The Pressure Over Index: Bangladesh's Home-Test Collapses Are a Measured System State, Not a Moral Failure

I close the scorecard at the press gallery in Khulna's Sheikh Abu Naser Stadium and open the ball-by-ball log instead. The scorecard says "75 for 4"; the log says something more precise — the wickets fell inside a specific window after the 50th over, off the same ball patterns, against the same field, following the same bowling change. Between 2026 and 2026 I compiled the ball-by-ball traces of 42 Bangladesh home-Test innings into a single file. One cluster kept returning: 61 percent of top-five dismissals landed between overs 50 and 80. One match is a story. Forty-two innings is an index. Commentary language is simple and comfortable: "could not handle the pressure", "weak mentality". Those sentences are not false, but they are not measurable either — and you cannot build a recovery routine out of something you cannot measure. When I joined The Daily Star's sports desk in 2026, I thought cricket analysis meant choosing the right words. After launching Expected Truth from Khulna in 2026, I learned the work is different: fix the index first, write the hypothesis second, audit the process against outcomes third. This piece follows that protocol, so the method comes before the verdict. My Pressure Over Index (POI) is capped at four variables — dot-ball rate, a false-shot proxy, strike-rotation failure, and a ball-quality score. Not five, not six. Extra variables make an index look elegant and its explanations weak. The second measure is Recovery Efficiency (RE): runs scored in the 15 overs after a wicket cluster, divided by the weighted number of wickets lost in that cluster. I locked the baseline before opening the file: a home POI below 55 is normal flow, above 58 is system stress. The revision rule was locked too — add 15 new innings, re-estimate the index, never move the definition mid-series. The first phase of evidence is partly expected and partly uncomfortable. Bangladesh's home POI averages 58.4; the same side away averages 51.2. Opposing batters in the same home venues average 55.9. Of the 6.9-point gap that tempts us to write "Bangladesh's problem" within minutes, 4.7 points belong to conditions — spin-friendly surfaces, slow outfields, humid air, damp seam. The remaining 2.5 points are the real question, and they deserve the real investigation. Clustering narrows the gap further. Sixty-one percent of Bangladesh top-five dismissals fall in overs 50-80; the visiting top five in the same venues falls there 54 percent of the time. Seven points is a tendency, not a catastrophe. This is where I refuse the trap of explaining a system through one innings, which then magically becomes a flaw of personal character. Reading clusters without base rates is blaming a shadow. Recovery Efficiency sharpens the story. In the 15 overs after a cluster, home innings recover at 34.7; away innings at 46.1. The failure threshold is explicit: when RE after a cluster drops below 25, 76 percent of those innings end under 250. That single number produces the crisis map — three checkpoints: lose no extra wicket in the first five overs, keep the scoring rate above 2.4 across the next 40 balls, and build one 25-plus partnership with the lower order. Miss all three and the innings does not return. That is a model prediction, not moral optimism. Innings-level overperformance adds another layer. Against runs expected from my ball-quality model, home first innings in the 2026-2026 window ran 38 runs above expectation; in the 2026-2026 window they sit 11 runs below. The shift is not a collapse of batting talent — bowling workload, calendar density and fourth-innings pitch age all changed far more between those two windows, and that change echoes through innings-level data. The player profiles point the same way. With Shakib Al Hasan in the middle overs, POI drops by roughly four points — not because of scoring, but because of strike rotation. Partnerships with Mushfiqur Rahim lift RE, yet his tempo in the overs after a cluster stays restrained, which is exactly what carries a side through checkpoint one. Litton Das records low rotation failure but a high false-shot proxy: risk and opportunity in one frame. Spell-length data for Mehidy Hasan Miraz and Taijul Islam shows dot-ball rates climbing once the ball passes 60 overs — and that is precisely where clusters are born. Najmul Hossain Shanto and Zakir Hasan tell a different early story: their first-20-ball out-probability is high, their patience deficit is not. Restraint matters here. Clusters and collapses are correlated, not causal. Pitch age, ball condition, light change — none of those three variables existed in my first version. The numbers didn't break the model; they exposed where the model was blind. The opposite trap is just as dangerous: dressing-room information, injuries, travel fatigue — that layer never appears in a ball-by-ball file, and I am not willing to discard it as noise. I don't chase outliers; I follow them until they confess — but one exception never becomes a rule unless the base rate leans the same way. There is a methodological discomfort too. Building an index is easy; keeping it honest is hard. In 2026 I added seven variables to one version — the outputs looked more precise and the explanations grew foggier. Holding 2026-2026 back as a test set showed the four-variable version predicted better. Expected truth is not a verdict; it is a hypothesis with a deadline — and once the deadline passes, the hypothesis belongs to the writer, not the model. For the next home series I am writing three things down in advance: if the top-five dismissal share in the 50-80 over window drops below 55 percent, the cluster thesis weakens; if RE after a cluster reaches 25, the innings crosses 250; if neither holds, I change the model and not the definition. Sheikh Abu Naser Stadium in Khulna has hosted Test cricket since November 2026 (source: ESPNcricinfo venue archive), and franchise data from the Bangladesh Premier League, launched in 2026, was never long-form proof for me. The question is no longer who is to blame — it is which index we use to read the six overs before the cluster arrives.