Not a Death-Over Collapse but a Workload Ledger: The PPDA Audit of a Regular Season

Sher-e-Bangla National Cricket Stadium, Mirpur. Match fourteen of the regular season. Over seventeen, Taskin Ahmed's second spell. My speed sheet reads 141.6 kph average in the first spell, 133.2 in the second. Between overs sixteen and twenty, runs per over climbed from 6.1 to 11.4. The scoreboard told a clean story: a bowler cracking under death-over pressure. My workload log tells a different one. In the previous twenty-one days that seamer bowled 92.4 overs, 37.8 percent of them in the death phase. Across the same window he played four matches at three venues — Mirpur, Chattogram, Sylhet. In his last three matches his second-spell average was 139.1 kph; here it was 133.2. Behind a visible collapse there is an invisible cause, and that cause is usually written in the calendar. I built the baseline before I trusted the outlier.

Every number here comes from my own coding. The sample is eighteen matches from the current regular season, 2,146 delivery events in total. Data comes from two sources — venue speed guns and my own frame-by-frame manual coding — and both are cross-checked against ball-by-ball scorecard logs. The coding rules go down first. Anything from over sixteen counts as death. A spell is at least three uninterrupted overs of bowling; a break under twenty-five minutes does not create a new spell. A speed dip counts only when the bowler took a full run-up; injury-warning shortened run-ups and slower balls are excluded.

PPDA creates confusion, so I will be explicit. In football, PPDA means passes allowed per defensive action. I have adapted it for cricket: the ratio of pressure-building dot balls and forced failed shots per fielding phase. There are three phases — powerplay, middle, death. In 2026, building an xG model for the BPL for a Dhaka sports-data startup, I hand-coded 1,240 shot events from 72 matches. What those four months taught me became my writing rule: sample and coding rules first, interpretation second. A metric without a baseline is just a rumor with decimals. In the 2026 regular season, teams are travelling more than ever — the calendar is now as important to me as tactics.

The speed decline is simple but orderly. In a twenty-one-day window, bowlers who delivered more than 85 overs saw second-spell averages drop between 7.4 and 11.2 kph. Those under 65 overs dropped less than 2.1. The correlation between that speed drop and runs conceded in the death phase is 0.71. The number is eye-catching, and this is exactly where I stay most careful — correlation is not causation.

Of the four seamers who passed 85 overs in that window, two did not play the next two rounds. Club statements cited muscle spasms. My log cites something different: flights across three venues, hotel changes, a 92-over load. An injury report dates the event; a workload log dates the cause.

The pressure curve paints a clearer picture. Across these eighteen matches, team PPDA reads 4.2 in the powerplay, 7.8 in the middle, 9.6 at the death. As the ball ages, fielding pressure roughly halves. The exception is where the work is. The three sides holding death-over PPDA under 7 have used spin for 41 percent of balls in the last five overs on average — bowlers like Mehidy Hasan Miraz are being brought into the death, moving the burden off the seamer's shoulder.

The match-up layer is missing if you skip it. In this sample, a right-arm seamer against a left-handed batter in the death phase concedes at 9.8 an over; a left-arm cutter bowler like Mustafizur Rahman concedes 7.1. In seven of ten innings, no left-arm seamer bowled in the last five overs. Captaincy is a bigger variable here than workload, and I cannot measure it — only count it.

Venue-specific pitch data matters too. This season spinners concede at 6.4 an over in Sylhet, 7.9 in Mirpur, with Chattogram in between. The same workload produces different results at different venues. In 2026, Abahani Limited Dhaka's defensive data showed they leaked 0.18 xG per shot from set fielding positions; the coaching staff dismissed it as bad luck. The pattern repeats in the regular season — bigger sides are slow to change their boundary fielding sets outside the powerplay, and they pay for it near the rope.

Keep the league-wide baseline in mind: this season, overs sixteen to twenty average 8.9 runs per over. That is the league mean, not my sample mean. Death-over scoring has eased slightly, yet specific spells still collapse as sharply as before. The cause is not format change but usage — who bowls how much is deciding who bowls the last over.

When the stadiums went empty in 2026, my fifteen-year crowd-noise coefficient model became worthless overnight. I sat in my Barishal study for eleven days and rebuilt it — crowd density replaced by travel distance, rest days and umpire home distance. The new framework correctly predicted 68 percent of Bundesliga outcomes in the first three rounds after resumption; the old model managed 41 percent. When the stadiums went empty, I recalibrated what home meant.

The market moves fast, but the baseline moves first. Last round, three bookmakers took an average of eleven hours to update their lines to my workload-based thresholds. By the time a death-over speed drop is visible on broadcast, the price has already moved.

The 2026 group stage taught me that chaos has a schedule. Germany's pressing fell from 7.2 PPDA in qualifying to 13.8 in the group opener; a 12.4 kilometre drop in average distance covered in the final twenty minutes of warm-up matches matched it. Mexico won 1-0, and that pre-match note was forwarded over 400 times on WhatsApp. I do not chase upsets. I chart the conditions that invite them.

Now I argue against myself, because the biggest trap for a data analyst is locking the door of the coding room. A 0.71 correlation is not causation. That seamer's second spell did lose pace, but he also bowled two extra overs because the captain trusts him — that is role, not workload. Rhythm, captaincy confidence, square-leg umpiring calls, dressing-room chemistry: none of it enters my coding sheet, and all of it changes outcomes. My team-level mean works at match level, not at delivery level.

One more limit, stated plainly. The new framework can be right, but its hit rate sits in the 60 to 70 percent band, not the 90s. Anyone selling certainty in my name is not using my method. I am willing to retire an old model, but not before piloting the replacement — the cricket adaptation of PPDA is still in pilot, and I will not run it as settled truth.

On valuation models my objection is old and unchanged. In an auction, a 22-year-old quick like Nahid Rana is priced on his best spell and his ceiling; whether he can take the pressure ball in a dressing room appears on nobody's table. Placing age and potential in the same column is what I call a half-sample wearing a full-sample label.

Two thresholds for the next round, named in advance. One: if a seamer past 85 overs in twenty-one days loses more than 6 kph in his second spell, I will support removing him from the death overs. Two: if a side's death-over PPDA sits above 8, that is my signal of a tactics problem, not a personal-failure story. If both thresholds cross, which one comes first? The calendar. It is written before the collapse arrives.