The Real Price of the Transfer Window: What My 1,200 Hand-Coded Events Say About Who Is Cheap

Seventeen balls, thirty-four runs, three fours and a six. Within a week that innings nearly tripled a finisher's asking price. In the same season another batter faced 230 deliveries between overs seven and fifteen, scored 268, produced not a single viral clip, and his base price in the following player draft did not move by a taka. I coded both of them, ball by ball. One number shook the market; the other the market never saw. The difference is not talent, it is measurement. I did not trust the Bangladesh Premier League's numbers until I had coded them by hand — sitting in Chattogram in 2026, tagging 1,200 events across 24 matches before I first felt that what gets said about this league and what happens in it are two separate things. No API, no shortcut, just ninety minutes of keystrokes and a monk's patience. I watched every match twice. The first time I only watched, taking no numbers down, because a scoreboard in your eye line locks your attention to it. The second time I tagged ball by ball: shot type, bowler type (right-arm pace, left-arm pace, off-spin, leg-spin, left-arm spin), line and length zone, field setting, wickets in hand, batting position, and whether the delivery belonged to a dead rubber or a live match. Without those columns, a shot has no value, because the same shot in over three and over eighteen are two entirely different acts. Strike rate has never satisfied me. The reason is ordinary: in a forty-ball sample two edges flying to the boundary lift strike rate by twenty-five points while the batter's skill does not move a millimetre. The second reason is that strike rate blends phases. In the powerplay, with the field up, six-hitting is easy; in the middle overs, with two spinners and a deep field, it is nearly impossible; in the death overs it is easy again but the risk is far higher. So I do not take the runs on the scorecard. I build the runs that should have come. My model is simple, deliberately simple. For every ball I estimate what it should be worth given phase, bowler type, wickets in hand, and strike-rate pressure. Then I subtract that baseline from what the batter actually did, and the difference is Expected Runs Added, xRA. I judge a batter by xRA per twelve balls faced, because twelve balls is two overs — a small but meaningful window of work. Not two or three innings per match, but how far above baseline they are per twelve balls across a season. That is what repeats. The first counter-intuitive result came from there. In my coded dataset, roughly a third of the batters who score fast in overs sixteen to twenty sat below baseline in overs seven to fifteen. They borrow six to nine runs a match and repay it later. The scoreboard shows the repayment; the loan is never recorded. In a small sample, one innings of 34 off 17 covers the debt entirely. That middle-overs cost is invisible unless boundary percentage and dot-ball percentage are read together. A batter can run at 145, but if 58 percent of his output depends on boundaries, then on the day boundaries do not come he stalls his side. A 128-strike-rate batter with a 28 percent dot-ball rate and 38 percent boundary dependence does more: he keeps the partner alive and keeps wickets for the seventeenth over. The market prices these two a factor of two apart; their real contribution is close to equal. The value of wickets in hand I modelled separately. Sides losing their second wicket before the tenth over clearly scored less per ball across the last six overs, because a new batter has to face the fastest, most hostile phase cold. A franchise that ignores experienced middle-order batters for a queue of power-hitters is not buying the arithmetic of the second half. A batting order is a risk-allocation structure, not a list of names. The same error runs through bowling. Wicket count is the worst single metric this league still uses, because a slogging wicket in over one is not worth a set batter's wicket in over nineteen. In my dataset, death-over economy was the most stable metric season to season; wicket count was the most volatile. Dropping a bowler because he takes fewer wickets means leaving a reliable asset in the market. Matchups gave a strange picture. Left-arm spinners against right-handers hold the best phase-adjusted economy through the middle overs, yet their average market price was about 40 percent below pace bowlers. Visibility politics: 140kph looks good in a transfer window, a flat left-arm orthodox bowler does not. The market pays a premium for what the eye can process quickly. Football has the same disease in goalkeeping — a keeper who can hit a long pass is priced up while the actual job of saving shots is unpriced. Cricket has recast it as delivery speed. With young players the data is more uncomfortable. We hand a teenager promoted from age-group cricket the death overs every week while his frame is still forming. In my coded injury timeline, players under nineteen who carried sustained heavy delivery or match loads in one season missed noticeably more cricket in the next two. Clubs lean hardest on those who look physically mature earliest, and load them fastest. In a transfer window they look ready; the injury record arrives the following year. The biggest blind spot in today's window is injury risk pricing. Data cannot predict injury, but risk can be added to a price. The market tries to buy quick returns cheaply while spending most on the position where one-sided stardom is most fragile. Injury-locked fast bowlers go unsold because the calculation happens in the emotion of the moment, not on the accounts sheet. That is not a verdict; it is an accounting failure. Expectation structures follow from this. Home advantage is not training, it is the crowd. During the 2026-20 Bundesliga restart I compared 83 matches before and after and found home advantage fell from +0.31 to +0.08 xG on average, with home win rate dropping from 43.3 percent to 33.3 percent. Where we assumed travel and adaptation were the causes, the edge dissolved the moment the stands emptied. In cricket the crowd effect cannot be measured like xG, but in my coded dataset boundary reaction times, umpiring decision patterns, and pace bowlers' tendency to alter their angles all differed between full and empty grounds. That is correlation, not causation — the cause sits somewhere else. A franchise that refuses to see this difference builds policy around an invisible club. The larger trap is that the metric which looks modern is usually decoration. In the name of data-rich decisions we often just accumulate more charts when a model should deliver a decision. A model without a decision is a diary, not a weapon. My own test this window is plain: prioritise batters whose xRA per run of transfer fee is highest, and discard extra metrics that cannot be used in a specific phase. Three numbers will tell the story next window — phase-adjusted run addition between overs seven and fifteen, reliable death-over economy, and a fast bowler's recent delivery load. None of them appear in a highlight reel, which is precisely why they remain available at uncontested prices. If the market gets this right, the story will not be yellow. The quietly winning asset will simply be collected in one place.