Writing
Analysis I'd want to read myself: the decision the number was for, the method that got there, and the part where it didn't work. Where a piece has runnable code or an interactive report, it's attached.
US install prices rise in the fourth quarter of an election year because presidential campaigns are bidding for the same impressions. It is real, it is specific to one market, and it is the one piece of seasonality you cannot fit.
Spend to CPI to installs to actives to spenders to revenue per user to LTV. Seven links, each with its own estimator, and a chain whose weakest point is not where anyone expects it.
A decade of running experiments on live games, reduced to seventeen rules anyone can apply, a worked case study for each way an experiment lies to you, and the module I built so the pipeline refuses to make the same mistakes twice.
Bias requires truth. A simulator lets you set the truth, hand an estimator only the observable view, and measure how far off it lands — the one measurement real market data can never give you.
I believe in the 80/20 rule — right up until the last 20% is the answer. What Savepoint is, why it exists, and the one idea it is organized around.