The Fatigue Curve of the Last Three Overs: The Arithmetic Tournament Cricket's Models Cannot See

A night match from the last cycle still sits unfinished in my notebook. Before the 19th over began, my small model put the probability of a wicket in that over at 0.11 — the most hopeful window of the six balls for the bowling side. The over went for nineteen. The match went with it. Nobody in the dugout made an error, nobody shirked responsibility. The bowler's pace had dropped eight kilometres that evening, the last ball was the fourth ball of his fourth straight over, and before starting his run-up he had rested on his back leg two seconds longer than usual. I wrote beside the page: workload, more than skill, decided this over. In tournament cricket that gap is the least discussed and the most expensive.

I have been building an expected-wickets database for cricket from Rangpur since 2026. The idea is borrowed from football's xG model, but the logic is the same: look at the chances before you look at the result. That year a Rangpur club lost 2-1 while outshooting the opponent 17-6. I handed the coaching staff a one-page xG breakdown, then installed pressing metrics; across the next six matches our PPDA fell from 14.2 to 9.8. Cricket has no direct equivalent of pressing, but it has dot-ball pressure and phase-adjusted economy. The lesson held anyway: one number starts a story, it never finishes it.

A tournament cycle means compressed time. In a league you play thirty matches and get to correct your mistakes; in a tournament seven matches make that a luxury. At Russia 2026 I tracked Croatia on a single spreadsheet — three straight knockout matches went to extra time, their xG was modest, yet they reached the final and lost 4-2 to France on 15 July 2026 (source: FIFA official match record). I built a small model and told colleagues France held a 62 per cent edge. The model was right, but the real lesson was in the gaps: penalties, fatigue and set pieces sat outside my arithmetic. I came home and added three layers — territory, pressing triggers and rest days.

Qatar 2026's winter window sharpened that lesson. Several group matches carried more than ten minutes of stoppage time, and the England-Iran fixture produced a record amount across the two halves. I logged every minute and found that late goals rose sharply, punishing squads with thin rotations. Before the knockout rounds I briefed two clubs on substitution timing for the final fifteen minutes; teams that followed my fatigue curve conceded measurably fewer goals after the 75th minute. In cricket the decision does not translate literally, but the logic lands exactly: tournament arithmetic is schedule arithmetic.

My expected-wickets model has four inputs. Line and length from ball tracking, with a cricket-specific pitch map laid over it. The bowler's historical matchup against the batter — not just total wickets, but broken down by delivery type. Phase weights: a wicket is worth most in the powerplay, less in the middle, and rises again in the last five overs. And an adjustment for ball age and conditions, which must be computed separately on a spin-friendly pitch. Put those four together and the number that emerges is not a verdict; it is a grammar of probability.

Across the Asian matches I logged last cycle, one pattern became clear. Teams that controlled their phase-adjusted strike rate before the 16th over conceded roughly one run less per over across the last five. In the end this is arithmetic of hands, not talent. A side with five reliable death bowlers never needs to hand the ball to a tired bowler in the 16th over. A side with three has to walk a fourth over in the 19th — and that is exactly where my 0.11 broke down. For Bangladesh, how the overs are shared between Taskin Ahmed and Mustafizur Rahman is the most important line on my table.

Asian conditions make the arithmetic messier. Evening dew turns a spinner from a weapon into a burden, and in the second innings batters rotate the bat far more easily. Last cycle, second-innings economy in evening matches ran about 0.4 runs higher than first innings, and the spinner wicket rate dropped by roughly nine per cent. Anyone picking a side without that adjustment is picking on the toss, not on cricket.

Phase-adjusted strike rate is one of the most honest indicators I keep. A 140 strike rate in the powerplay is not equal to another 140 if the second batter is batting in the death overs. I now weight every innings by where the balls were bowled, then compare against league averages to see who is genuinely ahead. This correction dissolves a lot of romance. An innings that looks spectacular may be standing on seven edges and three top-edges; the spectator's eye follows the batter, my ledger follows the bowler's legs.

The dot-ball pressure index is my cricket version of PPDA. It is not only how many runs I conceded, but in which over I built pressure, how many balls I forced the batter to defend, and how expensive that pressure became in the next over. In one BPL season I watched a side bowl more than four dot balls an over across its last six matches while the scoreboard hid it, because they conceded one six every over. That oscillation between dot and six is the real story to me.

Workload now goes in the first line of my previews, ahead of form. My table carries three columns: overs bowled in the last ten days, travel distance, and rest days. From the Qatar lesson I set a rule — for a seamer who has bowled a full quota in two straight matches, the wicket probability in his fourth over falls by 14 to 20 per cent on average, and that over's economy rises by 1.5 to 2 runs. The number does not blame the bowler; it puts a decision in the captain's hand.

In 2026 the German league returned to ghost games, and I treated it as the cleanest natural experiment of my career. Across the first forty matches the home win rate fell from roughly 43 per cent to 33, and added injury time dropped by nearly a minute per game. I wrote then that crowd noise measurably shifts referee decisions. In cricket the experiment is even cleaner, because DRS reviews, umpiring hesitation and batter pressure all connect directly to the crowd.

There is an uncomfortable edge here. The empty stadium gave me my cleanest data, and also my loneliest answer. When conditions are clean the model runs well, but the match stops being the same match. In a ground with thousands of throats, my fatigue curve can be cut by a single shout. So every report now carries a paragraph on what the model cannot see — dew, wind, the umpire's mood and dugout pressure, written out separately.

My first-year habit has not changed. Before I sit down to any column, I place three verifiable numbers, then the sentences. Last cycle they were the phase-adjusted strike rate of the last five overs, the dot-ball pressure index, and the pace bowlers' workload. Placed together, those three break many beautiful stories. And if the numbers and the narrative walk on separate paths in any column, I do not publish it — that is my only strict rule.

Then I stop. The gap between correlation and causation is the biggest trap in my profession. In my database, did the sides that conceded few runs before the 16th over actually have better death bowling, or did they land a better captain, or an easier group, or did their matches finish before the evening dew came down? In a single season those three causes are nearly impossible to separate. When a model gets too sure of itself, I open the xG notebook, and beside every claim I write a confidence range and the name of one clear limitation.

Croatia taught me that one number can start a story but never end it. My 0.11 wicket probability was an accurate calculation, not a wrong decision — it simply did not know about twenty thousand shouting throats, tired legs and two extra seconds of standing still. A dashboard's job is to survive in a coach's hands; and if a dashboard survives in front of a coach, it has done my job.

In the next round I will watch three things. First, rotation: a side that has shared its pacers' overs before the knockouts will look different in the final three overs. Second, the 16-to-20 window — who is bowling, a man walking in for his third or fourth over, or a fresh bowler. And third, dew: a spinner not bowling a second spell on an evening is a matter of the toss and humidity. The trophy ledger, in the end, is written on the schedule page. That is my next note.