The Restart Tax: Which Over Bangladesh's T20 Chase Actually Breaks

78 needed off 48. Required rate 9.75. Bangladesh has faced this equation dozens of times. Then the 13th over: four dots, a single, a wicket. Seven balls, three runs, one wicket. The scoreboard still says 'under control'. But the chase has already flipped. I have logged these small clusters separately since January 2026, and a pattern is now clear: Bangladesh's T20 chase failures are not separate match stories; they are the repetition of the same short DNA, returning each time in new clothing. If the structure repeats, match reviews cannot diagnose it. You need a repeatable metric.

Method, sample and context-integrity note\n\nMy log has two tiers. First: ball-by-ball records of 41 Bangladesh international T20 chases from January 2026 to February 2026, with over number, wicket events, bowler type and required rate. Second: 88 domestic T20 chases from the same window, where average fielding and bowling quality sit below the national side. I keep the second tier because if the pattern exists only internationally, the explanation becomes 'mentality' or 'talent'. If the domestic tier points the same way, the explanation is system.\n\nI split venues into three tiers: Tier-A (slow, two-spinner, big boundaries), Tier-B (flat, high-scoring), Tier-C (bounce and seam, true carry). Chases shortened by DLS, chases with extreme dew, and chases where the first innings score was rain-distorted are excluded. The effective sample is therefore 36, not 41. A structural claim is defensible on 36; a fine-grained one is not — I write that limit onto the report.\n\nOn metric importation I keep my own rule. xG in football is a shot's historical conversion probability; a shot is independent of the one before it. Cricket's nearest equivalent is expected runs value per ball. But the analogy breaks precisely there — in cricket a ball rewrites the context of the next one. A wicket redraws the whole run function. Football offers no such refuge. So I do not import xG vocabulary wholesale. I work with ball-level entropy and the 'restart tax'. Italy's PPDA machine once taught me that pressing is not chaos but a ledger — likewise, cricket's middle overs are a dot-ball ledger, not a mood.\n\nWhy the middle overs? Because that is the real battlefield of T20 in Asia. The powerplay is scripted. The death overs are compulsory risk. Overs seven to fifteen are where every ball is a decision — and more decisions mean more room for error.\n\nPowerplay surplus, middle-over deficit\n\nFirst finding, against expectation: Bangladesh's powerplay run rate in chases is 8.1, good by Asian standards. Death-overs run rate is 9.3, competitive. But overs 7-15 drop to 6.4 — the damage happens where no bowling quota is forced and control is most available.\n\nSecond: middle-over dot-ball share is 47% for Bangladesh, 41% for Pakistan, 39% for India, 44% for Sri Lanka. The averages sit close. The real anomaly is not the mean but the distribution. Dots are not independent; one dot raises the probability of the next. My 'dot-chain length' — maximum consecutive dots — averages 3.4 in Bangladesh's failed chases and 1.8 in successful ones. Beyond three dots, the bowler can safely bowl his best ball, the field drops a step deeper, and the batter is forced one step up the risk ladder.\n\nThe restart tax\n\nDefinition: the gap between the run rate over the 12 balls after a wicket falls and the match's own baseline run rate (taken from the first six overs, since that encodes the day's pitch, outfield and best opposition bowling).\n\nOver my 36 chases, Bangladesh's restart tax is 0.71 runs per delivery. India 0.42, Australia 0.38, Sri Lanka 0.66, Pakistan 0.69. 0.71 × 12 = 8.5 runs lost per wicket; at 2.3 middle-over wickets per chase, that is 17-25 runs. My log's average losing margin in failed chases is 19 runs. The mechanism is two-layered. First, the entry tax: Bangladesh's No.4 and No.5 average a career strike rate above 135 but roughly 105 across their first ten balls. That is role design, not talent deficit — they were selected for stability, not entry speed. Second, and more corrosive, the set batter at the other end also slows, by 11-14% over those fifteen balls. One wicket produces a double slowdown.\n\nWhich over the chase actually flips\n\nI flagged the over where win probability dropped below 50% and never recovered. Expectation said 17 or 18. 61% of failed chases flipped between overs 12 and 15. Death-over strike rates therefore record the consequence, not the test — which is why I read the RRR gradient between overs 10 and 16 first.\n\nContrarian: cause or symptom\n\nIf the top order is 60/3 in the tenth over, the middle order is already chasing an impossible rate; the slowdown is rational, and the restart tax is a symptom. I tested this. Of 36 chases, 11 had two-plus powerplay wickets: restart tax 0.94 there, 0.58 in the other 25. So roughly half the effect is conditional — the model survives, but its claim shrinks. Venue is a second confounder: Tier-A pitches raise everyone's restart tax, and it explains around 30% of Bangladesh's number. I also refuse to treat the 2026-21 empty-stadium window as a clean cricket experiment — unlike the Bundesliga's 83 fanless versus 306 attended matches, Asian cricket changed pitch preparation, dew patterns, travel and rotation simultaneously. I deploy that window only for home-advantage questions. The eye test gets a bounded role: hypothesis generator, never verdict. A model is a monastery: you enter with noise and leave with discipline — but denying the noise on exit makes you something other than an analyst.\n\nTakeaway\n\nPre-registered benchmark for the next three bilateral chase series: watch the No.4's first-ten-ball strike rate, and the 12-ball run rate after any wicket in overs 7-15. If the restart tax falls below 0.55 per ball, my model is wrong — and that would be good news. If it stays above 0.65, no amount of death-over power-hitting will shrink the average losing margin, because changing the speedometer does not change the engine.