Aditya

IPL CRUNCH '26 - Backed by 289,673 Balls of Data

Everyone has IPL opinions — "toss matters", "death overs decide everything" — but nobody backs them with data. With 19 seasons and 289,673 ball-by-ball deliveries available from Cricsheet, I set out to answer 3 questions with proof: Does the toss actually matter? Which phase separates winners from losers? Who are the real top performers across 5 seasons? The gap: most IPL analysis online is surface-level (team totals, averages) and ignores phase-level breakdowns, statistical significance testing, and proper cricket stat rules like excluding Super Overs and using legal balls only. Process I started by downloading ball-by-ball CSV data from Cricsheet (289,673 rows, 1,218 matches). Cleaned it in Python/pandas — fixed season labels (2020/21 → separate seasons using match dates), corrected Cricsheet's zero-indexed overs to real IPL phases (PP: 1-6, Middle: 7-15, Death: 16-20), excluded 16 ties and 9 no-results. For the toss question, I ran a chi-squared statistical test instead of just eyeballing percentages. For phases, I counted every innings once per phase (including 0 runs if a chase ended early) to avoid survivorship bias. For top performers, I excluded Super Overs and used legal-ball stats for SR/economy. Built an interactive Chart.js dashboard and a 168-point verification script that cross-checks every number from raw CSV through JSON, HTML, and the report. First attempt Results Key findings: Toss winners won 51.6% (p=0.284, not significant). Middle overs showed the biggest run gap (+7.5 runs for winners). Death overs had the biggest intensity gap (+1.97 RPO). Top batter: Shubman Gill (2,827 runs), Top bowler: Yuzvendra Chahal (90 wickets). Surprise: 78.7% of captains chose to field first in 2024-2026, but toss winners only won 52.1% — a 26.6pp strategy gap. Verification: 168/168 automated checks passed. Every single number on the dashboard is traceable to the raw CSV. Bonus: venue analysis revealed Chinnaswamy (8.89 RPO) vs Chepauk (8.12 RPO) scoring split. Reflection 1. I'd add player-level phase analysis — who dominates specifically in death overs vs powerplay, not just overall stats. 2. I'd include a "match simulator" where judges can pick two teams and see predicted phase-wise scoring based on historical data. 3. I'd test the dashboard on mobile earlier — some chart labels overlap on smaller screens. 4. I'd add confidence intervals to the phase run gaps, not just averages, so judges can see the spread. 5. I'd use a proper bundler instead of vanilla JS to keep the codebase cleaner as it grew to 900+ lines.