I'll bet it's line frequency derived, after all why wouldn't it be?
If so, imagine this: you need to read the 50 or 60Hz line frequency, so you look at the AC wave coming off the electronics power transformer using say a diode. Since the voltage is a bit high and unregulated, to get that into your micro you go through a resistor divider into the base of a transistor. Say the gain on the transistor is poor and/or the resistor divider isn't quite driving the transistor adequately, so that it gets a half decent signal out the collector but every time there's a bit of noise from the fridge coming on during a high cycle it drops out and gets an extra count. (BTW - last week I was messing around with a line derived clock signal in this fashion, and that's exactly what happened when my biasing was off.)
A nasty spike comes out down the power lines and gives the lower resistor (lower value) on the divider a zap, increasing it's value a bit. Now the transistor is getting a stronger signal and it doesn't count the dropouts anymore.
One can easily see how something like that can slip past QC at the plant. Maybe a combination of marginal components, marginal design and a little extra noise on your personal power lines caused the initial issue. Anyway I've learned not to complain when problems fix themselves.