Death-Overs Arithmetic: How Invisible Workload in the BPL Regular Season Breaks the Powerplay Baseline

Death-Overs Arithmetic: How Invisible Workload in the BPL Regular Season Breaks the Powerplay Baseline Sher-e-Bangla National Cricket Stadium, Mirpur. Round eight of the regular season. The 17th over of the innings. The bowler walking in has bowled more death overs than anyone in the tournament — an average of 8.4 overs across his last six games. Tracking data put his first ball of that over at 136.2 kph, against a season average of 141.7 for his first spell. By the third ball it was down to 132.4. The fifth slipped outside leg stump. Seventeen runs came off the over, 47 off the four. The scorecard will file this under death-overs bowling failure. My audit trail says the match was lost in the powerplay, where the same side had averaged 51.3 runs across its first six games and then produced 31, 29 and 34 in three straight matches. This is not a match report. It is a workload audit, and a warning: you cannot trust an outlier before you have built the baseline. In 2026, at 59, a Dhaka sports-data startup contracted me to build a standardised expected-runs model for the Bangladesh Premier League. Coding alone took four months: 1,240 shot events from 72 matches by hand — shot zone, foot position, bowler length bucket, distance from the boundary, swing plane — cross-referenced with distance-covered and pressure-per-ball data from local tracking providers. The model flagged Abahani Limited Dhaka's set-piece inefficiency at 0.18 expected goals per shot, which the coaching staff had dismissed as bad luck. A fourteen-page methodology brief showed it was a coding failure with a footwork solution. That brief became the startup's internal gold standard. The rule since then has not changed: readers should not meet a conclusion before they meet the sample size and the provenance. Sample here: the last eight rounds of the current regular season, 28 innings. Sources: ball-by-ball tracking feed, announced XIs, and my own spell log taken in the ground. Coding rules: "death overs" means 16 to 20; "workload" means total legal balls bowled in the last ten days, all formats combined; travel load is a separate column, because returning from Sylhet to Dhaka and bowling the next evening is not the same as bowling after a rest day. A metric announced without a baseline is just a rumour with decimals. Four thresholds, declared in advance. One: a powerplay scoring rate below 7.8 over six overs is a warning signal. Two: a pace drop above five kph across a four-over spell is a workload signal. Three: 90-plus legal balls in ten days is high risk. Four: a dew point above 3 lifts second-innings spin economy by roughly 1.4 times. The chain has five links. Link one: the powerplay collapse. Through round six that side's powerplay run rate was 8.54, scoring rate 1.38 per ball. Rounds seven and eight: run rate 5.17, scoring rate 0.86, against a baseline standard deviation of 0.71 — a fall of nearly six sigma inside 72 hours. Form does not move that fast. Link two: the workload log. The two frontline seamers bowled 142 and 118 legal balls in ten days; one of them worked three matches in seven days, two evening fixtures and a morning travel day in between. Past the 16th over, their spell efficiency sat at 72 to 76 per cent of baseline. Link three: the length map. Three of six balls in that 17th over were short or back-of-length; full-length percentage fell from 54 in the first six games to 31 in the last three. Same arm, same run-up, different body. Link four: why the powerplay breaks first. The new-ball spell is the leading indicator; the death-over economy is only where the bill arrives. Link five: bench depth. The third seamer has bowled eight overs all tournament, so nobody can relieve the front two. Bangladesh's pace-workload argument is not new. The careers of our most discussed fast bowlers have sat at the centre of it for two decades, Taskin Ahmed's injury cycles and the long conversation around Mustafizur Rahman included. But that argument is usually conducted in the language of selection politics rather than the language of workload, and workload indices are patient in a way selection debates are not. Spin is a separate problem because it is repetition, not new-ball burst. Among four spinners coded this season, those above 30 overs in eight days conceded 1.9 runs per over more than their equally used peers, and their failure showed up first in boundary rate — a lag that produces the classic selection error of dropping a spinner for four bad boundaries when the real cause was five days of workload. Then dew: Mirpur's dew point of 3.6 this round pushed second-innings spin economy to 9.2 against 7.1 in the first innings. Same pitch, same bowlers, different clock. When the stadiums emptied in 2026, my sixteen-year home-advantage model became worthless overnight. I rebuilt it in eleven days in my Barishal study around travel distance, rest days and referee nationality instead of crowd decibels. The new framework called 68 per cent of Bundesliga outcomes across the first three rounds after resumption, against 41 per cent for the old one. I have written a model-status line at the top of every piece since, because declared uncertainty does more work than concealed uncertainty. The 2026 group stage taught me that chaos has a schedule: Germany's pressing metrics jumped from 7.2 to 13.8 between qualifiers and the opener, and a pre-match note to three betting syndicates called the Mexico result before kickoff. Still, I stop short of causation. Three of those matches were batted first, forfeiting the dew advantage. One pitch behaved abnormally after heavy watering. One slip cordon changed because a fielder broke a finger. And 28 innings is a small sample. I do not chase upsets; I chart the conditions that invite them. The market side matters too. Franchises value innings — runs scored, wickets taken — while the column that decides matches, who bowls the 16th over without raising the opponent's win probability, stays invisible. That is precisely where young potential gets overpriced and dressing-room chemistry never gets priced at all. My model's limits are worth stating: I do not see injury reports, medical data or training conversations. A workload index is a proxy — it shows decline, not what is happening inside a body. Part of my own model is now retired. The powerplay component of the 2026 xG framework is switched off this season, because shot quality was zone-dependent and batters now hit fours from anywhere; a two-layer replacement is in pilot. Model status first, decisions later. For the next two rounds, watch three things. If those two seamers stay under 110 balls in ten days, does first-spell pace return? If the dew point stays under 3, can the second-innings spinners bowl freely for the first time this season? And if the powerplay scoring rate does not climb above 1.15, the problem is not the squad — it is the eleven. The market moves fast; the baseline moves first.