Correlation is not causation: the uninvited third variable
Two quantities rise together, but one doesn’t drive the other. How to spot a confounder, and why it matters when you want to intervene.
Note 3 · 2 min read
3 notes published
New notes every Tuesday and Friday.
The alarm paradox: why 9 out of 10 doesn’t mean 9 out of 10
A sensor catches 90% of failures, yet only one alarm in six is real. Bayes’ theorem explains why rare events fool us.
Note 2 · 2 min read
A model doesn’t make decisions: it prepares them
An accurate forecast is not enough: what matters is what happens next, what a mistake costs and who decides. Why we always start from the decision.
Note 1 · 2 min read
How to read them
Two levels: basic and advanced
Every note has two levels. “In plain words” needs no formulas; “Going deeper” uses technical language and, where useful, a formula. Figures with synthetic data say so.
Have a real case where these ideas could help?
Notes come out on Tuesdays and Fridays. On Bluesky you’ll find the daily workflow cards.
Let’s talk →