The 18th Over and an Auditable Notebook: Load, Sample Size, and Data That Does Not Guess

The third ball of the 18th over landed well outside leg stump, on the damp part of the deck. Sylhet needed 11 runs, a spinner was releasing the ball, and twenty-two thousand people in the stands exhaled at once. There was no dashboard in front of me; there was a notebook, in which I had hand-coded 1,024 deliveries from the first phase of this tournament — where the ball pitched, the batter's footwork, the fielder's starting position, the distance from the rope. Once the ball crossed the boundary, the easiest explanation assembled itself: the spinner could not absorb the pressure. The notebook tells the opposite story. I hand-coded 1,024 passes in Cardiff before I trusted a single dashboard. The Sylhet Data Room began with one notebook, one modem, and a stubborn refusal to guess. In tournament coverage that habit matters more than anywhere else, because a tournament calendar blends every number with emotion. The structure of the cycle is the primary variable here. World Cups and regional championships deliver six to eight matches in four weeks, repeated travel between two cities, start times shifted by rain and dew, and constant second-guessing of squad depth. Teams do not change how they play; what changes is how much each player can still give. I have kept a load ledger across 50-plus club matches, and it shows that after three matches inside a five-day window, the muscle-injury risk for fast bowlers rises roughly 2.3 times. That figure is not there to frighten anyone; it is a threshold for caution. The place where the notebook speaks loudest is the last four overs. Conventional statistics quote strike rate, but between overs 16 and 20 strike rate is an outcome, not a cause. My hand-coding says only two things separate teams under pressure: dot-ball percentage and the consistency of a spinner's release point. When Sylhet dew dampens the ball, a release point that drifts half a foot back does not lift the strike rate, but the dot-ball percentage climbs quietly. The story of the match then lives not on the scoreboard but in the camera frames. The notebook I keep is structured like a verifiable ledger. Every delivery is an entry; every entry carries time, over, bowler type and field setting. If someone later claims a certain spinner was poor in the last eight overs, I can reconcile the entries one by one — where runs came, where they did not. The core lesson of decentralisation and blockchain is identical: only a record that cannot be altered afterwards, and whose every step can be checked independently, is fit to base decisions on. Dashboards paint pictures; notebooks give testimony. In bowling load I separate three components. First, spell length and per-over delivery speed. Second, the minutes of rest between two spells. Third, the running done in the field, which no scorecard records. A bowler who delivers four overs and then chases in the deep for twelve has a load that cannot be read as four overs. That is why someone suddenly looks slow in a tournament's fifth match while the crowd concludes the form has gone. To measure fielding pressure I borrow football's PPDA: how many effective movements our fielders make before each opposition pass. Tournament sides split into two types — those who press in the first powerplay, and those who store pressure for after the 15th over. The second model is expensive in load terms, because a fast bowler must bowl at full tilt twice inside the last five overs. Squad selection becomes a list of names when rest windows are not reconciled against that cost. When the 64-match xG bracket called France, I learned that models can be quiet prophets. The cricket translation is this: when a number comes from ten matches of data, it needs a band around it. I never say a bowler will concede 7.2 in the 18th over; I say his expected range is 6 to 9.5, and the upper edge if dew is present. That difference sounds small in journalism, but it changes what a decision means. Now the point where I part company with colleagues. Good bowling in the last five overs and winning matches are correlated, not causally linked — those are two different statements. Teams that win naturally get their death overs watched more; the good death bowling in lost matches is recorded nowhere. That is survivorship bias. Declaring a bowler a 'closer' from seven tournament matches, and calling someone a 'finisher' from one match in a season, are the same hurry. Empty stadiums in 2026 taught me that atmosphere is a variable, not a verdict. Dew, wind speed, the moment daylight fades — these alter how the ball behaves. Under tournament pressure we often treat crowd noise as a cause of performance. In reality noise is a signal that speeds up a player's decision, and a faster decision is not always a better one. That ball in the 18th over is the evidence: the spinner did not err in his choice, he released a second early. At this point the biggest lesson from 43 years of watching is that sample size fixes the limits of any judgement. Two forties in three matches and we build a narrative; yet two successes in three attempts is a probability well within pure chance. In my count you need eight matches across at least two different conditions before calling anything a tendency. Before that it is only a probability, and printing a probability as a prophecy is a fraud on the reader. At 59, I still hand-code because trust is a manual process. Writing each entry after watching it myself leaves room to err but none to guess. That difference is the only asset my work has. In the next phase I will watch two things closely. One, the rest pattern of fast bowlers — where the gap drops below three days, I will log first-spell pace and line separately. Two, dot-ball percentage between overs 16 and 20 — the side that holds dot balls in that window has a better chance of winning than the one with the brighter strike-rate star. The question for the reader: which number will you trust — the scorecard's 42, or the notebook's 1,024 entries, where the ball's path is written alongside its time?