Why do bugs that won't reproduce cost me days and shake me?

You handle most bugs already. Give you a stack trace, a clear failure, a line number — you find it, you fix it, you move on. That is not the problem. The problem arrives when the bug will not sit still long enough to be seen. This does not mean you have gotten worse. It means you have run out of a tool. Reasoning about code works when the failure repeats on command. Intermittent failures do not obey your schedule. They obey conditions you have not yet named. The days you lose are not wasted on the bug. They are wasted on hunting without a trap. Choose to spend the first hour building the trap — logging, state capture, a way to catch the failure once — instead of the tenth hour guessing at the cause.