IPL Crunch: Toss, Phase & Players Decoded from 278K Deliveries

IPL "common knowledge" says winning the toss is a big edge and death overs decide games. But these claims are repeated without evidence. The data also arrives messy: Cricsheet ships matches as 1,235 separate JSON files, with franchise renames (Delhi Daredevils→Capitals, RCB Bangalore→Bengaluru), tied/no-result matches, super-overs, and run-outs wrongly credited to bowlers — all of which silently skew any naive analysis. Goal: convert the raw JSON to one clean CSV (278,034 deliveries, 2008–2025) and answer three questions with hard numbers instead of folklore. Process 1. Converted all 1,235 match JSONs into a single ball-by-ball CSV with a Python script (one row per delivery). 2. Cleaned the six traps that corrupt results: unified franchise renames, derived a clean season_year from mixed formats ("2009" vs "2007/08"), dropped 25 tie/no-result matches from win-rate maths, removed super-over innings, and counted only bowler-credited dismissals. 3. Answered the toss question against a 50% coin-flip baseline, then split by the captain's decision (bat vs field) to test WHY. 4. For phases, I summed runs per innings per phase and compared winning vs losing teams — the widest gap is the most decisive phase. 5. Validated on BOTH datasets (my self-converted CSV and the ready-made attachment) to prove the findings don't depend on the source. Results Three evidence-backed answers from 278,034 deliveries: • Toss: winners win just 51.7% of decided matches — only +1.7 pts over a coin flip. The toss is NOT a real edge. • Phase: the MIDDLE overs separate winners from losers most (+7.8 runs vs +5.9 powerplay, +4.1 death). • Top 5 (2021–2025): Batters — Gill, Kohli, Rahul, du Plessis, Buttler; Bowlers — HV Patel, Chahal, Arshdeep, Rashid, Avesh. Both datasets agreed: identical top-5 lists, toss within 0.1%. Delivered as a 4-page cricket-themed PDF. Reflection The toss result has a real confound I'd control for next time: chasing teams stop batting once they pass the target, which deflates the winner's death-over runs and understates how much death overs matter. A fairer phase comparison would normalize by balls faced or use only first-innings totals. I'd also add a venue/era breakdown (run rates have climbed sharply since 2008, so an "all-seasons" average hides real change) and a quick significance test on the toss gap to confirm 51.7% is within noise of 50% — which I suspect it is.