Dumb mistake from when I was still working:

My work laptop’s SSD went haywire, and I/O would spike every 10 minutes or so for ~50 ms. The hardware guy said he could replace the SSD right away, or I could endure it for a few weeks and get a new laptop instead. Obviously, I agreed to wait. The stutter noticeably affected screen rendering, but I didn’t notice any other issues. Little did I know that every time it happened, all input was ignored (as in: not queued). Normally it wouldn’t matter, because hitting a random ~50 ms window is hard. How-the-f×ck-ever…

A few days later — without getting into “why” — I was forced to apply a patch in production. So I opened an SSH session to prod in one terminal, spun up a dev environment in another, copied the database schema from prod to dev, and made sure to test everything. No issues, so I jumped to prod, applied the patch, restarted services, jumped back to dev, and cleaned up the now-unnecessary database. Only to discover that my “jumped back to dev” keystroke didn’t register.