It should have been a trivial task: an incandescent bulb in a standard AC-powered lamp at home burned out. I replaced the bulb with one of the same wattage, and thought that would be the end of it.
Imagine my surprise when the new bulb went dark after about an hour. While it is very, very unusual for a new bulb to have such a short lifetime, I figured it could happen, so I replaced the replacement bulb. It, too, lasted about an hour.
Now I was suspicious. I left the second replacement bulb in for a while, and it came back on. My guts and experience told me this was a thermal problem, but I was not sure how it was manifesting itself. But I tackled it with enthusiasm, since a standard AC-lamp should be easy to figure out. No software, no processor, no fancy supply rails, no active components. All you need is a basic meter to measure resistance and voltage and you're all set. And if it is not plugged in, there is no danger to anybody (there's no stored energy) nor danger to the lamp itself (from inadvertent short-circuiting of internal components. In other words), it's a debugger's dream-vacation situation.
Long story short: After many careful measurements, false paths, trying different bulbs, and using my precision calipers, here's what I am pretty sure happened. The center contact “finger” of the socket was slightly depressed and therefore not making solid contact with the bulb's corresponding contact. As the lamp stayed on and heated up, the metal of the finger changed its dimensions and position slightly, and broke the circuit's continuity.
Then things cooled off, the finger went back to its nominal situation, the overall contact was re-established, the flow of current resumed, and the cycle began again. Some bulbs had a more prominent center contact bump and so made better contact, despite the thermally induced change, and so did not have this cycling.
Sure, this was a minor problem in the scope of things, and easily fixed: I pried the socket contact up a little, so it made better contact with the bulb, and the problem has not reappeared. But the entire incident reminded me that even simple devices can have subtle problems, and just because you're an experienced EE who has done lots of complex debugging doesn't mean you won't be (temporarily) puzzled by simple products. It's always good to be a little humble, since products and devices have a mind of their own, don't answer to you, and don't care about your feelings. ♦