For years, sports commentators have repeated the same clichés about T20 cricket: "You have to win the powerplay," "You need an anchor to win," and "Team X always chokes."
I wanted to know if any of that was actually true.
The problem is that querying 15+ years of historical ball-by-ball telemetry (over 294,000 deliveries) via commercial sports APIs is painfully slow and brutally rate-limited.
So, I built Midwicket—an open-source SDK that bypasses APIs entirely. It pulls raw open data into a local DuckDB and PyArrow engine, turning your laptop into a sub-millisecond sports data warehouse.
To test the architecture, I wrote a logistic regression Win Probability model (AUC 0.87) trained on 1,239 IPL matches. Because the query engine is entirely local, I could calculate the exact Win Probability Added (WPA) for every single ball ever bowled.
