Start With the Match ID: Where Bangladesh Cricket's Data Pipeline Fails Its Own Audit

Start with the pipeline, not the prediction.

Nothing dramatic happened. A T20 match, local start 10:40pm. The game ended, the scorecard was fine, the highlights were fine. But two data feeds filed the same match under two different match IDs. One carried the next day's date because the feed splits on UTC; in the other, home and away had flipped in a single cell. From that one cell, the home side's powerplay run rate reads 7.8 in one table and 8.4 in the other.

I watch cricket from Khulna, but my real work starts after the game. Across recent seasons, reconciling domestic and international scorecards, I keep getting stuck in the same place — not batting, bowling or fielding, but feeds, IDs and definitions. The biggest error in the accounting is never on the field; it sits in the blank cell between the scorecard and the feed. That blank cell is today's subject.

Context: rules first, interpretation later

When I built a standard data template for the BPL in 2026, the goal was simple: any editor should be able to verify my table themselves. The template carries four layers — source, match ID, cleaning rule, sample window. In football I worked with xG and PPDA. Cricket has no direct substitute, so I defined two proxies and published them openly.

First, Dot-Pressure Conversion (DCR): the share of dot balls in the last five overs that turn into a wicket within the next three deliveries. Second, Strike-Rotation Failure (SRF): the share of deliveries between overs seven and fifteen from which no run comes at all. Both have limits — DCR does not separate bowling changes, dew or pitch character; SRF ignores innings state. That is why every piece I write carries a sample size. Publishing an SRF on a small sample means pressing confidence onto noise.

Data production in Bangladesh cricket has grown — board feeds, broadcaster graphics, market score APIs. Production and verification are not the same thing. More sources mean more cross-checks and less control. The result: analyst, editor and bettor discuss the same match with three different numbers, and none of them knows which one is real.

Core: exactly where the accounting breaks

First crack: time zone. Many Bangladesh matches start at 9:30 or 10:30pm. On a UTC-based feed that lands on the following date. If match IDs are date-derived, the same match is filed twice. After catching such duplicates in my own log, I made a rule: verify by venue code plus both teams' three-letter codes plus innings start time — not by date. Once that rule is in place, duplicates drop to nearly zero.

Second crack: definitions. What is a dot ball? A delivery where the batter takes no run but leg byes come — dot or runs? A delivery where a catch is dropped but the batter never left the crease — dot or chance? One feed answers one way, another the other. Using both together shifts powerplay run rate by 0.3 to 0.6. That gap is the most dangerous kind, because the number does not look wrong — it only looks different.

Third crack: the toss. Long the most-read variable and the least controlled. At Mirpur on a winter evening, when dew settles, the ball does not grip in the second innings — physics, not fashion. But "winning the toss equals an advantage" is poison for a model. My frame does not score the toss separately; it scores the toss decision plus the innings-time dew load. On the same pitch, an afternoon match and a night match do not carry equal home advantage, whatever the toss says.

Fourth crack: rain and DLS. A rain-shortened game means a new target, new ball-boundary accounting, a new innings shape. There is no tactics here, only bookkeeping. Pressing audits are just bookkeeping for chaos. Drop a DLS-revised match into a normal list and average it, and the list becomes meaningless. In my table, DLS matches sit in their own layer and never enter a rate index.

Fifth crack: bowling workload. A team's problem is never the raw count of dot balls; it is who is bowling them. I have seen a spinner's SRF hold steady through the middle overs while a seamer's DCR nearly doubles in the powerplay. Aggregate the team and both vanish. Every outlier is a question the data is asking you.

One citable fact on home advantage. On 10 January 2026 at Chittagong, Bangladesh won their first Test, beating Zimbabwe by 226 runs. On 17 March 2026 at Port of Spain, they beat India by five wickets at a World Cup. The second was a neutral venue; the first was their own pitch. When someone says Bangladesh are strong at home, I want those two kinds of win separated — because whether the edge comes from the crowd or from pitch preparation is a different story entirely.

Contrarian angle: what I do not believe in my own index

In 2026, crowdless cricket handed us a control group nobody asked for. Bangladesh's series against West Indies, among many others, was played in empty grounds. In the subcontinent, a large share of what is called home advantage is attributed to crowd pressure; without crowds, much of it should have vanished. It did not all vanish. Because the real pressure was never in the noise — it was in how the pitch behaved, how the ball seamed, how a spinner turned it on a damp surface. That is where my confidence cracks. Some of the "environment" I use to explain the advantage at 22 yards is actually a decision made in the preparation room. And decisions have no feed.

Still, I will not revise my index unless evidence arrives. If two independent feeds over three seasons return the same DCR on the same sample, and that value moves with pitch age rather than with crowd presence, then I will remove "home advantage equals crowd" from my framework. I am willing to re-test. That is not the same as being willing to back down.

And the market. In betting, the edge hides in the boring columns. Everyone watches powerplay run rate; nobody checks which feed left three deliveries unassigned, or which venue had rain the day before. What I have seen over three years is plain: a model standing on an unclean match ID has a beautiful prediction and a zero audit trail. Some buyers only purchase the number, never the proof of the accounting.

Comparing India's and Bangladesh's data systems matters here, because a metric's meaning shifts when the environment shifts. India's domestic structure produces so many matches, broadcasts and physio rotations that a spell's decline shows up the next day. In Bangladesh, the same sample takes two seasons. One index can be used in both places; the limits of the decision cannot. The pools are not the same size.

Takeaway

Next season I carry three revision triggers: if a feed fails my match ID rule; if the dot-ball definition shifts again; and if dew load in any match runs double the norm. Any one of those and my index changes, not my reporting. I will leave the question open: the match number in your hand — which match is it, and who verified it?