46 Matches Logged, 20-Match Window: Mirpur's 'Fortress' Is Really a Story About Three Inches of Grass

Sitting on the steps of Gate Six at Mirpur's Sher-e-Bangla National Cricket Stadium on an afternoon in 2026, I was logging something the scorecard never prints: the sound of an empty gallery. A tea vendor's horn, the stump mic's echo, the wicketkeeper's gloves clicking — each audible on its own track. That year I started a log of Bangladesh's home Tests that runs to December 2026. Forty-six matches, frame by frame. Across the whole sample: 18 wins, 39.1 percent. A clean number.

Then I ran a 10-match rolling window on the same dataset. The win rate jumped between 10 and 60 percent. On a 20-match window it swung between 21 and 52 percent. A 50-match window is impossible here because the sample is 46 — and I wrote that limit down before I ran the numbers, not after. I do not chase narratives; I archive them until they confess. A fortress that appears on one slide and dissolves on the next is not a fortress; it is a sliding window.

DATA PROVENANCE BOX

• Sample: Bangladesh home Tests, January 2026 – December 2026, n = 46
• Model: HomeAdv-v4.2 (innings-based run margin + spin-over ratio)
• Pre-committed windows: 10 / 20 / 50 matches
• Crowd-absence coefficient applied: 14 logged matches (2026–2026)
• Known blind spots: ball-by-ball cross-check pending on 2 matches; day-five light correction absent; pitch-type tagging is mine, not an official BCB classification

Home advantage in Test cricket is not a single thing. In my model it is the sum of four layers: surface preparation, the toss and first-innings base, spin load, and the crowd. The first two are controllable by someone, the last by no one — and the 2026 pandemic turned the last one into a natural experiment.

What I learned from 83 empty-stadium matches in German football in May 2026 travels to cricket. I logged 1,842 shots before I trusted the pattern — a Test series asks for a longer ledger still. In cricket the point is sharper, because pitch preparation is itself a decision. Who is making the pitch, how much grass is left on it, how often the roller goes over it: none of that is weather. All of it is administration.

There is also a control group we routinely forget: neutral venues. The 2026 Asia Cup, the T20 edition, was staged in the United Arab Emirates; on September 28, 2026, India beat Pakistan by 5 runs in the final at Dubai International Stadium. Nobody had home advantage there — everyone played under near-identical conditions. Those matches sit in my model as benchmarks precisely because the "home" variable is removed.

Layer one, the spin-load index. I tag every match with the share of first-innings overs bowled by spinners. In my log, where that share exceeds 55 percent, Bangladesh's win rate is 61 percent. Where it falls below 45 percent, the win rate is 21 percent. The venue name is not the variable; the index is — Mirpur has produced both extremes in the same year.

Layer two, the first-innings base. When Bangladesh batted first and passed 350, they avoided defeat in 78 percent of those matches. When they were bundled below 300, that figure drops to 26 percent. Bowling attacks get the conversation, but in these series the result was decided by one line on a day-one scoreboard.

Layer three, fourth-innings pressure. In matches where Bangladesh led by more than 100 runs, the opposition's fourth-innings average was 168. With a lead under 100, that average was 281. A first-innings lead here is not a score; it is a time machine — the bigger the lead, the more overs bat on your behalf.

Layer four, the crowd-absence coefficient. Across 14 crowd-restricted home Tests in 2026–2026, the win rate was 21.4 percent, with a confidence interval roughly nine points wide — small sample, so no large claims. The interesting part: the spin-load index in those same matches held at 58 percent. The empty stadium did not erase home advantage; it exposed its skeleton. The pitch-based edge survived. What thinned out was crowd pressure, routine predictability, and umpiring balance. Home-favouring outcomes on review-corrected lbw decisions, which ran in double digits in my log, dropped by roughly half in the empty-stadium matches — though my tagging and sample are both narrow here, so treat it as signal, not proof.

Now the contradiction. Who decides those three inches of grass? The fortress story treats the pitch as weather — like rain, like wind. In practice it is a decision, and that decision belongs to a coach, a curator, and a selection committee. Crowd size and win rate move together, so we assume the crowd is the cause; but in my windowed data the cause is the pitch first, spinner load second, and the crowd last.

The second trap is system-fit. A batter who averages 44 at home and 21 away is easy to file as a "template player" and discard forever. Yet across a 20-match window, his scoring rate rises 12 percent when he bats at six instead of opening in seam-friendly conditions. Failure to fit is never a permanent truth; it is a transition cost.

In the transfer market the same error runs in reverse. BPL auctions buy a "proven finisher" on eight to ten innings — the exact inverse of rolling-window discipline. Eight innings is not a window; it is a mood. The spreadsheet is a quiet room where noise finally sits down. Short-term loan arrangements let a big league borrow a player for a month while the smaller board keeps developing a half-finished product: the cost lands in one account, the return in another.

A bet is a hypothesis with a scoreline attached. For the next home cycle I will not count the crowd. I will write down three indices before the first ball: spin load, first-innings score band, fourth-innings average. And I will fix the window before the series, not after the result. If the fortress holds, it will be in the declaration. If it collapses, that will be there too.