The WS2812B controller in this knockoff Neopixel failed:
It used to live in the Noval socket:
As with the one atop the big incandescent bulb, it failed by emitting random flashes of primary colors. This time, the Octal and Duodecar sockets were downstream and I got to watch four randomly flashing RGB LEDs, which says the controller failed enough to corrupt the data stream, but not enough to make the downstream controllers regard it as completely invalid.
I replaced it with another one, just like the other ones, and it’s been running happily ever since.
Fairly obviously, cheap knockoff Neopixels aren’t a good deal; the strip and these PCB versions have racked up three or four (I’m losing track) out of less than a dozen deployed. I won’t hold the overtemperature failures against the strip versions, but, still …
The Smell of Molten Projects in the Morning 
Not going to decapsulate the controller and test to see if it’s a chip fail or a wirebonder fail? ;-)
Careful microscopic scrutination suggests a subtle failure that didn’t leave a smoking crater in the chip. I should wire up the entire PCB panel and run it for a few months to see how many more go toes-up under duress. [sigh]
In my work, I support the little rooms at the peripheral of the semiconductor fabs where they have all sorts of magical machines to determine why these subtle failures occur. Opening the packages with acid, plasma, micro milling, or laser, testing bond wire strengths, examining internals with acoustics before opening (SAM), SEM, AFM, microprobing, etc etc. it’s a fascinating, yet incredibly droll part of the process.
Failure analysis can be fun (sometime…). In a previous life, I did F/A on wafer sort failures for analog/interface devices on a 5 micron (min geometry) process. Amazing how much havoc a 1 um defect can cause. At least the defects were big enough to see optically, though we often got creative. We’d use a Nomarski microscope, and/or Sirtl etch for various defects. Things usually got lively whenever we’d update the process, especially wafer size. (Over 20 years, we went from 2″ to 6″ wafers before the powers that be pulled the plug on the department.)