, and the venue changed twice. Letting a single explanation dominate means suppressing the other.

Hand-coded versus herd-coded

An uncomfortable point follows. Much of data journalism now rests on hosted databases and automated logging. That is good, because scale grows. But what automated logging does not measure is the reason behind a decision. A shot map shows the ball came from the left; it does not show that a defender was out of position, which is what made it possible.

Herd-coded data has another problem: everyone pulls the same numbers from the same source, reaches the same conclusion, and then mistakes that for a majority of evidence. Hand-coded beats herd-coded — but hand-coding means being alone with your own errors. In 2026 nobody caught mine, because there was no second notebook to compare against.

So in 2026 I follow a new rule. Whenever a number enters a piece, three things travel with it: sample size, collection method, and the weakness of that method. How dramatic the number is matters less to me now. In an earlier draft of this piece I placed 61 percent and 59 percent in the same paragraph. On rereading I realised their samples differ, so putting them side by side is itself an assumption — one I have banned for myself.

The real signal for readers in the regular season

Back from football to cricket. What to watch in the coming series is not only who is scoring runs, but which side bowls whom, and in which over. Sending a spinner into the powerplay means the captain is betting on a specific matchup; whether that bet is right cannot be known in one match, only in ten. If a reader logs that decision every match, by season's end they hold a small dataset of their own — exactly what I had.

But the biggest point concerns our behaviour around numbers. We all want a statistic to hand us a prediction. It does not. A number is a question, not an answer. The 44-match notebook taught me that, and learning it never required me to give up my affection for those 44 matches.

Next season, if someone asks what the Rangpur notebook's greatest contribution was, I will say this: it proves that empty stands, shifting schedules and a hand-drawn column, if collected honestly, outlast a viral hot take. One question stays with me: how many matches are we willing to hand-code, if we do not like the answer?