The Price of the Death Over: A Workload Curve, a Scoring Ledger and One Expired Assumption

Bangladesh's death-over economy from overs 17 to 20 has moved to 11.2 runs per over across the last three matches. In the February window, the same phase sat at 8.4. That gap is 2.8 runs an over — eleven and a half runs across four overs, which in T20 is usually the difference between winning and losing. Over those three nights in Mirpur I hand-logged 214 deliveries, frame by frame, long before any scorecard was reconciled. Two things surfaced. Boundary percentage barely moved (14.8 to 15.1). The share of soft, hittable deliveries rose 14 percent, and average pace after the 17th over fell 4.1 kph against the first spell. I logged every shot by hand before the market learned to price it.

If boundaries did not rise, the batters did not suddenly get aggressive. The ball is arriving slower, hitting softer areas. The signal is execution, not batting intent. Yet franchise auction valuations for these bowlers have barely shifted. My hand-logged ledger shows a curve bending downward; the market says nothing happened.

From my Khulna desk I treat Bangladesh's calendar as a portfolio, and every asset in a portfolio carries an expiry date. June to August: nine competitive matches in 21 days — BPL playoffs, a bilateral series, two travel blocks. In that window, three of the four frontline seamers bowled more than 40 overs inside a rolling 14 days. The question is not who is tired. The question is at what bowling load delivery quality starts to decay, and where the market has priced that point.

In 2026, on a 12-person desk in Dhaka, I held the only data seat. I hand-logged 1,140 shots from 96 BPL matches off grainy streams. Abahani Limited Dhaka won the title that year. The senior columnist called my table "a girl counting shots." Two BPL head coaches later asked for the spreadsheet. That table taught me to open with a number instead of an adjective, and to attach margin of error to every claim.

Then 6 July 2026, Kazan. Belgium 2-1 Brazil. Shots 9 against 21, created 1.1 against 2.4 xG. Every front page in Dhaka called it a robbery. I filed at 3 a.m., arguing 41 percent possession was a deliberate low block built on 18 recoveries inside their own third. It became the outlet's most-read piece of the year, 480,000 reads. Belgium — Root: 2026 defending Belgium. Since then the rule is fixed: I publish a counter-consensus read only when the model's edge clears a threshold set in advance, and I print that threshold inside the article. In cricket my threshold is 0.6 runs per over in death-over economy. The current divergence is 2.8, more than four times the threshold.

So I opened the table: 31 spells across nine matches, 214 hand-logged deliveries. Death-over economy by rolling 14-day overs: up to 30 overs, 8.1; 31 to 40, 9.6; 41 and above, 11.9. Inside a single spell the decay is cleaner: first over 7.4, second 9.1, third 10.3, fourth 12.6. The spreadsheet is my monastery; every formula is a vow of clarity. The decay is not linear — it is a staircase, with the second-to-third-over jump at 1.2 runs against 1.7 from first to second.

Why the staircase forms is a tactical question. With five fielders outside the circle in overs 17 to 20, run-denial options shrink; the bowler must commit to a yorker and slower-cutter mix. The more manufactured the mix, the larger the error. In my log, the 41-plus group shows 23 percent wrong-length deliveries against 11 percent in the under-30 group. Fielders do not reach short third-man early in the first spell; that delay is visible in the third spell. Physical fatigue and decision fatigue arrive together.

A structural question follows. Ball-by-ball integrity in cricket still rests on a central scoring system, and corrections happen quietly. A timestamped, immutable ledger — where every delivery, field placement and review decision is written once and cannot be altered — moves the argument from "whose table is right" to "what does the log say." I hand-log because I know where the gaps are. If every ball sat on an open ledger, my job would not be easier; it would be more credible, and market suspicion would fall.

On price: franchise bands imply an economy of 7.2 per over in overs 17 to 20. My hand-logged curve implies 8.4 to 9.1. A transfer rumor is an unhedged position until the medical clears. The gap is 1.2 to 1.9 runs per over — the market is still buying old reputation, not technical decay. I read that as a pricing error, not a moral verdict.

Here I argue against myself. The divergence is 2.8, but crediting all of it to workload would be sloppy. Two of those three matches were at smaller grounds with pulled-in boundaries, and the batting line-up was deliberately hunting match-ups, shielding its right-handed finisher from the left-arm spinner's overs. Across 31 spells I attribute 40 to 50 percent to workload; the rest to conditions, boundary dimensions and match-up planning. Correlation is never causation. A bowler with 41 overs behind him was not bad; he was bowled in bad conditions. A model that cannot say that is overfitted.

My pre-registered minimum sample was 40 spells. I am at 31. This piece is a pre-registered notice, not a verdict — threshold crossed, sample not yet. Living between those two positions is the work. I do not chase edges. I audit the assumptions that create them.

Next round I will watch new-ball economy alongside the death overs. If overs one to six also drift above 8.5 once conditions are stripped out, the problem is not fatigue but plan. If the new ball behaves but third-spell pace drop clears 5 kph, the rotation has to be broken before the BPL final. This assumption expires on 30 September 2026, or the day the spell sample reaches 40, whichever comes first.

One question stays. When the stadiums fill and the market turns confident, does anyone remember which over the ball landed in the wrong place — and who recorded it?