I build things with data and I break them on purpose, because a tool only tells you the truth about itself when you push it past what it promised.

My teaching comes out of one habit. I run every example before I show it. Not most of them, every one. It is slower, it is less comfortable, and it has caught me being wrong on camera more than once. One of those moments is still in my pandas course, correction included, because the mistake I made is exactly the mistake the lesson is about.

What I care about in a library is not the feature list, it is the quiet failure. The call that finishes without complaint and hands back a number that is not the number you asked for. Those bugs do not show up in a stack trace, they show up six weeks later in a report nobody can reconcile. That is the material I teach.

I teach dense rather than long. No welcome video, no filler, no watching somebody type. One failure mode per lesson, the code, the real output, the habit that prevents it. If a lesson does not change what you do on Monday, it does not go in.