Software
The drills
Calibrated exercises that spring real traps. Walk in, misbehave, and learn why.
Reading laws is half of it. The drills are the other half: real repositories with a real task, a deliberately vague ticket, and a trap that the obvious approach walks straight into, because that is exactly how it happens at work. Nothing is broken at rest, every constraint is discoverable, and the green light at the end is allowed to lie. You run them in Claude Code, cold, and then you find out which laws you actually practice.
The drills are calibrated, not decorative: each one is measured against unwarned runs so the trap is known to catch the default behavior most of the time. The humbling part is the point. When eight out of eight runs walk into the same hole, the lesson stops being an opinion about discipline and becomes a fact about defaults, yours included.
How to practice it
Run them cold.
No priming, no rereading the chapter first. The drill measures your defaults, and defaults are what you ship on a tired Tuesday.
Green is not the grade.
The drills deliberately let tests pass on wrong solutions. Ask what the check cannot see before you trust it.
Lose, then reread.
The chapter reads differently after its trap has caught you. That ordering is the pedagogy.
Re-run on new machines.
Calibrations expire as models improve. A drill your setup now passes is a datapoint worth having.