IPL Crunch '26: 10 Cricket Clichés Tested. Most Didn't Survive.

Toss winners win 51.3% of IPL matches, statistically a coin flip. Yet commentary treats it as decisive and captains agonise over it. That gap between confident assertion and actual evidence is everywhere in cricket analysis. I took Wooble's three questions as a starting point and ran every major IPL cliché through 294,757 balls of data. The toss myth was just the beginning. The powerplay, the death overs, the 200-run wall and even CSK's legendary consistency all look different when you stop taking them on faith. Process Spent two days on data engineering before writing a single chart. That decision caught two bugs that would have poisoned everything downstream: an RCB name mismatch splitting their stats in half, and an era-bias where pre-2022 averages were pulling win-probability baselines 4.4 runs lower than reality. Once the spine was clean, ran statistical tests against cricket's most repeated claims. Found a structural break in 2022 scoring confirmed by Chow test. Sixes per match nearly doubled, meaning Kohli's 2016 record and Gill's 2024 season are not the same sport. Built player archetypes via UMAP on 12-feature batter vectors. Finished with a walk-forward prediction model, zero data leakage. Results Over 7 is the cheapest scoring over in T20, cheaper than the opening ball. The real chase wall is 175, not 200. Winners outscore losers by nearly twice as much in death overs as in the powerplay. CSK wins at exactly 60% whether they bat or field, identical to the decimal point, and no other franchise shows this pattern. SRH left roughly 6 wins on the table across their history by consistently choosing to bat when the data said field. The model called RCB to beat GT in Q1. They did. Reflection Build the win-probability model in week one, not week two. Every finding in this project can be expressed as a delta-WP: how much does losing a powerplay wicket move the needle, what is Watson's 80 actually worth in probability terms. Framing everything through one primitive would have made the story tighter. I'd also run a grid search over the 11 signal weights rather than hand-tuning on intuition. The backtest infrastructure was already there, I just didn't use it that way.